📎 Update changelog

Merge remote-tracking branch 'origin/staging'
🐛 Fix nil dereference crash during flex layout drag operations (#10845 )
2026-09-12 21:59:43 -04:00 · 2026-08-17 12:08:38 +02:00 · 2026-08-03 09:15:03 +02:00 · 2026-07-31 12:51:35 +02:00 · 2026-07-31 12:06:19 +02:00 · 2026-07-30 14:21:42 +02:00
2172 changed files with 141735 additions and 83607 deletions
@@ -0,0 +1,133 @@
+# `.devenv/` — Per-Workspace AI-Client MCP Configs
+
+This directory carries the pieces needed to point an AI coding agent
+(currently Claude Code, opencode, VS Code Copilot, and the OpenAI Codex CLI)
+at the MCP servers running inside the parallel devenv instance the developer
+is currently working in. Every parallel workspace (`ws0`, `ws1`, …) has its
+own copy because the Penpot MCP and Serena MCP host ports are
+workspace-specific.
+
+## Layout
+
+```
+.devenv/
+  README.md
+  scripts/
+    merge-mcp-config.py    # generator helper invoked by manage.sh
+  shared/                  # committed; workspace-independent entries
+    claude-code.json       # Playwright — same for every workspace
+    opencode.json
+    vscode.json
+    codex.toml
+  templates/               # committed; entries with ${...} port placeholders
+    claude-code.json       # Penpot MCP, Serena MCP — port is the only diff
+    opencode.json
+    vscode.json
+    codex.toml
+  mcp/                     # gitignored; written by manage.sh per workspace
+    claude-code.json       # loaded via Claude Code's --mcp-config flag
+    opencode.json          # loaded via OPENCODE_CONFIG env var
+```
+
+One more file is generated outside `.devenv/`, in the directory VS Code itself
+auto-discovers (gitignored):
+
+```
+.vscode/mcp.json           # auto-loaded by GitHub Copilot in VS Code
+```
+
+Codex is the exception: it has no way to load an MCP config from an arbitrary
+path, and its only project-level config file (`.codex/config.toml`) is one a
+developer may already own. So we do **not** write a file for Codex at all —
+`start-coding-agent codex` injects our servers as `-c` command-line overrides
+built fresh from `shared/codex.toml` + `templates/codex.toml` at launch.
+
+* **`shared/`** holds MCP entries that don't depend on the workspace — the
+  browser-driving Playwright server today, plus any other workspace-independent
+  servers we add later. Same content in every workspace, so it's a static
+  checked-in file.
+* **`templates/`** holds the workspace-specific entries (Penpot MCP, Serena
+  MCP) with `${PENPOT_MCP_PORT}` and `${SERENA_MCP_PORT}` placeholders. The
+  placeholders are resolved per-workspace from the port-base constants in
+  `manage.sh`.
+* **`mcp/`** (Claude Code, opencode) is the result of merging `shared/` with
+  the port-substituted `templates/`. `manage.sh` writes these on every
+  `run-devenv --agentic` pass. Gitignored, dedicated paths with no developer
+  content — never edit by hand, your edits will be overwritten on the next
+  reconcile.
+* **`.vscode/mcp.json`** is the same merge, but written to the path VS Code
+  auto-discovers. Because on `ws0` that path *is* the live repo's own file, the
+  reconcile **deep-merges** into it: any servers you added yourself are kept,
+  and only the entries we manage (`penpot`, `serena-devenv`, `playwright`) are
+  (re)written to the current ports. On `ws1+` the file doesn't exist yet, so it
+  is created from scratch.
+* **`scripts/merge-mcp-config.py`** is the generator. Its `json` mode does the
+  JSON deep-merge (with `--merge-into-existing` for the VS Code path); its
+  `codex-args` mode prints the `-c` assignments for Codex. `manage.sh`'s
+  `_merge-mcp-config-json` helper is a thin shim over the former, and
+  `start-coding-agent` calls the latter directly. Run
+  `python3 .devenv/scripts/merge-mcp-config.py --help` for the CLI.
+
+## Launching a coding agent
+
+The easiest path is the wrapper command, which knows the right flags per
+client, `cd`'s into the target workspace, and refuses to launch unless the
+target instance is running and its MCP config has been generated:
+
+```bash
+# Default target is ws0 (the live repo).
+./manage.sh start-coding-agent claude               [...args to forward]
+./manage.sh start-coding-agent opencode             [...args to forward]
+./manage.sh start-coding-agent vscode               [...args to forward to 'code']
+./manage.sh start-coding-agent codex                [...args to forward]
+
+# Target a parallel workspace with --ws N. N is an integer (non-negative);
+# 'main', 'ws1' and similar spellings are rejected.
+./manage.sh start-coding-agent claude --ws 1
+./manage.sh start-coding-agent opencode --ws 2
+```
+
+Equivalents by hand (run from inside the workspace directory):
+
+```bash
+claude --mcp-config .devenv/mcp/claude-code.json
+OPENCODE_CONFIG=.devenv/mcp/opencode.json opencode
+code "$PWD"                  # VS Code auto-discovers .vscode/mcp.json
+# Codex: pass our servers as -c overrides (no config file is written).
+codex $(python3 .devenv/scripts/merge-mcp-config.py --format codex-args \
+          .devenv/shared/codex.toml .devenv/templates/codex.toml \
+          | sed 's/^/-c /')
+```
+
+`start-coding-agent codex` does the `-c` wiring for you (and resolves the
+workspace's ports first). Because our servers arrive as command-line
+overrides, no "trusted project" prompt is involved for them — that prompt only
+gates Codex's own `.codex/config.toml`, which we never write.
+
+## Overriding our entries
+
+Both the auto-discovered configs and the launcher-loaded configs sit *on top
+of* the developer's global config (with varying precedence rules). All four
+clients offer escape hatches for shadowing entries we ship:
+
+* **Claude Code** — `claude mcp add --scope local …` installs a private entry
+  that overrides the one in `mcp/claude-code.json`. Local scope wins.
+* **opencode** — drop an `opencode.json` at the repo root with the override
+  entries you need. opencode's precedence chain is *global → `OPENCODE_CONFIG`
+  → project*, so the project file always wins. The root `opencode.json` is
+  gitignored on purpose, since these overrides are personal.
+* **VS Code Copilot** — the reconcile deep-merges into `.vscode/mcp.json`, so
+  any servers you add there yourself are preserved (only `penpot`,
+  `serena-devenv` and `playwright` are rewritten). To shadow one of *ours*,
+  put an entry under the same name in your VS Code user-profile MCP config —
+  it is loaded alongside the workspace file and wins.
+* **Codex CLI** — our servers arrive as `-c` overrides, which are Codex's
+  highest-precedence layer, so they win over a same-named `[mcp_servers.<name>]`
+  in your `~/.codex/config.toml` or a project `.codex/config.toml`. To override
+  one of ours, append your own `-c` after the client name — extra args are
+  forwarded after ours and the later `-c` wins, e.g.
+  `./manage.sh start-coding-agent codex -- -c 'mcp_servers.penpot.url="…"'`.
+
+See `docs/technical-guide/developer/agentic-devenv.md` for the broader
+client-configuration story (browser remote debugging, AI-client config
+schemas, manual setup for unsupported clients).
@@ -0,0 +1,204 @@
+#!/usr/bin/env python3
+"""Combine a shared MCP-server config with a port-substituted template for one
+AI coding-agent client.
+
+Invoked per workspace by manage.sh's `write-instance-mcp-configs` (JSON
+clients) and by `start-coding-agent` (Codex). Each supported client ships a
+`.devenv/shared/<tool>.{json,toml}` (workspace-independent entries, e.g.
+Playwright) and a `.devenv/templates/<tool>.{json,toml}` (per-workspace entries
+with `${PENPOT_MCP_PORT}` / `${SERENA_MCP_PORT}` placeholders). This script
+combines the two for the target client.
+
+Two output modes are supported:
+
+  json         Deep-merge two JSON documents under a configurable top-level key
+               (`mcpServers` for Claude Code, `mcp` for opencode, `servers` for
+               VS Code Copilot) and write the result to <out>. Same-name
+               entries in the template override entries in shared. With
+               --merge-into-existing, any pre-existing <out> file is loaded as
+               the lowest-precedence layer first, so entries the developer
+               already had are preserved (ours win on name collision). This is
+               used for VS Code's auto-discovered `.vscode/mcp.json`, which on
+               ws0 IS the live repo's file and may hold the developer's own
+               servers; the Claude/opencode outputs live in a dedicated,
+               gitignored `.devenv/mcp/` path and are written without the flag
+               (a clean overwrite).
+
+  codex-args   Deep-merge the two TOML chunks and print one
+               `dotted.key=<toml-value>` assignment per line to stdout (no
+               <out> file). The caller wraps each line in a `codex -c` flag.
+               Codex has no way to load an MCP config from an arbitrary file
+               path (CODEX_HOME would relocate auth/history too), so rather than
+               writing the auto-discovered `.codex/config.toml` we inject our
+               servers as ephemeral per-invocation overrides. This never
+               touches the developer's project- or user-level Codex config.
+
+In both modes, `${VAR}` placeholders inside *either* chunk are resolved from
+the current environment (only template chunks carry placeholders in practice,
+but the substitution is uniform either way) using Python's
+`os.path.expandvars`. Undefined placeholders are left as `${VAR}` literal text
+-- callers (i.e. manage.sh) are responsible for exporting the variables before
+invoking the script.
+
+Usage:
+  merge-mcp-config.py --format json --key <key> [--merge-into-existing] \
+      <shared> <template> <out>
+  merge-mcp-config.py --format codex-args <shared> <template>
+
+Exit codes:
+  0  success
+  2  argparse error (missing required option, bad value, unreadable input)
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import re
+import sys
+import tomllib
+from pathlib import Path
+
+
+def merge_json(
+    shared_path: Path,
+    tpl_path: Path,
+    out_path: Path,
+    key: str,
+    merge_into_existing: bool,
+) -> None:
+    """Deep-merge JSON documents under a single top-level dict key into out.
+
+    Precedence (lowest to highest): an existing <out> file (only when
+    merge_into_existing is set), then shared, then the template. Entries under
+    `key` are merged by name, so the template wins on a name collision while
+    every other entry the lower layers contributed is kept. Top-level keys
+    other than `key` come from the existing file and shared (shared wins).
+    """
+    shared = json.loads(shared_path.read_text())
+    tpl = json.loads(os.path.expandvars(tpl_path.read_text()))
+
+    base: dict = {}
+    if merge_into_existing and out_path.exists():
+        base = json.loads(out_path.read_text())
+
+    merged: dict = {**base, **shared}
+    merged[key] = {**base.get(key, {}), **shared.get(key, {}), **tpl.get(key, {})}
+
+    out_path.write_text(json.dumps(merged, indent=2) + "\n")
+
+
+def _deep_merge(base: dict, overlay: dict) -> dict:
+    """Recursively merge overlay into base; overlay wins on scalar/list keys."""
+    out = dict(base)
+    for k, v in overlay.items():
+        if isinstance(out.get(k), dict) and isinstance(v, dict):
+            out[k] = _deep_merge(out[k], v)
+        else:
+            out[k] = v
+    return out
+
+
+def _toml_value(value: object) -> str:
+    """Serialize a scalar/list as a TOML literal for a `codex -c` value.
+
+    bool is checked before int because `isinstance(True, int)` is True. Strings
+    are emitted as JSON strings, which are valid TOML basic strings for the
+    ASCII values our configs carry (commands, args, URLs). Tables never reach
+    here -- they are flattened into dotted keys by _flatten.
+    """
+    if isinstance(value, bool):
+        return "true" if value else "false"
+    if isinstance(value, (int, float)):
+        return repr(value)
+    if isinstance(value, str):
+        return json.dumps(value)
+    if isinstance(value, list):
+        return "[" + ", ".join(_toml_value(v) for v in value) + "]"
+    raise TypeError(f"unsupported TOML value type: {type(value).__name__}")
+
+
+_BARE_KEY = re.compile(r"^[A-Za-z0-9_-]+$")
+
+
+def _key_segment(seg: str) -> str:
+    """A dotted-key segment: bare if TOML-safe, else a quoted key."""
+    return seg if _BARE_KEY.match(seg) else json.dumps(seg)
+
+
+def _flatten(obj: dict, prefix: list[str]):
+    """Yield (dotted-path-segments, leaf-value) for every non-table leaf.
+
+    Lists are leaves (TOML arrays), so we do not recurse into them; nested
+    tables (e.g. an `env` table) are flattened into further dotted keys.
+    """
+    for k, v in obj.items():
+        path = prefix + [k]
+        if isinstance(v, dict):
+            yield from _flatten(v, path)
+        else:
+            yield path, v
+
+
+def emit_codex_args(shared_path: Path, tpl_path: Path) -> None:
+    """Print `dotted.key=<toml-value>` lines from the merged TOML chunks."""
+    shared = tomllib.loads(os.path.expandvars(shared_path.read_text()))
+    tpl = tomllib.loads(os.path.expandvars(tpl_path.read_text()))
+    merged = _deep_merge(shared, tpl)
+    for path, value in _flatten(merged, []):
+        dotted = ".".join(_key_segment(s) for s in path)
+        sys.stdout.write(f"{dotted}={_toml_value(value)}\n")
+
+
+def main(argv: list[str]) -> int:
+    parser = argparse.ArgumentParser(
+        description=__doc__.split("\n\n", 1)[0],
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+    )
+    parser.add_argument(
+        "--format",
+        choices=("json", "codex-args"),
+        required=True,
+        help="Output mode: 'json' writes a merged file; 'codex-args' prints -c assignments.",
+    )
+    parser.add_argument(
+        "--key",
+        help="Top-level JSON key under which MCP entries live (required for --format json).",
+    )
+    parser.add_argument(
+        "--merge-into-existing",
+        action="store_true",
+        help="json only: layer the merge on top of an existing <out> file, "
+        "preserving entries already there (ours still win on name collision).",
+    )
+    parser.add_argument("shared", type=Path, help="Path to the shared chunk.")
+    parser.add_argument("template", type=Path, help="Path to the port-placeholder template chunk.")
+    parser.add_argument(
+        "out",
+        type=Path,
+        nargs="?",
+        help="Path the merged result is written to (json only; codex-args writes stdout).",
+    )
+    args = parser.parse_args(argv)
+
+    if args.format == "json":
+        if not args.key:
+            parser.error("--key is required when --format json")
+        if args.out is None:
+            parser.error("out is required when --format json")
+        merge_json(args.shared, args.template, args.out, args.key, args.merge_into_existing)
+    else:  # codex-args
+        if args.key:
+            parser.error("--key is not accepted when --format codex-args")
+        if args.merge_into_existing:
+            parser.error("--merge-into-existing is not accepted when --format codex-args")
+        if args.out is not None:
+            parser.error("out path is not accepted when --format codex-args (result goes to stdout)")
+        emit_codex_args(args.shared, args.template)
+
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv[1:]))
@@ -0,0 +1,8 @@
+{
+  "mcpServers": {
+    "playwright": {
+      "command": "npx",
+      "args": ["@playwright/mcp@latest", "--cdp-endpoint=http://127.0.0.1:9222"]
+    }
+  }
+}
@@ -0,0 +1,8 @@
+# Workspace-independent MCP servers for the OpenAI Codex CLI.
+# This block is concatenated with the port-substituted templates/codex.toml
+# by manage.sh's write-instance-mcp-configs to produce .codex/config.toml at
+# the workspace root.
+
+[mcp_servers.playwright]
+command = "npx"
+args = ["@playwright/mcp@latest", "--cdp-endpoint=http://127.0.0.1:9222"]
@@ -0,0 +1,9 @@
+{
+  "mcp": {
+    "playwright": {
+      "type": "local",
+      "command": ["npx", "@playwright/mcp@latest", "--cdp-endpoint=http://127.0.0.1:9222"],
+      "enabled": true
+    }
+  }
+}
@@ -0,0 +1,9 @@
+{
+  "servers": {
+    "playwright": {
+      "type": "stdio",
+      "command": "npx",
+      "args": ["@playwright/mcp@latest", "--cdp-endpoint=http://127.0.0.1:9222"]
+    }
+  }
+}
@@ -0,0 +1,12 @@
+{
+  "mcpServers": {
+    "penpot": {
+      "command": "npx",
+      "args": ["-y", "mcp-remote", "http://localhost:${PENPOT_MCP_PORT}/mcp", "--allow-http"]
+    },
+    "serena-devenv": {
+      "command": "npx",
+      "args": ["-y", "mcp-remote", "http://localhost:${SERENA_MCP_PORT}/mcp", "--allow-http"]
+    }
+  }
+}
@@ -0,0 +1,10 @@
+# Workspace-specific MCP servers for the OpenAI Codex CLI. The PENPOT_MCP_PORT
+# and SERENA_MCP_PORT placeholders below are filled in per workspace by
+# manage.sh's write-instance-mcp-configs, then the result is concatenated
+# with shared/codex.toml to produce .codex/config.toml.
+
+[mcp_servers.penpot]
+url = "http://localhost:${PENPOT_MCP_PORT}/mcp"
+
+[mcp_servers.serena-devenv]
+url = "http://localhost:${SERENA_MCP_PORT}/mcp"
@@ -0,0 +1,14 @@
+{
+  "mcp": {
+    "penpot": {
+      "type": "remote",
+      "url": "http://localhost:${PENPOT_MCP_PORT}/mcp",
+      "enabled": true
+    },
+    "serena-devenv": {
+      "type": "remote",
+      "url": "http://localhost:${SERENA_MCP_PORT}/mcp",
+      "enabled": true
+    }
+  }
+}
@@ -0,0 +1,12 @@
+{
+  "servers": {
+    "penpot": {
+      "type": "http",
+      "url": "http://localhost:${PENPOT_MCP_PORT}/mcp"
+    },
+    "serena-devenv": {
+      "type": "http",
+      "url": "http://localhost:${SERENA_MCP_PORT}/mcp"
+    }
+  }
+}
@@ -0,0 +1,3 @@
+# Penpot API configuration for error-reports CLI tool
+PENPOT_API_URI=http://localhost:3450
+PENPOT_ACCESS_TOKEN=your-access-token-here
@@ -1,7 +1,8 @@
 description: Create a report to help us improve
-labels: ["bug"]
 name: Bug report
-title: "bug: "
+type: Bug
+labels: ["needs triage"]
+
 body:
  - type: markdown
    attributes:
@@ -1,39 +0,0 @@
---
-name: Bug report
-about: Create a report to help us improve
-title: ''
-labels: ''
-assignees: ''
-type: Bug
-
---
-
-**Describe the bug**
-A clear and concise description of what the bug is.
-
-**To Reproduce**
-Steps to reproduce the behavior:
-1. Go to '...'
-2. Click on '....'
-3. Scroll down to '....'
-4. See error
-
-**Expected behavior**
-A clear and concise description of what you expected to happen.
-
-**Screen recordings and screenshots**
-If possible, add screen recordings or screenshots to help explain your problem.
-
-**Desktop (please complete the following information):**
- - OS: [e.g. iOS]
- - Browser [e.g. chrome, safari]
- - Version [e.g. 22]
-
-**Smartphone (please complete the following information):**
- - Device: [e.g. iPhone6]
- - OS: [e.g. iOS8.1]
- - Browser [e.g. stock browser, safari]
- - Version [e.g. 22]
-
-**Additional context**
-Add any other context about the problem here.
@@ -0,0 +1 @@
+blank_issues_enabled: true
@@ -1,7 +1,8 @@
 description: Suggest an idea for this project.
-labels: ["needs triage", "enhancement"]
+labels: ["needs triage"]
 name: "Feature request"
-title: "feature: "
+type: Enhancement
+
 body:
  - type: markdown
    attributes:
@@ -1,21 +0,0 @@
---
-name: Feature request
-about: Suggest an idea for this project
-title: ''
-labels: ''
-assignees: ''
-type: Feature
-
---
-
-**Is your feature request related to a problem? Please describe.**
-A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]
-
-**Describe the solution you'd like**
-A clear and concise description of what you want to happen.
-
-**Describe alternatives you've considered**
-A clear and concise description of any alternative solutions or features you've considered.
-
-**Additional context**
-Add any other context or screenshots about the feature request here.
@@ -15,6 +15,5 @@
 - [ ] Add or modify existing integration tests in case of bugs or new features, if applicable.
 - [ ] Refactor any modified SCSS files following the refactor guide.
 - [ ] Check CI passes successfully.
- [ ] Update the `CHANGES.md` file, referencing the related GitHub issue, if applicable.

 <!-- For more details, check the contribution guidelines: https://github.com/penpot/penpot/blob/develop/CONTRIBUTING.md -->
@@ -0,0 +1,76 @@
+name: Auto Label and Add to Project
+
+on:
+  issues:
+    types: [opened]
+  pull_request:
+    types: [opened]
+
+jobs:
+  triage:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Generate GitHub App token
+        id: triage-app-token
+        uses: actions/create-github-app-token@v1
+        with:
+          app-id: ${{ secrets.TRIAGE_APP_ID }}
+          private-key: ${{ secrets.TRIAGE_APP_PRIVATE_KEY }}
+          owner: penpot
+
+      - name: Process Issue or PR
+        uses: actions/github-script@v7
+        with:
+          github-token: ${{ steps.triage-app-token.outputs.token }}
+          script: |
+            // === 1. CONFIGURATION ===
+            const PROJECT_NUMBER = 8; // <--- Replace with your project board number
+            const IS_ORG = true;      // <--- Set to false if this is a personal project, true if an organization
+            const OWNER = context.repo.owner;
+            const REPO = context.repo.repo;
+
+            const issueNumber = context.issue.number;
+            const isPR = !!context.payload.pull_request;
+            const contentId = isPR ? context.payload.pull_request.node_id : context.payload.issue.node_id;
+
+            // Define your labels here
+            const labelToApply = 'needs triage';
+
+            // === 2. APPLY THE LABEL ===
+            console.log(`Applying label "${labelToApply}" to ${isPR ? 'PR' : 'Issue'} #${issueNumber}...`);
+            await github.rest.issues.addLabels({
+              issue_number: issueNumber,
+              owner: OWNER,
+              repo: REPO,
+              labels: [labelToApply]
+            });
+
+            // === 3. ADD TO PROJECT BOARD ===
+            console.log(`Fetching Project #${PROJECT_NUMBER} ID...`);
+            const projectQuery = `
+              query($owner: String!, $number: Int!) {
+                ${IS_ORG ? 'organization' : 'user'}(login: $owner) {
+                  projectV2(number: $number) {
+                    id
+                  }
+                }
+              }
+            `;
+
+            const projectRes = await github.graphql(projectQuery, { owner: OWNER, number: PROJECT_NUMBER });
+            const projectId = IS_ORG ? projectRes.organization.projectV2.id : projectRes.user.projectV2.id;
+
+            console.log(`Adding item to project board...`);
+            const addToProjectMutation = `
+              mutation($projectId: ID!, $contentId: ID!) {
+                addProjectV2ItemById(input: {projectId: $projectId, contentId: $contentId}) {
+                  item {
+                    id
+                  }
+                }
+              }
+            `;
+
+            await github.graphql(addToProjectMutation, { projectId, contentId });
+            console.log("Automation successfully completed!");
+
@@ -37,6 +37,10 @@ on:
        required: false
        default: 'yes'

+concurrency:
+  group: ${{ github.workflow }}-${{ inputs.gh_ref }}
+  cancel-in-progress: true
+
 jobs:
  build-bundle:
    name: Build and Upload Penpot Bundle
@@ -6,7 +6,6 @@ on:
 jobs:
  build-and-push:
    name: Build and push DevEnv Docker image
-    environment: release-admins
    runs-on: penpot-runner-02

    steps:
@@ -16,6 +16,10 @@ on:
        required: true
        default: 'develop'

+concurrency:
+  group: ${{ github.workflow }}-${{ inputs.gh_ref }}
+  cancel-in-progress: true
+
 jobs:
  build-and-push:
    name: Build and Push Penpot Docker Images
@@ -1,22 +0,0 @@
-name: _MAIN-STAGING
-
-on:
-  workflow_dispatch:
-  schedule:
-    - cron: '26 5-20 * * 1-5'
-
-jobs:
-  build-bundle:
-    uses: ./.github/workflows/build-bundle.yml
-    secrets: inherit
-    with:
-      gh_ref: "main-staging"
-      build_wasm: "yes"
-      build_storybook: "yes"
-
-  build-docker:
-    needs: build-bundle
-    uses: ./.github/workflows/build-docker.yml
-    secrets: inherit
-    with:
-      gh_ref: "main-staging"
@@ -19,7 +19,6 @@ permissions:

 jobs:
  release:
-    environment: release-admins
    runs-on: ubuntu-24.04
    outputs:
      version: ${{ steps.vars.outputs.gh_ref }}
@@ -0,0 +1,85 @@
+name: "CI: Backend"
+
+defaults:
+  run:
+    shell: bash
+
+on:
+  pull_request:
+    paths:
+      - 'backend/**'
+      - 'common/**'
+
+    types:
+      - opened
+      - synchronize
+      - ready_for_review
+
+  push:
+    branches:
+      - develop
+      - staging
+
+    paths:
+      - 'backend/**'
+      - 'common/**'
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  test-backend:
+    if: ${{ !github.event.pull_request.draft }}
+    name: "Backend Tests"
+    runs-on: penpot-runner-02
+    container:
+      image: penpotapp/devenv:latest
+      volumes:
+        - /var/cache/github-runner/m2:/root/.m2
+        - /var/cache/github-runner/gitlib:/root/.gitlibs
+
+    services:
+      postgres:
+        image: postgres:17
+        # Provide the password for postgres
+        env:
+          POSTGRES_USER: penpot_test
+          POSTGRES_PASSWORD: penpot_test
+          POSTGRES_DB: penpot_test
+
+        # Set health checks to wait until postgres has started
+        options: >-
+          --health-cmd pg_isready
+          --health-interval 10s
+          --health-timeout 5s
+          --health-retries 5
+
+      redis:
+        image: valkey/valkey:9
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Fmt
+        working-directory: ./backend
+        run: |
+          cljfmt check --parallel=true src/ test/
+
+      - name: Lint
+        working-directory: ./backend
+        run: |
+          clj-kondo --parallel --lint ../common/src/ src/
+
+      - name: Tests
+        working-directory: ./backend
+        env:
+          PENPOT_TEST_DATABASE_URI: "postgresql://postgres/penpot_test"
+          PENPOT_TEST_DATABASE_USERNAME: penpot_test
+          PENPOT_TEST_DATABASE_PASSWORD: penpot_test
+          PENPOT_TEST_REDIS_URI: "redis://redis/1"
+
+        run: |
+          mkdir -p /tmp/penpot;
+          clojure -M:dev:test
@@ -0,0 +1,57 @@
+name: "CI: Common"
+
+defaults:
+  run:
+    shell: bash
+
+on:
+  pull_request:
+    paths:
+      - 'common/**'
+
+    types:
+      - opened
+      - synchronize
+      - ready_for_review
+
+  push:
+    branches:
+      - develop
+      - staging
+
+    paths:
+      - 'common/**'
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  test-common:
+    if: ${{ !github.event.pull_request.draft }}
+    name: "Common Tests"
+    runs-on: penpot-runner-02
+    container:
+      image: penpotapp/devenv:latest
+      volumes:
+        - /var/cache/github-runner/m2:/root/.m2
+        - /var/cache/github-runner/gitlib:/root/.gitlibs
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Lint
+        working-directory: ./common
+        run: |
+          corepack enable;
+          corepack install;
+          pnpm install;
+          pnpm run check-fmt:clj
+          pnpm run check-fmt:js
+          pnpm run lint:clj
+
+      - name: Tests
+        working-directory: ./common
+        run: |
+          ./scripts/test
@@ -0,0 +1,71 @@
+name: "CI: Frontend"
+
+defaults:
+  run:
+    shell: bash
+
+on:
+  pull_request:
+    paths:
+      - 'frontend/**'
+      - 'common/**'
+      - 'render-wasm/**'
+
+    types:
+      - opened
+      - synchronize
+      - ready_for_review
+
+  push:
+    branches:
+      - develop
+      - staging
+
+    paths:
+      - 'frontend/**'
+      - 'common/**'
+      - 'render-wasm/**'
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  test-frontend:
+    if: ${{ !github.event.pull_request.draft }}
+    name: "Frontend Tests"
+    runs-on: penpot-runner-02
+    container:
+      image: penpotapp/devenv:latest
+      volumes:
+        - /var/cache/github-runner/m2:/root/.m2
+        - /var/cache/github-runner/gitlib:/root/.gitlibs
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Lint
+        working-directory: ./frontend
+        run: |
+          corepack enable;
+          corepack install;
+          pnpm install;
+          pnpm run check-fmt:js
+          pnpm run check-fmt:clj
+          pnpm run check-fmt:scss
+          pnpm run lint:clj
+          pnpm run lint:js
+          pnpm run lint:scss
+
+      - name: Unit Tests
+        working-directory: ./frontend
+        run: |
+          ./scripts/test
+
+      - name: Component Tests
+        working-directory: ./frontend
+        env:
+          VITEST_BROWSER_TIMEOUT: 120000
+        run: |
+          ./scripts/test-components
@@ -0,0 +1,93 @@
+name: "CI: Integration"
+
+defaults:
+  run:
+    shell: bash
+
+on:
+  pull_request:
+    paths:
+      - 'frontend/**'
+      - 'common/**'
+      - 'render-wasm/**'
+
+    types:
+      - opened
+      - synchronize
+      - ready_for_review
+
+  push:
+    branches:
+      - develop
+      - staging
+
+    paths:
+      - 'frontend/**'
+      - 'common/**'
+      - 'render-wasm/**'
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  build-integration:
+    if: ${{ !github.event.pull_request.draft }}
+    name: "Build Integration Bundle"
+    runs-on: penpot-runner-02
+    container:
+      image: penpotapp/devenv:latest
+      volumes:
+        - /var/cache/github-runner/m2:/root/.m2
+        - /var/cache/github-runner/gitlib:/root/.gitlibs
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Build Bundle
+        working-directory: ./frontend
+        run: |
+          ./scripts/build
+
+      - name: Store Bundle Cache
+        uses: actions/cache@v5
+        with:
+          key: "integration-bundle-${{ github.sha }}"
+          path: frontend/resources/public
+
+  test-integration:
+    if: ${{ !github.event.pull_request.draft }}
+    name: "Integration Tests"
+    runs-on: penpot-runner-02
+    container:
+      image: penpotapp/devenv:latest
+      volumes:
+        - /var/cache/github-runner/m2:/root/.m2
+        - /var/cache/github-runner/gitlib:/root/.gitlibs
+
+    needs: build-integration
+
+    steps:
+      - name: Checkout Repository
+        uses: actions/checkout@v6
+
+      - name: Restore Cache
+        uses: actions/cache/restore@v5
+        with:
+          key: "integration-bundle-${{ github.sha }}"
+          path: frontend/resources/public
+
+      - name: Run Tests
+        working-directory: ./frontend
+        run: |
+          ./scripts/test-e2e
+
+      - name: Upload test result
+        uses: actions/upload-artifact@v7
+        if: always()
+        with:
+          name: integration-tests-result
+          path: frontend/test-results/
+          overwrite: true
+          retention-days: 3
@@ -0,0 +1,58 @@
+name: "CI: Library"
+
+defaults:
+  run:
+    shell: bash
+
+on:
+  pull_request:
+    paths:
+      - 'common/**'
+      - 'library/**'
+
+    types:
+      - opened
+      - synchronize
+      - ready_for_review
+
+  push:
+    branches:
+      - develop
+      - staging
+
+    paths:
+      - 'common/**'
+      - 'library/**'
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  test-library:
+    if: ${{ !github.event.pull_request.draft }}
+    name: "Library Tests"
+    runs-on: penpot-runner-02
+    container:
+      image: penpotapp/devenv:latest
+      volumes:
+        - /var/cache/github-runner/m2:/root/.m2
+        - /var/cache/github-runner/gitlib:/root/.gitlibs
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Lint
+        working-directory: ./library
+        run: |
+          corepack enable;
+          corepack install;
+          pnpm install;
+          pnpm run check-fmt
+          pnpm run lint
+
+      - name: Tests
+        working-directory: ./library
+        run: |
+          ./scripts/test
@@ -45,3 +45,8 @@ jobs:
          pnpm run fmt:check;
          pnpm -r run build;
          pnpm -r run types:check;
+
+      - name: Tests
+        working-directory: ./mcp
+        run: |
+          pnpm run test;
@@ -0,0 +1,133 @@
+name: "CI: Plugin API Test Suite"
+
+# Runs the Plugin API Test Suite (it exercises the real Penpot Plugin API, so it
+# needs a running frontend + the plugin runtime). Two jobs:
+#
+# - api-test-suite-mocked (pull_request / push): the per-PR gate. Serves the
+#   prebuilt frontend bundle and intercepts every backend RPC with Playwright
+#   (MOCK_BACKEND=1). No backend / no login. Validates the frontend Plugin API
+#   binding + in-memory store; backend-result-dependent tests are skipped via the
+#   `skipIfMocked` tag. See plugins/apps/plugin-api-test-suite/README.md.
+#
+# - api-test-suite-live (workflow_dispatch): true end-to-end against a LIVE
+#   instance. Point PENPOT_BASE_URL at a reachable instance and provide login
+#   credentials via repo secrets. Manual because the CI runner has no Docker to
+#   stand up a full stack.
+
+defaults:
+  run:
+    shell: bash
+
+on:
+  workflow_dispatch:
+    inputs:
+      base_url:
+        description: "Penpot base URL (e.g. https://localhost:3449)"
+        required: false
+        default: "https://localhost:3449"
+
+  pull_request:
+    paths:
+      - 'plugins/**'
+      - 'frontend/**'
+      - 'common/**'
+    types:
+      - opened
+      - synchronize
+      - ready_for_review
+
+  push:
+    branches:
+      - develop
+      - staging
+    paths:
+      - 'plugins/**'
+      - 'frontend/src/app/plugins/**'
+      - 'common/**'
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  api-test-suite-mocked:
+    if: ${{ github.event_name != 'workflow_dispatch' && !github.event.pull_request.draft }}
+    name: "Run Plugin API Test Suite (mocked)"
+    runs-on: penpot-runner-02
+    container:
+      image: penpotapp/devenv:latest
+      volumes:
+        - /var/cache/github-runner/m2:/root/.m2
+        - /var/cache/github-runner/gitlib:/root/.gitlibs
+
+    steps:
+      - uses: actions/checkout@v6
+
+      # Mocked mode serves the prebuilt bundle from frontend/resources/public.
+      - name: Build frontend bundle
+        working-directory: ./frontend
+        run: ./scripts/build
+
+      - name: Install deps
+        working-directory: ./plugins
+        run: |
+          corepack enable;
+          corepack install;
+          pnpm install;
+
+      - name: Install Playwright Chromium
+        working-directory: ./plugins
+        run: pnpm --filter plugin-api-test-suite exec playwright install --with-deps chromium
+
+      - name: Generate API surface
+        working-directory: ./plugins
+        run: pnpm --filter plugin-api-test-suite run gen:api
+
+      - name: Run API test suite (mocked)
+        working-directory: ./plugins
+        env:
+          MOCK_BACKEND: "1"
+        run: pnpm --filter plugin-api-test-suite run test:ci
+
+  ## The following job will launch the whole suite of tests but we need
+  ## to have a full environment in the CI for this to work.
+
+  # api-test-suite-live:
+  #   if: ${{ github.event_name == 'workflow_dispatch' }}
+  #   name: Run Plugin API Test Suite (live)
+  #   runs-on: penpot-runner-02
+  #   container:
+  #     image: penpotapp/devenv:latest
+  # 
+  #   env:
+  #     PENPOT_BASE_URL: ${{ github.event.inputs.base_url }}
+  #     E2E_LOGIN_EMAIL: ${{ secrets.E2E_LOGIN_EMAIL }}
+  #     E2E_LOGIN_PASSWORD: ${{ secrets.E2E_LOGIN_PASSWORD }}
+  # 
+  #   steps:
+  #     - uses: actions/checkout@v6
+  # 
+  #     - name: Setup Node
+  #       uses: actions/setup-node@v6
+  #       with:
+  #         node-version-file: .nvmrc
+  # 
+  #     - name: Install deps
+  #       working-directory: ./plugins
+  #       run: |
+  #         corepack enable;
+  #         corepack install;
+  #         pnpm install;
+  # 
+  #     - name: Install Playwright Chromium
+  #       working-directory: ./plugins
+  #       run: pnpm --filter plugin-api-test-suite exec playwright install --with-deps chromium
+  # 
+  #     - name: Generate API surface
+  #       working-directory: ./plugins
+  #       run: pnpm --filter plugin-api-test-suite run gen:api
+  # 
+  #     # Note: requires a running Penpot instance reachable at PENPOT_BASE_URL.
+  #     - name: Run API test suite
+  #       working-directory: ./plugins
+  #       run: pnpm --filter plugin-api-test-suite run test:ci
@@ -0,0 +1,77 @@
+name: "CI: Plugins"
+
+defaults:
+  run:
+    shell: bash
+
+on:
+  pull_request:
+    paths:
+      - 'plugins/**'
+
+    types:
+      - opened
+      - synchronize
+      - ready_for_review
+
+  push:
+    branches:
+      - develop
+      - staging
+
+    paths:
+      - 'plugins/**'
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  test-plugins:
+    if: ${{ !github.event.pull_request.draft }}
+    name: Plugins Runtime Linter & Tests
+    runs-on: penpot-runner-02
+    container:
+      image: penpotapp/devenv:latest
+      volumes:
+        - /var/cache/github-runner/m2:/root/.m2
+        - /var/cache/github-runner/gitlib:/root/.gitlibs
+
+    steps:
+      - uses: actions/checkout@v6
+
+      - name: Install deps
+        working-directory: ./plugins
+        shell: bash
+        run: |
+          corepack enable;
+          corepack install;
+          pnpm install -r;
+
+      - name: Run Lint
+        working-directory: ./plugins
+        run: pnpm run lint
+
+      - name: Run Format Check
+        working-directory: ./plugins
+        run: pnpm run format:check
+
+      - name: Run Test
+        working-directory: ./plugins
+        run: pnpm run test
+
+      - name: Build runtime
+        working-directory: ./plugins
+        run: pnpm run build:runtime
+
+      - name: Build doc
+        working-directory: ./plugins
+        run: pnpm run build:doc
+
+      - name: Build plugins
+        working-directory: ./plugins
+        run: pnpm run build:plugins
+
+      - name: Build styles
+        working-directory: ./plugins
+        run: pnpm run build:styles-example
@@ -0,0 +1,57 @@
+name: "CI: WASM"
+
+defaults:
+  run:
+    shell: bash
+
+on:
+  pull_request:
+    paths:
+      - 'render-wasm/**'
+
+    types:
+      - opened
+      - synchronize
+      - ready_for_review
+
+  push:
+    branches:
+      - develop
+      - staging
+
+    paths:
+      - 'render-wasm/**'
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  test-render-wasm:
+    if: ${{ !github.event.pull_request.draft }}
+    name: "Render WASM Tests"
+    runs-on: penpot-runner-02
+    container:
+      image: penpotapp/devenv:latest
+      volumes:
+        - /var/cache/github-runner/m2:/root/.m2
+        - /var/cache/github-runner/gitlib:/root/.gitlibs
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Format
+        working-directory: ./render-wasm
+        run: |
+          cargo fmt --check
+
+      - name: Lint
+        working-directory: ./render-wasm
+        run: |
+          ./lint
+
+      - name: Test
+        working-directory: ./render-wasm
+        run: |
+          ./test
@@ -1,410 +0,0 @@
-name: "CI"
-
-defaults:
-  run:
-    shell: bash
-
-on:
-  pull_request:
-    types:
-      - opened
-      - synchronize
-      - ready_for_review
-  push:
-    branches:
-      - develop
-      - staging
-
-concurrency:
-  group: ${{ github.event.pull_request.number || github.ref }}
-  cancel-in-progress: true
-
-jobs:
-  lint:
-    if: ${{ !github.event.pull_request.draft }}
-    name: "Linter"
-    runs-on: penpot-runner-02
-    container:
-      image: penpotapp/devenv:latest
-      volumes:
-        - /tmp/.m2:/root/.m2
-        - /tmp/.gitlibs:/root/.gitlibs
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v6
-
-      - name: Lint Common
-        working-directory: ./common
-        run: |
-          corepack enable;
-          corepack install;
-          pnpm install;
-          pnpm run check-fmt:clj
-          pnpm run check-fmt:js
-          pnpm run lint:clj
-
-      - name: Lint Frontend
-        working-directory: ./frontend
-        run: |
-          corepack enable;
-          corepack install;
-          pnpm install;
-          pnpm run check-fmt:js
-          pnpm run check-fmt:clj
-          pnpm run check-fmt:scss
-          pnpm run lint:clj
-          pnpm run lint:js
-          pnpm run lint:scss
-
-      - name: Lint Backend
-        working-directory: ./backend
-        run: |
-          corepack enable;
-          corepack install;
-          pnpm install;
-          pnpm run check-fmt
-          pnpm run lint
-
-      - name: Lint Exporter
-        working-directory: ./exporter
-        run: |
-          corepack enable;
-          corepack install;
-          pnpm install;
-          pnpm run check-fmt
-          pnpm run lint
-
-      - name: Lint Library
-        working-directory: ./library
-        run: |
-          corepack enable;
-          corepack install;
-          pnpm install;
-          pnpm run check-fmt
-          pnpm run lint
-
-  test-common:
-    if: ${{ !github.event.pull_request.draft }}
-    name: "Common Tests"
-    runs-on: penpot-runner-02
-    container:
-      image: penpotapp/devenv:latest
-      volumes:
-        - /tmp/.m2:/root/.m2
-        - /tmp/.gitlibs:/root/.gitlibs
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v6
-
-      - name: Run tests
-        working-directory: ./common
-        run: |
-          ./scripts/test
-
-  test-plugins:
-    if: ${{ !github.event.pull_request.draft }}
-    name: Plugins Runtime Linter & Tests
-    runs-on: penpot-runner-02
-    container:
-      image: penpotapp/devenv:latest
-      volumes:
-        - /tmp/.m2:/root/.m2
-        - /tmp/.gitlibs:/root/.gitlibs
-
-    steps:
-      - uses: actions/checkout@v6
-
-      - name: Setup Node
-        id: setup-node
-        uses: actions/setup-node@v6
-        with:
-          node-version-file: .nvmrc
-
-      - name: Install deps
-        working-directory: ./plugins
-        shell: bash
-        run: |
-          corepack enable;
-          corepack install;
-          pnpm install;
-
-      - name: Run Lint
-        working-directory: ./plugins
-        run: pnpm run lint
-
-      - name: Run Format Check
-        working-directory: ./plugins
-        run: pnpm run format:check
-
-      - name: Run Test
-        working-directory: ./plugins
-        run: pnpm run test
-
-      - name: Build runtime
-        working-directory: ./plugins
-        run: pnpm run build:runtime
-
-      - name: Build doc
-        working-directory: ./plugins
-        run: pnpm run build:doc
-
-      - name: Build plugins
-        working-directory: ./plugins
-        run: pnpm run build:plugins
-
-      - name: Build styles
-        working-directory: ./plugins
-        run: pnpm run build:styles-example
-
-  test-frontend:
-    if: ${{ !github.event.pull_request.draft }}
-    name: "Frontend Tests"
-    runs-on: penpot-runner-02
-    container:
-      image: penpotapp/devenv:latest
-      volumes:
-        - /tmp/.m2:/root/.m2
-        - /tmp/.gitlibs:/root/.gitlibs
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v6
-
-      - name: Unit Tests
-        working-directory: ./frontend
-        run: |
-          ./scripts/test
-
-      - name: Component Tests
-        working-directory: ./frontend
-        env:
-          VITEST_BROWSER_TIMEOUT: 120000
-        run: |
-          ./scripts/test-components
-
-  test-render-wasm:
-    if: ${{ !github.event.pull_request.draft }}
-    name: "Render WASM Tests"
-    runs-on: penpot-runner-02
-    container:
-      image: penpotapp/devenv:latest
-      volumes:
-        - /tmp/.m2:/root/.m2
-        - /tmp/.gitlibs:/root/.gitlibs
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v6
-
-      - name: Format
-        working-directory: ./render-wasm
-        run: |
-          cargo fmt --check
-
-      - name: Lint
-        working-directory: ./render-wasm
-        run: |
-          ./lint
-
-      - name: Test
-        working-directory: ./render-wasm
-        run: |
-          ./test
-
-  test-backend:
-    if: ${{ !github.event.pull_request.draft }}
-    name: "Backend Tests"
-    runs-on: penpot-runner-02
-    container:
-      image: penpotapp/devenv:latest
-      volumes:
-        - /tmp/.m2:/root/.m2
-        - /tmp/.gitlibs:/root/.gitlibs
-
-    services:
-      postgres:
-        image: postgres:17
-        # Provide the password for postgres
-        env:
-          POSTGRES_USER: penpot_test
-          POSTGRES_PASSWORD: penpot_test
-          POSTGRES_DB: penpot_test
-
-        # Set health checks to wait until postgres has started
-        options: >-
-          --health-cmd pg_isready
-          --health-interval 10s
-          --health-timeout 5s
-          --health-retries 5
-
-      redis:
-        image: valkey/valkey:9
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v6
-
-      - name: Run tests
-        working-directory: ./backend
-        env:
-          PENPOT_TEST_DATABASE_URI: "postgresql://postgres/penpot_test"
-          PENPOT_TEST_DATABASE_USERNAME: penpot_test
-          PENPOT_TEST_DATABASE_PASSWORD: penpot_test
-          PENPOT_TEST_REDIS_URI: "redis://redis/1"
-
-        run: |
-          clojure -M:dev:test --reporter kaocha.report/documentation
-
-  test-library:
-    if: ${{ !github.event.pull_request.draft }}
-    name: "Library Tests"
-    runs-on: penpot-runner-02
-    container:
-      image: penpotapp/devenv:latest
-      volumes:
-        - /tmp/.m2:/root/.m2
-        - /tmp/.gitlibs:/root/.gitlibs
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v6
-
-      - name: Run tests
-        working-directory: ./library
-        run: |
-          ./scripts/test
-
-  build-integration:
-    if: ${{ !github.event.pull_request.draft }}
-    name: "Build Integration Bundle"
-    runs-on: penpot-runner-02
-    container:
-      image: penpotapp/devenv:latest
-      volumes:
-        - /tmp/.m2:/root/.m2
-        - /tmp/.gitlibs:/root/.gitlibs
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v6
-
-      - name: Build Bundle
-        working-directory: ./frontend
-        run: |
-          ./scripts/build
-
-      - name: Store Bundle Cache
-        uses: actions/cache@v5
-        with:
-          key: "integration-bundle-${{ github.sha }}"
-          path: frontend/resources/public
-
-  test-integration-1:
-    if: ${{ !github.event.pull_request.draft }}
-    name: "Integration Tests 1/3"
-    runs-on: penpot-runner-02
-    container:
-      image: penpotapp/devenv:latest
-      volumes:
-        - /tmp/.m2:/root/.m2
-        - /tmp/.gitlibs:/root/.gitlibs
-
-    needs: build-integration
-
-    steps:
-      - name: Checkout Repository
-        uses: actions/checkout@v6
-
-      - name: Restore Cache
-        uses: actions/cache/restore@v5
-        with:
-          key: "integration-bundle-${{ github.sha }}"
-          path: frontend/resources/public
-
-      - name: Run Tests
-        working-directory: ./frontend
-        run: |
-          ./scripts/test-e2e --shard="1/3";
-
-      - name: Upload test result
-        uses: actions/upload-artifact@v7
-        if: always()
-        with:
-          name: integration-tests-result-1
-          path: frontend/test-results/
-          overwrite: true
-          retention-days: 3
-
-  test-integration-2:
-    if: ${{ !github.event.pull_request.draft }}
-    name: "Integration Tests 2/3"
-    runs-on: penpot-runner-02
-    container:
-      image: penpotapp/devenv:latest
-      volumes:
-        - /tmp/.m2:/root/.m2
-        - /tmp/.gitlibs:/root/.gitlibs
-
-    needs: build-integration
-
-    steps:
-      - name: Checkout Repository
-        uses: actions/checkout@v6
-
-      - name: Restore Cache
-        uses: actions/cache/restore@v5
-        with:
-          key: "integration-bundle-${{ github.sha }}"
-          path: frontend/resources/public
-
-      - name: Run Tests
-        working-directory: ./frontend
-        run: |
-          ./scripts/test-e2e --shard="2/3";
-
-      - name: Upload test result
-        uses: actions/upload-artifact@v7
-        if: always()
-        with:
-          name: integration-tests-result-2
-          path: frontend/test-results/
-          overwrite: true
-          retention-days: 3
-
-  test-integration-3:
-    if: ${{ !github.event.pull_request.draft }}
-    name: "Integration Tests 3/3"
-    runs-on: penpot-runner-02
-    container:
-      image: penpotapp/devenv:latest
-      volumes:
-        - /tmp/.m2:/root/.m2
-        - /tmp/.gitlibs:/root/.gitlibs
-
-    needs: build-integration
-
-    steps:
-      - name: Checkout Repository
-        uses: actions/checkout@v6
-
-      - name: Restore Cache
-        uses: actions/cache/restore@v5
-        with:
-          key: "integration-bundle-${{ github.sha }}"
-          path: frontend/resources/public
-
-      - name: Run Tests
-        working-directory: ./frontend
-        run: |
-          ./scripts/test-e2e --shard="3/3";
-
-      - name: Upload test result
-        uses: actions/upload-artifact@v7
-        if: always()
-        with:
-          name: integration-tests-result-3
-          path: frontend/test-results/
-          overwrite: true
-          retention-days: 3
@@ -9,10 +9,13 @@
 .clj-kondo
 .cpcache
 .lsp
+.env
 .nrepl-port
 .nyc_output
 .rebel_readline_history
 .repl
+opencode.json
+.opencode/package-lock.json
 /*.jpg
 /*.md
 !CHANGES.md
@@ -30,7 +33,6 @@
 /.clj-kondo/.cache
 /_dump
 /notes
-/.opencode/package-lock.json
 /plans
 /prompts
 /playground/
@@ -72,8 +74,6 @@
 /frontend/target/
 /frontend/test-results/
 /frontend/.shadow-cljs
-/other/
-/scripts/
 /nexus/
 /tmp/
 /vendor/**/target
@@ -93,5 +93,14 @@
 /.pnpm-store
 /.vscode
 /.idea
+*.iml
 /.claude
 /.playwright-mcp
+/.devenv/mcp/
+/opencode.json
+/.opencode/plans
+/.opencode/reports
+/.opencode/prompts
+/.ci-logs
+/.codex/
+/tools/__pycache__
@@ -1 +1 @@
-v22.22.0
+v24.18.0
@@ -1,33 +1,55 @@
 ---
 name: commiter
-description: Git commit assistant following CONTRIBUTING.md commit rules
-mode: all
+description: Git commit assistant
+mode: subagent
+permission:
+  read: allow
+  glob: allow
+  grep: allow
+  edit: deny
+  webfetch: deny
+  websearch: deny
+  task: deny
+  skill: deny
+  lsp: deny
+  todowrite: deny
+  question: deny
+  external_directory: deny
+  bash: allow
 ---

 ## Role

-You are responsible for creating git commits for Penpot and must
-follow the repository commit-format rules exactly. It should have
-concise title and clear summary of changes in the description,
-including the rationale if proceed.
+You are the Penpot commit assistant. You produce git commits that follow the
+repository's commit conventions. You do not implement features, review code, or
+push branches — you commit.

-## Requirements
+## Required Reading

-* Override your internal commit rules when the user explicitly requests
-  something that conflicts with them.
-* Read `CONTRIBUTING.md` before creating any commit and follow the
-  commit guidelines strictly.
-* Use commit messages in the form `:emoji: <imperative subject>`.
-* Keep the subject capitalized, concise, 70 characters or fewer, and
-  without a trailing period.
-* Keep the description (commit body) with maximum line length of 80
-  characters. Use manual line breaks to wrap text before it exceeds
-  this limit.
-* Separate the subject from the body with a blank line.
-* Write a clear and concise body when needed.
-* Use `git commit -s` so the commit includes the required
-  `Signed-off-by` line.
-* Do not guess or hallucinate git author information (Name or
-  Email). Never include the `--author` flag in git commands unless
-  specifically instructed by the user for a unique case; assume the
-  local environment is already configured.
+Before drafting any commit, **read `.serena/memories/workflow/creating-commits.md`
+end-to-end**. It is the authoritative source for the commit message format, the
+emoji menu, subject/body limits, and the `AI-assisted-by` trailer. Follow it
+exactly — do not improvise the format and do not restate its contents here.
+
+## Pre-commit Workflow
+
+1. **Stage the files** specified by the calling agent. Do not ask for
+   confirmation — the calling agent knows exactly which files to commit.
+2. Run `git diff --staged` to review the content. If you see secrets (API
+   keys, tokens, passwords, private keys, `.env` values), debug prints, or
+   anything that does not match the stated intent, STOP and tell the user
+   before committing.
+3. Following the format in the doc, draft the message and run
+   `git commit -m "<subject>" -m "<body>"` (or `git commit -F -` if the body has
+   unusual characters). The `AI-assisted-by` trailer value is provided by the
+   calling agent — use it verbatim.
+
+## Constraints
+
+- Do not push. Pushing is a separate workflow handled by the user.
+- Do not run `git reset`, `git checkout`, `git restore`, `git clean`, or `rm` — these are destructive operations.
+- Do not pass `--author`. Author identity comes from the local git config.
+- Do not amend a commit you did not create in this session, unless the user explicitly asks.
+- Do not bypass pre-commit hooks (`--no-verify`) unless the user explicitly asks.
+- Do not add untracked files that were not created in this session.
+- Do not ask questions. The calling agent provides all necessary information. If something is unclear, proceed with what you know and note any assumptions in your response.
@@ -1,37 +0,0 @@
---
-name: Penpot Engineer
-description: Senior Full-Stack Software Engineer
-mode: primary
---
-
-Role: You are a high-autonomy Senior Full-Stack Software Engineer working on
-Penpot, an open-source design tool. You have full permission to navigate the
-codebase, modify files, and execute commands to fulfill your tasks. Your goal is
-to solve complex technical tasks with high precision while maintaining a strong
-focus on maintainability and performance.
-
-Tech stack: Clojure (backend), ClojureScript (frontend/exporter), Rust/WASM
-(render-wasm), TypeScript (plugins/mcp), SCSS.
-
-Requirements:
-
-* Read the root `AGENTS.md` to understand the repository and application
-  architecture. Then read the `AGENTS.md` **only** for each affected module.
-  Not all modules have one — verify before reading.
-* Before writing code, analyze the task in depth and describe your plan. If the
-  task is complex, break it down into atomic steps.
-* When searching code, prefer `ripgrep` (`rg`) over `grep` — it respects
-  `.gitignore` by default.
-* Do **not** touch unrelated modules unless the task explicitly requires it.
-* Only reference functions, namespaces, or APIs that actually exist in the
-  codebase. Verify their existence before citing them. If unsure, search first.
-* Be concise and autonomous — avoid unnecessary explanations.
-* After making changes, run the applicable lint and format checks for the
-  affected module before considering the work done (see module `AGENTS.md` for
-  exact commands).
-* Make small and logical commits following the commit guideline described in
-  `CONTRIBUTING.md`. Commit only when explicitly asked.
- Do not guess or hallucinate git author information (Name or Email). Never include the
-  `--author` flag in git commands unless specifically instructed by the user for a unique
-  case; assume the local environment is already configured. Allow git commit to
-  automatically pull the identity from the local git config `user.name` and `user.email`.
@@ -1,64 +0,0 @@
---
-name: Penpot Planner
-description: Software architect for planning and analysis only
-mode: primary
-permission:
-  edit: ask
---
-
-# Penpot Planner
-
-## Role
-
-You are a Senior Software Architect working on Penpot, an open-source design
-tool. Your sole responsibility is planning and analysis — you do NOT write,
-modify any code.
-
-You help users understand the codebase, design solutions, and create detailed
-implementation plans that other agents or developers can execute. Document
-everything they need to know: which files to touch for each task, code, testing,
-docs they might need to check, how to test it. Give them the whole plan as
-bite-sized tasks. DRY. YAGNI. TDD. Frequent commits.
-
-Do **not** suggest commit messages or commit names anywhere in your plans or
-responses — committing is the developer's responsibility.
-
-Assume they are a skilled developer, but know almost nothing about our toolset
-or problem domain. Assume they don't know good test design very well.
-
-## Requirements
-
-* Analyze the codebase architecture and identify affected modules.
-* Read `AGENTS.md` files (root and per-module) to understand structure and
-  conventions.
-* Search code using `ripgrep` skill (`rg`) to trace dependencies, find patterns,
-  and understand existing implementations.
-* Break down complex features or bugs into atomic, actionable steps.
-* Propose solutions with clear rationale, trade-offs, and sequencing.
-* Identify risks, edge cases, and testing considerations.
-
-Save plans to: plans/YYYY-MM-DD-<plan-one-line-title>.md
-
-## Constraints
-
-* You are **read-only** — never create, edit, or delete files.
-* You do **not** run builds, tests, linters, or any commands that modify state.
-* You do **not** create git commits or interact with version control.
-* You do **not** execute shell commands beyond read-only searches (`rg`, `ls`,
-  `find`, `cat`).
-* Your output is a structured plan or analysis, ready for handoff to an
-  engineer agent or developer.
-
-## Output format
-
-When producing a plan, structure it as:
-
-1. **Context** — What is the problem or feature request?
-2. **Affected modules** — Which parts of the codebase are involved?
-3. **Approach** — Step-by-step implementation plan with file paths and
-   function names where applicable.
-4. **Risks & considerations** — Edge cases, performance implications, breaking
-   changes.
-5. **Testing strategy** — How to verify the implementation works correctly.
-
-
@@ -1,59 +0,0 @@
---
-name: Prompt Assistant
-description: Refines and improves prompts for maximum clarity and effectiveness
-mode: all
---
-
-# Prompt Assistant
-
-## Role
-
-You are an expert Prompt Engineer with strong knowledge of
-penpot. Your sole responsibility is to take a prompt provided by the
-user and transform it into the most effective, clear, and
-well-structured version possible — ready to be used with any AI model.
-
-## Requirements
-
-* You do NOT execute tasks. You do NOT write code. You only design and
-  refine prompts
-* Read the root `AGENTS.md` to understand the repository and application
-  architecture. Then read the `AGENTS.md` **only** for each affected module.
-* Analyze the original prompt: identify its intent, target audience,
-   ambiguities, missing context, and structural weaknesses
-* Ask clarifying questions if the intent is unclear or if critical
-  information is missing (e.g. target model, expected output format,
-  tone, constraints). Keep questions concise and grouped
-* Rewrite the prompt using prompt engineering best practices
-
-
-## Prompt Engineering Principles
-
-Apply these techniques when refining prompts:
-
- **Be specific and explicit**: Replace vague instructions with precise ones.
- **Set the context**: Include background information the model needs to
-  perform well.
- **Specify the output format**: State the desired structure, length, tone,
-  or format (e.g. bullet list, JSON, step-by-step).
- **Add constraints**: Include what the model should avoid or not do.
- **Use examples** (few-shot): When applicable, suggest adding examples to
-  anchor the model's behaviour.
- **Break down complexity**: Split multi-step tasks into clear numbered steps.
- **Avoid ambiguity**: Remove pronouns and references that could be
-  misinterpreted.
- **Chain of thought**: For reasoning tasks, include "Think step by step."
-
-## Constraints
-
- Do NOT execute the prompt yourself.
- Do NOT answer the question inside the prompt.
- Do NOT add unnecessary verbosity — prompts should be as short as they can
-  be while remaining complete.
- Always preserve the user's original intent.
-
-## Output
-
-Refined Prompt: The improved, ready-to-use prompt. Print it for
-immediate use and save it to
-prompts/YYYY-MM-DD-N-<prompt-one-line-title>.md for future use.
@@ -0,0 +1,43 @@
+---
+description: Execute a ready plan end-to-end — create a GitHub issue, branch issue-NNNN, implement the plan, then commit via the commiter subagent
+agent: build
+---
+
+# Implement Plan
+
+This command is run once a plan is ready (for example, from plan mode). Execute
+the plan already prepared in the current session context — it does not take
+extra arguments. Follow these steps in order.
+
+## 1. Create the issue
+
+Use the **`create-issue`** skill, following the *Creating Issues from Draft Body*
+flow in `mem:workflow/creating-issues`. Derive the issue title and body from the
+plan. Capture the new issue's number — call it **NNNN** (needed for the branch
+name and the commit reference).
+
+## 2. Create the branch
+
+Create and switch to a branch named after the issue:
+
+```
+git checkout -b issue-NNNN
+```
+
+(Replace NNNN with the issue number from step 1.)
+
+## 3. Execute the plan
+
+Implement the prepared plan from the session context. Work methodically, keeping
+changes focused on what the issue requires. Do not commit — the commit happens in
+step 4.
+
+## 4. Commit with the commiter subagent
+
+After the implementation is complete, delegate the commit to the **`commiter`**
+subagent. Give it a brief summary of what was implemented and why, the issue
+reference (`issue-NNNN`), and the model name you are running as so it sets the
+`AI-assisted-by` trailer correctly. The subagent owns the commit format and
+conventions.
+
+Do not push. Pushing is handled separately by the user.
@@ -0,0 +1,17 @@
+Act as a senior software engineer and perform a thorough code review.
+
+## Instructions
+
+1. Load the **`code-review-and-quality`** skill — it defines the five axes, core principles (DRY, KISS, YAGNI), severity taxonomy, and output format.
+2. Determine the diff or code to review from the provided context.
+3. Read the diff and the surrounding context for each changed file.
+4. Review across all five axes: correctness, readability, architecture, security, performance.
+5. Produce the review using the **Review Output** format from the skill (Summary → Critical/High → Other Findings → Refactoring → Testing Recommendations → Positive Observations → Final Verdict).
+6. For each finding: state the severity (Critical / High / Medium / Low / Suggestion), identify the file and line, describe failure circumstances, and propose a concrete fix.
+7. Do not invent problems. Every finding must be real and actionable.
+
+Do not modify any code and do not create a commit — this command only reviews.
+
+## Context
+
+$ARGUMENTS
@@ -0,0 +1,299 @@
+import { tool } from "@opencode-ai/plugin"
+import path from "path"
+import { spawn } from "child_process"
+
+const penpotPsqlTool = tool({
+  description:
+    "Execute a SQL command against the Penpot database. Uses the defaults from scripts/psql.",
+
+  args: {
+    sql: tool.schema
+      .string()
+      .describe("SQL command to execute"),
+
+    test: tool.schema
+      .boolean()
+      .describe("Use the penpot_test database")
+      .optional(),
+  },
+
+  async execute(args, context) {
+    const host = process.env.PENPOT_DB_HOST || "postgres"
+    const user = process.env.PENPOT_DB_USER || "penpot"
+    const db = args.test
+      ? "penpot_test"
+      : process.env.PENPOT_DB_NAME || "penpot"
+    const password = process.env.PENPOT_DB_PASSWORD || "penpot"
+
+    const psqlArgs = ["-h", host, "-U", user, "-d", db, "-c", args.sql]
+
+    return new Promise((resolve) => {
+      let stdout = ""
+      let stderr = ""
+
+      const proc = spawn("psql", psqlArgs, {
+        cwd: context.worktree,
+        env: { ...process.env, PGPASSWORD: password },
+      })
+
+      proc.stdout.on("data", (data) => {
+        stdout += data.toString()
+      })
+
+      proc.stderr.on("data", (data) => {
+        stderr += data.toString()
+      })
+
+      proc.on("error", (error) => {
+        resolve(`Error: ${error.message}`)
+      })
+
+      proc.on("close", (exitCode) => {
+        const output =
+          exitCode === 0
+            ? stdout.trim() || "Query executed successfully"
+            : `Error (exit ${exitCode}): ${
+                (stderr || stdout).trim() || "No error output"
+              }`
+        resolve(output)
+      })
+    })
+  },
+})
+
+const parenRepairTool = tool({
+  description:
+    "Fix mismatched parentheses/braces in Clojure files (.clj, .cljs, .cljc) then reformat with cljfmt.",
+
+  args: {
+    // A string is used instead of an array so OpenCode displays it
+    // in the generic tool invocation.
+    files: tool.schema
+      .string()
+      .describe(
+        "Comma-separated file paths to fix, for example: frontend/src/app/config.cljs, backend/src/core.clj",
+      )
+      .optional(),
+
+    code: tool.schema
+      .string()
+      .describe("Code string to fix via stdin")
+      .optional(),
+  },
+
+  async execute(args, context) {
+    const script = path.join(context.worktree, "scripts/paren-repair")
+
+    const files = args.files
+      ? args.files
+          .split(",")
+          .map((file) => file.trim())
+          .filter(Boolean)
+      : []
+
+    const paramInfo =
+      files.length > 0
+        ? `files=[${files.join(", ")}]`
+        : args.code !== undefined
+          ? `code=(${args.code.length} chars)`
+          : "none"
+
+    return new Promise((resolve) => {
+      const childArgs =
+        files.length > 0
+          ? [script, ...files]
+          : [script]
+
+      const proc = spawn("bb", childArgs, {
+        cwd: context.worktree,
+      })
+
+      let stdout = ""
+      let stderr = ""
+
+      proc.stdout.on("data", (data) => {
+        stdout += data.toString()
+      })
+
+      proc.stderr.on("data", (data) => {
+        stderr += data.toString()
+      })
+
+      proc.on("error", (error) => {
+        resolve(`Error: ${error.message}`)
+      })
+
+      proc.on("close", (exitCode) => {
+        const output =
+          exitCode === 0
+            ? stdout.trim() || "No changes needed"
+            : `Error (exit ${exitCode}): ${
+                (stderr || stdout).trim() || "No error output"
+              }`
+
+        resolve(output)
+      })
+
+      // Close stdin in all cases so the process cannot wait indefinitely.
+      if (args.code !== undefined) {
+        proc.stdin.end(args.code)
+      } else {
+        proc.stdin.end()
+      }
+    })
+  },
+})
+
+export default async function plugin() {
+  return {
+    tool: {
+      "paren-repair": parenRepairTool,
+      "penpot-psql": penpotPsqlTool,
+    },
+  }
+}
+
+
+
+
+
+
+
+
+
+
+
+// import { tool } from "@opencode-ai/plugin"
+// import path from "path"
+// import { spawn } from "child_process"
+
+// function formatFiles(files) {
+//   if (files.length === 0) return "stdin"
+
+//   // Keep the visible tool title reasonably short.
+//   if (files.length <= 3) return files.join(", ")
+
+//   return `${files.slice(0, 3).join(", ")} (+${files.length - 3} more)`
+// }
+
+// const parenRepairTool = tool({
+//   description:
+//     "Fix mismatched parentheses/braces in Clojure files, then reformat with cljfmt.",
+
+//   args: {
+//     files: tool.schema
+//       .array(tool.schema.string())
+//       .describe("Array of file paths to fix")
+//       .optional(),
+
+//     code: tool.schema
+//       .string()
+//       .describe("Code string to fix via stdin")
+//       .optional(),
+//   },
+
+//   async execute(args, context) {
+//     const script = path.join(context.worktree, "scripts/paren-repair")
+
+//     const files = (args.files ?? []).map((file) => {
+//       const absolute = path.isAbsolute(file)
+//         ? file
+//         : path.resolve(context.worktree, file)
+
+//       return path.relative(context.worktree, absolute)
+//     })
+
+//     const targetSummary =
+//       files.length > 0
+//         ? formatFiles(files)
+//         : args.code !== undefined
+//           ? `stdin (${args.code.length} chars)`
+//           : "no input"
+
+//     // This updates the tool-call title immediately, while it is running.
+//     await context.metadata({
+//       title: `Paren repair: ${targetSummary}`,
+//       metadata: {
+//         files,
+//         codeChars: args.code?.length,
+//       },
+//     })
+
+//     const childArgs =
+//       args.files && args.files.length > 0
+//         ? [script, ...args.files]
+//         : [script]
+
+//     return new Promise((resolve) => {
+//       const proc = spawn("bb", childArgs, {
+//         cwd: context.worktree,
+//       })
+
+//       let stdout = ""
+//       let stderr = ""
+
+//       if (args.code !== undefined) {
+//         proc.stdin.end(args.code)
+//       }
+
+//       proc.stdout.on("data", (data) => {
+//         stdout += data.toString()
+//       })
+
+//       proc.stderr.on("data", (data) => {
+//         stderr += data.toString()
+//       })
+
+//       proc.on("close", (exitCode) => {
+//         const successful = exitCode === 0
+
+//         const commandOutput = successful
+//           ? stdout.trim() || "No changes needed"
+//           : `Error (exit ${exitCode}): ${(stderr || stdout).trim()}`
+
+//         const parameterOutput =
+//           files.length > 0
+//             ? `Files passed:\n${files.map((file) => `- ${file}`).join("\n")}`
+//             : args.code !== undefined
+//               ? `Input passed through stdin: ${args.code.length} characters`
+//               : "No files or stdin input were passed"
+
+//         resolve({
+//           title: `Paren repair: ${targetSummary}`,
+//           output: `${parameterOutput}\n\n${commandOutput}`,
+//           metadata: {
+//             files,
+//             codeChars: args.code?.length,
+//             exitCode,
+//             successful,
+//           },
+//         })
+//       })
+
+//       proc.on("error", (error) => {
+//         resolve({
+//           title: `Paren repair failed: ${targetSummary}`,
+//           output: [
+//             files.length > 0
+//               ? `Files passed:\n${files.map((file) => `- ${file}`).join("\n")}`
+//               : `Input: ${targetSummary}`,
+//             `Failed to start bb: ${error.message}`,
+//           ].join("\n\n"),
+//           metadata: {
+//             files,
+//             codeChars: args.code?.length,
+//             successful: false,
+//           },
+//         })
+//       })
+//     })
+//   },
+// })
+
+// export default async function plugin() {
+//   return {
+//     tool: {
+//       "paren-repair": parenRepairTool,
+//     },
+//   }
+// }
@@ -1,90 +0,0 @@
---
-name: backport-commit
-description: Port changes from a specific Git commit to the current branch by manually applying the diff, avoiding cherry-pick when it would introduce complex conflicts.
---
-
-# Backport Commit
-
-Port changes from a specific Git commit to the current branch by manually
-applying the diff, avoiding `git cherry-pick` when it would introduce
-complex conflicts.
-
-## When to Use
-
-Use this skill whenever the user asks to backport a commit, especially when:
-
- The commit touches multiple modules or files with significant divergence
- `git cherry-pick` is explicitly ruled out ("do not use cherry-pick")
- The target commit is old enough that conflicts are likely
- The commit introduces both source changes AND new files (tests, etc.)
- You need full control over how each hunk is applied
-
-## Workflow
-
-### 1. Identify the target commit
-
-```bash
-# Verify the commit exists and understand what it does
-git log --oneline -1 <commit-sha>
-
-# Get the full diff (including new/deleted files)
-git show <commit-sha>
-
-# Capture the original commit message for later reuse
-git log --format='%B' -1 <commit-sha>
-```
-
-### 2. Identify affected modules
-
-From the file paths in the diff, determine which Penpot modules are affected
-(frontend, backend, common, render-wasm, etc.) and read their `AGENTS.md`
-files **before** making any changes. If a module has no `AGENTS.md`, skip
-that step — verify with `ls <module>/AGENTS.md` first.
-
-### 3. Read the current state of each affected file
-
-For every file the diff touches, read the current version on disk to understand
-context and ensure correct placement before editing.
-
-### 4. Apply changes manually (the core of this approach)
-
-Process every hunk in the diff using the appropriate tool:
-
-| Diff action | Tool to use |
-|-------------|-------------|
-| Modify existing file | `edit` — use enough surrounding context in `oldString` to uniquely match the location |
-| Add new file | `write` — include proper license header and namespace conventions matching project style |
-| Delete file | `bash rm <path>` |
-| Rename/move file | `bash mv <old> <new>`, then apply any content changes with `edit` |
-
-> **Tip:** Group nearby hunks from the same file into a single `edit` call.
-> Use separate calls when hunks are far apart to keep `oldString` short and
-> unambiguous.
-
-Repeat until **all** hunks in the diff are ported.
-
-### 5. Validate
-
-Run **lint**, **check-fmt**, and **tests** for every affected module (see each
-module's `AGENTS.md` for the exact commands). If the formatter auto-fixes
-indentation, verify the logic is still semantically correct. All checks must
-pass before moving on.
-
-### 6. Port the changelog entry (if any)
-
-If the original commit added or modified a `CHANGES.md` entry, port that entry
-too — adapting wording and version references for the target branch.
-
-### 7. Commit
-
-Ask the `commiter` sub-agent to create a commit. Stage all relevant files
-(exclude unrelated untracked files) and provide the original commit message as
-a reference, adapting it as needed for the target branch context.
-
-## Key Principles
-
- **Context matters** — always read files before editing; never guess
-  indentation or surrounding code
- **Lint + format + test** — never skip validation before committing
- **Preserve intent** — keep the original commit message meaning; the
-  `commiter` agent handles formatting
@@ -0,0 +1,255 @@
+---
+name: code-review-and-quality
+description: Conducts multi-axis code review. Use before merging any change. Use when reviewing code written by yourself, another agent, or a human. Use when you need to assess code quality across multiple dimensions before it enters the main branch.
+---
+
+# Code Review and Quality
+
+## Overview
+
+Multi-dimensional code review with quality gates. Every change gets reviewed before merge — no exceptions. Review covers five axes: correctness, readability, architecture, security, and performance.
+
+**The approval standard:** Approve a change when it definitely improves overall code health, even if it isn't perfect. Perfect code doesn't exist — the goal is continuous improvement. Don't block a change because it isn't exactly how you would have written it. If it improves the codebase and follows the project's conventions, approve it.
+
+## When to Use
+
+- Before merging any PR or change
+- After completing a feature implementation
+- When another agent or model produced code you need to evaluate
+- When refactoring existing code
+- After any bug fix (review both the fix and the regression test)
+
+## Core Principles
+
+These principles underpin every axis. When in doubt, default to them.
+
+- **DRY (Don't Repeat Yourself):** Every piece of knowledge has one authoritative representation. If the same logic appears in two places, extract it into a shared helper, model, or type. Reviewers: flag duplicated logic as a required change — it's not "just similar," it's drift that will diverge.
+- **KISS (Keep It Simple, Stupid):** The simplest solution that works is the best solution. Complexity must earn its place. Reviewers: if you need more than one sentence to explain what a piece of code does, it's too complex — push for simplification before merge.
+- **YAGNI (You Aren't Gonna Need It):** Don't add abstractions, hooks, or generalizations for hypothetical future use cases. Generalize on the third occurrence, not the first. Reviewers: delete speculative generality.
+- **Don't invent problems:** Do not manufacture issues to produce more feedback. Every finding must be a real risk, a real readability barrier, or a real architectural concern — not a hypothetical or a stylistic preference disguised as a problem.
+
+## The Five-Axis Review
+
+Every review evaluates code across these dimensions.
+
+### 1. Correctness
+
+Does the code do what it claims to do?
+
+- Does it match the spec or task requirements?
+- Are edge cases handled (null, empty, boundary values)?
+- Are error paths handled (not just the happy path)?
+- Does it pass all tests? Are the tests actually testing the right things?
+- Are there off-by-one errors, race conditions, or state inconsistencies?
+
+### 2. Readability & Simplicity
+
+Can another engineer (or agent) understand this code without the author explaining it?
+
+- Are names descriptive and consistent with project conventions? (No `temp`, `data`, `result` without context)
+- Is the control flow straightforward (avoid nested ternaries, deep callbacks)?
+- Are there any "clever" tricks that should be simplified?
+- **KISS check:** Is this the simplest approach that solves the problem? A 20-line straightforward function beats a 5-line clever one that requires a comment to explain.
+- Could this be done in fewer lines? (1000 lines where 100 suffice is a failure)
+- Are abstractions earning their complexity? (Don't generalize until the third use case)
+- Is a new conditional bolted onto an unrelated flow? Push the logic into its own helper, state, or policy.
+- Do repeated conditionals on the same shape appear? They signal a missing model or dispatcher.
+- Are there dead code artifacts: no-op variables, backwards-compat shims, or `// removed` comments?
+
+### 3. Architecture
+
+Does the change fit the system's design?
+
+- Does it follow existing patterns or introduce a new one? If new, is it justified?
+- Does it maintain clean module boundaries?
+- **DRY check:** Is there existing code that does the same thing? Reuse the canonical helper instead of writing a near-duplicate. If two branches do nearly the same thing, collapse them.
+- Are dependencies flowing in the right direction (no circular dependencies)?
+- Is the abstraction level appropriate (not over-engineered, not too coupled)?
+- Does this refactor reduce complexity or just relocate it? Count the concepts a reader must hold. Prefer the restructuring that makes whole branches disappear over one that re-centralizes the same logic. Prefer deleting an abstraction to polishing it.
+- Is feature-specific logic leaking into a shared or general-purpose module?
+- Are type boundaries explicit? Question gratuitous `any`/`unknown`/optional/casts and silent fallbacks.
+- **Structural remedies:** When you flag a problem, propose the move — not just the problem. Replace conditionals with dispatchers, collapse duplicate branches, separate orchestration from business logic, extract helpers, split large files. Prefer the remedy that removes moving pieces over one that spreads the same complexity around.
+
+### 4. Security
+
+For detailed security guidance, see `security-and-hardening`.
+
+- Is user input validated and sanitized?
+- Are secrets kept out of code, logs, and version control?
+- Is authentication/authorization checked where needed?
+- Are SQL queries parameterized (no string concatenation)?
+- Are outputs encoded to prevent XSS?
+- Are dependencies from trusted sources with no known vulnerabilities?
+- Is data from external sources (APIs, logs, user content, config files) treated as untrusted?
+
+### 5. Performance
+
+- Any N+1 query patterns?
+- Any unbounded loops or unconstrained data fetching?
+- Any synchronous operations that should be async?
+- Any unnecessary re-renders in UI components?
+- Any missing pagination on list endpoints?
+- Any large objects created in hot paths?
+
+## Review Process
+
+1. **Understand the intent** — What is this change trying to accomplish? What spec or task does it implement?
+2. **Review tests first** — Tests reveal intent and coverage. Do they test behavior, not implementation details? Are edge cases covered?
+3. **Review the implementation** — Walk through each file with the five axes in mind.
+4. **Categorize findings** — Label every comment with its severity:
+
+| Prefix | Meaning | Author Action |
+|--------|---------|---------------|
+| **Critical:** | Blocks merge | Security vulnerability, data loss, broken functionality |
+| **High:** | Required change | Must address before merge |
+| **Medium:** | Should fix | Strongly recommended, not a blocker |
+| **Low:** | Minor, optional | Author may ignore — formatting, style preferences |
+| **Suggestion:** | Worth considering | Not required, but improves the code |
+
+For each finding, describe the circumstances under which it could fail: specific inputs, load conditions, timing, or user actions that trigger the problem. "This crashes when input is null" is actionable; "this might crash" is not.
+
+Lead with what matters: correctness and security first, then structural issues, then everything else. A few high-conviction comments beat a long list.
+
+5. **Verify the verification** — What tests were run? Did the build pass? Was the change tested manually? Screenshots for UI changes?
+
+## Review Output
+
+Structure every review using this format:
+
+### Summary
+
+Briefly explain what the code does and give an overall assessment.
+
+### Critical and High-Priority Issues
+
+List problems that could cause security incidents, data loss, crashes, incorrect behavior, or major performance degradation. For each: state the severity, identify the file/function/code section, explain why it's a problem, describe failure circumstances, and provide a concrete improvement with corrected code when useful.
+
+### Other Findings
+
+List medium- and low-priority issues, including maintainability and design concerns.
+
+### Suggested Refactoring
+
+Provide focused code changes or revised snippets. Preserve existing behavior unless a behavior change is explicitly justified.
+
+### Testing Recommendations
+
+Identify missing tests and describe specific test cases, including edge cases and failure scenarios.
+
+### Positive Observations
+
+Mention implementation choices that are clear, safe, efficient, or well designed. This is not fluff — it reinforces good patterns and tells the author what to keep doing.
+
+### Final Verdict
+
+Choose one:
+
+- **Approve** — Ready to merge
+- **Approve with minor changes** — Good to merge after addressing low/medium issues
+- **Request changes** — Critical or high issues must be resolved before merge
+
+## Change Sizing
+
+Small, focused changes are easier to review, faster to merge, and safer to deploy.
+
+```
+~100 lines changed   → Good. Reviewable in one sitting.
+~300 lines changed   → Acceptable if it's a single logical change.
+~1000 lines changed  → Too large. Split it.
+```
+
+**Watch file size, not just diff size.** Around 1000 *total* lines in a single file is a common inspection signal. When a change materially grows an already-large file, decompose first.
+
+**Splitting strategies:**
+
+| Strategy | How | When |
+|----------|-----|------|
+| **Stack** | Submit a small change, start the next one based on it | Sequential dependencies |
+| **By file group** | Separate changes for groups needing different reviewers | Cross-cutting concerns |
+| **Horizontal** | Create shared code/stubs first, then consumers | Layered architecture |
+| **Vertical** | Break into smaller full-stack slices of the feature | Feature work |
+
+**Separate refactoring from feature work.** A change that refactors and adds new behavior is two changes — submit them separately.
+
+## Change Descriptions
+
+- **First line:** Short, imperative, standalone. "Delete the FizzBuzz RPC" not "Deleting the FizzBuzz RPC."
+- **Body:** What is changing and why. Include context and reasoning not visible in the code itself.
+- **Anti-patterns:** "Fix bug," "Fix build," "Add patch," "Phase 1."
+
+## Dependencies
+
+Before adding any dependency:
+
+1. Does the existing stack solve this? (Often it does.)
+2. How large is the dependency? (Check bundle impact.)
+3. Is it actively maintained? (Check last commit, open issues.)
+4. Does it have known vulnerabilities? (`npm audit`)
+5. What's the license? (Must be compatible with the project.)
+
+**Rule:** Prefer standard library and existing utilities over new dependencies. Every dependency is a liability.
+
+**Upgrading dependencies:**
+
+- Read the changelog, not just the version number. Semver is a promise the maintainer may not have kept.
+- One dependency per change. When a bulk bump breaks the build, you've lost which package did it.
+- Let the tests decide — a green suite before *and* after, not just "it installed."
+- Review the lockfile diff, not just `package.json`. Commit it and never hand-edit it.
+
+For supply-chain risk triage, follow the `security-and-hardening` skill.
+
+## Common Rationalizations
+
+| Rationalization | Reality |
+|---|---|
+| "It works, that's good enough" | Working code that's unreadable, insecure, or architecturally wrong creates debt that compounds. |
+| "I wrote it, so I know it's correct" | Authors are blind to their own assumptions. Every change benefits from another set of eyes. |
+| "We'll clean it up later" | Later never comes. The review is the quality gate — use it. |
+| "AI-generated code is probably fine" | AI code needs more scrutiny, not less. It's confident and plausible, even when wrong. |
+| "The tests pass, so it's good" | Tests are necessary but not sufficient. They don't catch architecture, security, or readability problems. |
+| "The refactor makes it cleaner" | Relocating complexity isn't reducing it. If the reader still holds the same number of concepts, the structure didn't improve. |
+| "It's only a small addition to this file" | Small diffs still push files past healthy size and bolt branches onto unrelated flows. |
+| "It's just a version bump" | A bump is a behavior change you didn't write. Read the changelog. |
+| "I'll upgrade everything in one PR" | A bulk bump hides which package broke the build. One per change. |
+| "It's duplicated but it's only two places" | Two becomes three becomes five. Extract now, before the copies diverge. |
+| "The abstraction is future-proof" | YAGNI. Delete speculative generality — generalize on the third occurrence, not the first. |
+| "It's clever but efficient" | Cleverness is a readability tax. If it needs a comment to understand, simplify it. |
+
+## Red Flags
+
+- PRs merged without any review
+- Review that only checks if tests pass (ignoring other axes)
+- "LGTM" without evidence of actual review
+- Security-sensitive changes without security-focused review
+- Large PRs that are "too big to review properly" (split them)
+- No regression tests with bug fix PRs
+- Accepting "I'll fix it later" — it never happens
+- A refactor that moves code around without reducing the number of concepts a reader must hold
+- New conditionals scattered into unrelated code paths (a missing abstraction)
+- A bespoke helper that duplicates an existing canonical one
+- A bulk "bump dependencies" PR with no changelog review
+
+## Verification
+
+After review is complete:
+
+- [ ] All Critical issues are resolved
+- [ ] All Required (no-prefix) changes are resolved or explicitly deferred with justification
+- [ ] Tests pass
+- [ ] Build succeeds
+- [ ] The verification story is documented (what changed, how it was verified)
+- [ ] Dependency upgrades reviewed against changelog, isolated per package, verified by green suite
+
+## Multi-Model Review Pattern
+
+Use different models for different review perspectives:
+
+```
+Model A writes the code → Model B reviews → Model A addresses feedback → Human makes the final call
+```
+
+Different models have different blind spots.
+
+## See Also
+
+- For detailed security review guidance, see `security-and-hardening`
@@ -0,0 +1,27 @@
+---
+name: create-issue
+description: Create or update GitHub issues (from PR, from draft body, retitle existing). Routes to the canonical flow in `mem:workflow/creating-issues`.
+---
+
+# Skill: create-issue
+
+Entry point for all GitHub issue work. All rules (title derivation, metadata,
+body templates, Issue Type IDs), all flows, and all `gh` / GraphQL commands
+live in `mem:workflow/creating-issues` (file:
+`.serena/memories/workflow/creating-issues.md`). This skill routes to the
+right flow.
+
+## When to Use
+
+- **Create from PR** — PR exists; the issue is the changelog/release unit,
+  the PR is the implementation. Issue = WHAT, PR = HOW.
+  → memory section **Creating Issues from PRs**
+- **Create from draft body** — Taiga story, user report, discussion; no PR
+  yet.
+  → memory section **Creating Issues from Draft Body**
+- **Retitle existing issue** — current title is vague, prefixed, or stale.
+  → memory section **Retitling an Existing Issue**
+
+Everything else (title derivation, metadata policy, body templates, Issue
+Type IDs, create/verify/cleanup commands) lives in the memory — go to the
+matching section there.
@@ -0,0 +1,39 @@
+---
+name: create-pr
+description: Create or update a GitHub PR following Penpot conventions.
+---
+
+# Skill: create-pr
+
+Create or update a GitHub PR. Read and follow:
+- `mem:workflow/creating-prs` — title format, description structure, writing principles
+- `mem:workflow/creating-commits` — commit type emojis
+
+## When to Use
+
+- Creating a new PR from a feature branch
+- Updating an existing PR's title or description to match conventions
+
+## Prerequisites
+
+- `gh` CLI authenticated (`gh auth status`)
+
+## Commands
+
+**Create:**
+
+```bash
+gh pr create --repo penpot/penpot --title "<TITLE>" --body-file /tmp/pr-body.md
+```
+
+**Update:**
+
+```bash
+gh pr edit <NUMBER> --repo penpot/penpot --title "<TITLE>" --body-file /tmp/pr-body.md
+```
+
+**Verify:**
+
+```bash
+gh pr view <NUMBER> --repo penpot/penpot --json title,body
+```
@@ -1,230 +0,0 @@
---
-name: gh-issue-from-pr
-description: Create a user-facing GitHub issue from a PR, separating the WHAT from the HOW, with correct milestone, project, labels, and issue type.
---
-
-# Skill: gh-issue-from-pr
-
-Create a GitHub issue that captures the **WHAT** (user-facing feature or
-bug) from an existing PR that describes the **HOW** (implementation).
-Used when the project board needs an issue as the primary changelog/release unit.
-
-## When to Use
-
- Create a tracking issue from a PR for changelog purposes
- Extract the user-facing problem/feature from a PR's implementation details
- Assign milestone, project, labels, and issue type to a new issue derived from a PR
-
-## Prerequisites
-
- `gh` CLI authenticated (`gh auth status`)
- Permission to create issues and edit PRs in the target repository
-
-## Workflow
-
-### 1. Understand the PR
-
-```bash
-gh pr view <PR_NUMBER> --repo penpot/penpot \
-  --json title,body,author,labels,baseRefName,mergedAt,state,milestone
-```
-
-Identify:
-
- **WHAT** — user-facing problem or feature. Goes into the issue.
-  Describe symptoms and impact, not internal mechanisms.
- **HOW** — implementation details. These belong in the PR, not the issue.
-
-### 2. Determine metadata
-
-| Field | Source | Rule |
-|-------|--------|------|
-| **Title** | PR title | Rewrite from user perspective. Strip leading emoji prefixes (`:bug:`, `:sparkles:`, `:tada:`). Focus on observable behavior. Use imperative mood. |
-| **Labels** | PR labels | Copy user-facing labels (`bug`, `enhancement`, `community contribution`). Skip workflow labels (`backport candidate`, `team-qa`). |
-| **Milestone** | PR milestone | **Always copy what's on the PR.** Fetch with: `gh pr view <PR_NUMBER> --json milestone --jq '.milestone.title'` If the PR has no milestone, create the issue without one. |
-| **Project** | Always `Main` | Penpot uses the `Main` project (number 8) for all issues. |
-| **Body** | PR's user-facing section | Extract steps to reproduce or feature description. Omit internal details. Use templates below. |
-| **Issue Type** | PR labels / title | Map: `bug` label or `:bug:` title → `Bug`. `enhancement` label or `:sparkles:` title → `Enhancement`. Feature/epic → `Feature`. Default → `Task`. |
-
-### 3. Write the issue body
-
-**Bug template:**
-
-```markdown
-### Description
-
-<what breaks, what the user experiences>
-
-### Steps to reproduce
-
-1. <step 1>
-2. <step 2>
-
-### Expected behavior
-
-<what should happen instead>
-
-### Affected versions
-
-<version>
-```
-
-**Enhancement template:**
-
-```markdown
-### Description
-
-<what the user can now do that they couldn't before>
-
-### Use case
-
-<why this is useful, who benefits>
-
-### Affected versions
-
-<version>
-```
-
-### 4. Create the issue
-
-Write the body to a temp file to avoid shell quoting issues:
-
-```bash
-cat > /tmp/issue-body.md << 'ISSUE_BODY'
-<body content here>
-ISSUE_BODY
-```
-
-Create:
-
-```bash
-gh issue create \
-  --repo penpot/penpot \
-  --title "<Title>" \
-  --label "<label1>" \
-  --label "<label2>" \
-  --milestone "<milestone>" \
-  --project "Main" \
-  --body-file /tmp/issue-body.md
-```
-
-Output: `https://github.com/penpot/penpot/issues/<NUMBER>`
-
-### 5. Assign to the PR author
-
-Assign the issue to the PR author so they're responsible for it:
-
-```bash
-AUTHOR=$(gh pr view <PR_NUMBER> --repo penpot/penpot --json author --jq '.author.login')
-gh issue edit <ISSUE_NUMBER> --repo penpot/penpot --add-assignee "$AUTHOR"
-```
-
-### 6. Set the Issue Type
-
-`gh issue create` can't set the Issue Type directly. Use GraphQL.
-
-Get the issue's GraphQL node ID:
-
-```bash
-ISSUE_ID=$(gh api graphql -f query='
-query { repository(owner: "penpot", name: "penpot") {
-  issue(number: <ISSUE_NUMBER>) { id }
-}}' --jq '.data.repository.issue.id')
-```
-
-Issue Type IDs for the Penpot repo:
-
-| Type | ID |
-|------|----|
-| Bug | `IT_kwDOAcyBPM4AX5Nb` |
-| Enhancement | `IT_kwDOAcyBPM4B_IQN` |
-| Feature | `IT_kwDOAcyBPM4AX5Nf` |
-| Task | `IT_kwDOAcyBPM4AX5NY` |
-| Question | `IT_kwDOAcyBPM4B_IQj` |
-| Docs | `IT_kwDOAcyBPM4B_IQz` |
-
-Set it:
-
-```bash
-gh api graphql -f query='
-mutation {
-  updateIssue(input: {
-    id: "'"$ISSUE_ID"'"
-    issueTypeId: "<TYPE_ID>"
-  }) {
-    issue { number issueType { name } }
-  }
-}'
-```
-
-### 7. Verify
-
-```bash
-gh issue view <ISSUE_NUMBER> --repo penpot/penpot \
-  --json title,milestone,projectItems,labels \
-  --jq '{title, milestone: .milestone.title, projects: [.projectItems[].title], labels: [.labels[].name]}'
-
-gh api graphql -f query='
-query { repository(owner: "penpot", name: "penpot") {
-  issue(number: <ISSUE_NUMBER>) { issueType { name } }
-}}' --jq '.data.repository.issue.issueType.name'
-```
-
-### 8. Link the PR to the issue
-
-Append `Closes #<ISSUE_NUMBER>` to the PR body:
-
-```bash
-gh pr view <PR_NUMBER> --repo penpot/penpot --json body --jq '.body' > /tmp/pr-body.md
-printf "\n\nCloses #<ISSUE_NUMBER>\n" >> /tmp/pr-body.md
-gh pr edit <PR_NUMBER> --repo penpot/penpot --body-file /tmp/pr-body.md
-
-# Verify
-gh pr view <PR_NUMBER> --repo penpot/penpot --json body \
-  --jq '.body | test("Closes #<ISSUE_NUMBER>")'
-```
-
-**Note:** If the PR is already merged, `Closes` won't auto-close the issue
-— it only creates the "Development" sidebar link. This is the desired
-behavior since the issue is a tracking artifact.
-
-### 9. Clean up
-
-```bash
-rm -f /tmp/issue-body.md /tmp/pr-body.md
-```
-
-## Label rules
-
-| PR has | Issue gets |
-|--------|-----------|
-| `bug` | `bug` |
-| `enhancement` | `enhancement` |
-| `community contribution` | `community contribution` |
-| `backport candidate` | *(skip — workflow label)* |
-| `team-qa` | *(skip — workflow label)* |
-| No user-facing label | Infer from title: `:bug:` → `bug`, `:sparkles:` → `enhancement` |
-
-## Issue Type mapping
-
-| PR label(s) / title prefix | Issue Type |
-|----------------------------|-----------|
-| `bug` or `:bug:` | Bug |
-| `enhancement` or `:sparkles:` or `:tada:` | Enhancement |
-| Feature / epic | Feature |
-| Documentation | Docs |
-| None of the above | Task |
-
-## Key Principles
-
- **Issue = WHAT, PR = HOW.** Never put implementation details in the
-  issue body. The issue is for users, QA, and changelog readers.
- **Copy the milestone from the PR.** Don't guess based on branch names.
-  If the PR has no milestone, create the issue without one.
- **Set Issue Type via GraphQL** — `gh issue create` can't set it.
- **Link via PR body** — `Closes #<NUMBER>` creates the "Development"
-  sidebar link automatically.
- **One issue per PR** — even if a PR fixes multiple things, create a
-  single issue that summarizes the overall change.
- **Community attribution:** if the PR has the `community contribution`
-  label or the author is not a core team member, add the label to the issue.
@@ -1,31 +1,25 @@
 ---
 name: nrepl-eval
-description: Evaluate Clojure code via nREPL using the standalone tools/nrepl-eval.mjs CLI tool.
+description: Evaluate Clojure code via nREPL using the standalone scripts/nrepl-eval.mjs CLI tool.
 ---

 # nREPL Eval

 Evaluate Clojure (or ClojureScript) code via a running nREPL server using
-`tools/nrepl-eval.mjs` — a standalone CLI application.
+`scripts/nrepl-eval.mjs`.

-Session state (defs, in-ns, etc.) persists across invocations via a stored
-session ID, so you can build up state incrementally.
+Full documentation: `mem:scripts/nrepl-eval` (file: `.serena/memories/scripts/nrepl-eval.md`)

-## Usage
+## Quick Reference

 ```bash
-node tools/nrepl-eval.mjs [options] [<code>]
+./scripts/nrepl-eval.mjs [options] [<code>]
 ```

-The tool is also executable directly:
-```bash
-./tools/nrepl-eval.mjs [options] [<code>]
-```
-
-## Options
-
 | Flag | Description | Default |
 |------|-------------|---------|
+| `--backend` | Connect to backend nREPL (port 6064) | — |
+| `--frontend` | Connect to frontend nREPL (port 3447) | — |
 | `-p, --port PORT` | nREPL server port | `6064` |
 | `-H, --host HOST` | nREPL server host | `127.0.0.1` |
 | `-t, --timeout MS` | Timeout in milliseconds | `120000` |
@@ -33,88 +27,11 @@ The tool is also executable directly:
 | `-e, --last-error` | Evaluate `*e` to retrieve the last exception | — |
 | `-h, --help` | Show help message | — |

-## When to Use
-
-Use this tool when you need to:
-
-1. **Evaluate Clojure code** during development — test functions, inspect
-   state, or run experiments against a running Clojure process.
-2. **Verify that edited files compile** — require namespaces with `:reload`
-   to pick up changes.
-3. **Inspect the last exception** after a failed evaluation — use `-e` to
-   print the error stored in `*e`.
-
-## Workflow
-
-### 1. Session management
-
-Sessions are persisted to `/tmp/penpot-nrepl-session-<host>-<port>`. State
-carries across calls automatically:
+## Examples

 ```bash
-./tools/nrepl-eval.mjs '(def x 42)'
-./tools/nrepl-eval.mjs 'x'
-# => 42
+./scripts/nrepl-eval.mjs '(+ 1 2 3)'
+./scripts/nrepl-eval.mjs --backend '(+ 1 2 3)'
+./scripts/nrepl-eval.mjs --frontend '(js/alert "hi")'
+./scripts/nrepl-eval.mjs -e
 ```
-
-Reset the session to start fresh:
-
-```bash
-./tools/nrepl-eval.mjs --reset-session '(def x 0)'
-```
-
-### 2. Evaluate code
-
-**Single expression (inline) — uses default port 6064:**
-```bash
-./tools/nrepl-eval.mjs '(+ 1 2 3)'
-```
-
-**Multiple expressions via heredoc (recommended — avoids escaping issues):**
-```bash
-./tools/nrepl-eval.mjs <<'EOF'
-(def x 10)
-(+ x 20)
-EOF
-```
-
-**Override with a different port:**
-```bash
-./tools/nrepl-eval.mjs -p 7888 '(+ 1 2 3)'
-```
-
-### 3. Inspect last exception
-
-After code throws an error, retrieve the full exception details:
-
-```bash
-./tools/nrepl-eval.mjs -e
-```
-
-## Common Patterns
-
-**Require a namespace with reload:**
-```bash
-./tools/nrepl-eval.mjs "(require '[my.namespace :as ns] :reload)"
-```
-
-**Test a function:**
-```bash
-./tools/nrepl-eval.mjs "(ns/my-function arg1 arg2)"
-```
-
-**Long-running operation with custom timeout:**
-```bash
-./tools/nrepl-eval.mjs -t 300000 "(long-running-fn)"
-```
-
-## Key Principles
-
- **Default port is 6064** — just pass code directly, no `-p` needed when
-  your nREPL server is on 6064. Use `-p <PORT>` for a different port.
- **Always use `:reload`** when requiring namespaces to pick up file changes.
- **Session is reused** across invocations — defs, in-ns, and var bindings
-  persist. Use `--reset-session` to clear.
- **Do not start any server** — the tool connects to an existing nREPL
-  server, it is not the agent's responsibility to start the nREPL server
-  (assume the server is already running on the specified port).
@@ -0,0 +1,271 @@
+---
+name: planner
+description: Read-only planning and architecture analysis for Penpot — produce a structured implementation plan (Context, Affected modules, Approach, Risks, Testing). Always output to the user; additionally save to .opencode/plans/YYYY-MM-DD-<title>.md.
+---
+
+# Planner
+
+Read-only senior software architect role for Penpot. Produces structured
+implementation plans that engineers or other agents can execute. Never writes
+or modifies code.
+
+## When to Use
+
+- The user asks for a plan, design, or analysis of a feature or bug.
+- The user wants to understand which parts of the codebase a task will touch.
+- The user needs a step-by-step implementation plan with file paths, function
+  names, and test strategy.
+- The user asks "how would I implement X?" or "what's involved in fixing Y?".
+- The user is about to start non-trivial work and wants a bite-sized task
+  breakdown.
+
+Do **not** use this skill to actually implement anything — it is read-only.
+
+## Role
+
+You are a Senior Software Architect working on Penpot, an open-source design
+tool. Your sole responsibility is planning and analysis — you do NOT write or
+modify code.
+
+You help users understand the codebase, design solutions, and create detailed
+implementation plans that other agents or developers can execute. Document
+everything they need to know: which files to touch for each task, code patterns,
+tests, and how to verify correctness. Apply DRY and KISS principles.
+
+Do **not** suggest commit messages or commit names anywhere in your plans or
+responses — committing is the developer's responsibility.
+
+## Required Reading Before Planning
+
+Before drafting any plan, work through the project's own guidance:
+
+1. Read `critical-info` (`.serena/memories/critical-info.md`) — the entry point
+   that describes the monorepo structure and module dependency graph.
+2. From `critical-info`, identify which modules your task affects.
+3. Read each affected module's core memory, e.g. `mem:frontend/core`,
+   `mem:backend/core`, `mem:common/core`, `mem:exporter/core`,
+   `mem:render-wasm/core`. Follow `mem:` references deeper as needed.
+4. For each affected module, note its lint, format, and test commands so the
+   plan can include concrete verification steps.
+
+Skipping this step is the #1 cause of incorrect or incomplete plans.
+
+## The Planning Process
+
+### Phase 1: Architecture Analysis
+
+1. Read the spec, requirements, or feature request.
+2. Analyze the codebase architecture and identify affected modules.
+3. Read project conventions (starting with `critical-info` and module core
+   memories) before drafting.
+4. Map dependencies between components (see the dependency graph in
+   `critical-info`).
+5. Identify risks, edge cases, performance implications, and breaking changes.
+
+### Phase 2: Task Breakdown
+
+Implementation order follows the monorepo's dependency graph:
+`frontend -> common`, `backend -> common`, `exporter -> common`,
+`frontend -> render-wasm`. Build shared foundations first, then layer
+consumers on top.
+
+#### Slice Vertically
+
+Instead of building all of common, then all of backend, then all of frontend —
+build one complete feature path at a time:
+
+```
+Task 1: common data types + schema             ← foundation
+Task 2: backend RPC handler + persistence
+Task 3: frontend UI component + API integration
+```
+
+Each vertical slice delivers working, testable functionality.
+
+#### Write Tasks
+
+Each task follows this structure:
+
+```markdown
+## Task [N]: [Short descriptive title]
+
+**Description:** One paragraph explaining what this task accomplishes.
+
+**Acceptance criteria:**
+- [ ] [Specific, testable condition]
+- [ ] [Specific, testable condition]
+
+**Verification:**
+- [ ] Tests pass (module-specific test command)
+- [ ] Lint/formatter passes (module-specific check command)
+
+**Dependencies:** [Task numbers this depends on, or "None"]
+
+**Files likely touched:**
+- `path/to/file.clj`
+- `path/to/file_test.clj`
+```
+
+Replace "module-specific test command" with the actual commands for the module
+(e.g. `clojure -M:dev:test` for backend/common, `npx shadow-cljs compile test && npx karma start` for frontend,
+or the commands noted in the module's core memory).
+
+#### Estimate Scope
+
+| Size | Files | Scope |
+|------|-------|-------|
+| **XS** | 1 | Single function, config change, or schema tweak |
+| **S** | 1-2 | One handler or component method |
+| **M** | 3-5 | One vertical feature slice |
+| **L** | 5-8 | Multi-component feature |
+| **XL** | 8+ | **Too large — break it down further** |
+
+If a task is L or larger, break it into smaller tasks. Agents perform best on
+S and M tasks.
+
+**When to break a task down further:**
+- It would take more than one focused session
+- You cannot describe the acceptance criteria in 3 or fewer bullet points
+- It touches two or more independent subsystems
+- You find yourself writing "and" in the task title (a sign it is two tasks)
+
+#### Order and Checkpoints
+
+Arrange tasks so that:
+
+1. Dependencies are satisfied (build foundation first)
+2. Each task leaves the system in a working state
+3. Verification checkpoints occur after every 2-3 tasks
+4. High-risk tasks are early (fail fast)
+
+Add explicit checkpoints with the relevant module commands:
+
+```markdown
+## Checkpoint: After Tasks 1-3
+- [ ] All tests pass (module-specific command)
+- [ ] Lint/format passes (module-specific command)
+- [ ] Core flow works end-to-end
+- [ ] Review with human before proceeding
+```
+
+## Requirements
+
+- Analyze the codebase architecture and identify affected modules.
+- Read project conventions before drafting (start with `critical-info` and
+  affected module core memories).
+- Break down complex features or bugs into atomic, actionable steps.
+- Propose solutions with clear rationale, trade-offs, and sequencing.
+- Identify risks, edge cases, performance implications, and breaking changes.
+- Apply DRY and KISS principles to the proposed implementation.
+- Define a testing strategy aligned with each affected module's tooling.
+- Every task must have acceptance criteria and verification steps.
+- Checkpoints must exist between major phases.
+
+## Constraints
+
+- You are **analysis-only** — never create, edit, or delete source code.
+- The only file write you may attempt is the plan itself, saved to
+  `.opencode/plans/`.
+- You do **not** run builds, tests, linters, or any commands that modify state.
+- You do **not** create git commits or interact with version control.
+- You do **not** execute shell commands beyond read-only searches.
+- Your output is a structured plan or analysis, ready for handoff to an
+  engineer agent or developer.
+
+## Output Format
+
+The plan is always delivered in the response so the user sees it regardless
+of which agent is running the skill.
+
+Additionally, save the plan to:
+
+```
+.opencode/plans/YYYY-MM-DD-<plan-one-line-title>.md
+```
+
+Use today's date in the user's local timezone. The `<plan-one-line-title>`
+slug is lowercase, hyphen-separated, and a short summary of the task
+(e.g. `add-batch-get-profiles-for-file-comments`). Create the
+`.opencode/plans/` directory if it does not exist.
+
+Always attempt the write. If the user explicitly provides a target file path,
+use that path instead of the default.
+
+### Plan Document Template
+
+```markdown
+# Plan: [Feature/Project Name]
+
+## Context
+[One paragraph: what is the problem or feature request? Why is it needed?]
+
+## Affected Modules
+[Which modules of the monorepo are involved? Reference module paths and any
+`mem:` memories that were consulted.]
+
+## Architecture Decisions
+- [Key decision 1 and rationale]
+- [Key decision 2 and rationale]
+
+## Risks & Considerations
+[Edge cases, performance implications, breaking changes, migration concerns,
+security implications.]
+
+## Approach
+[Step-by-step implementation plan with file paths, function names, and code
+shape where applicable. Group steps into atomic, ordered tasks.]
+
+## Task List
+
+### Phase 1: Foundation
+- [ ] Task 1: ...
+- [ ] Task 2: ...
+
+### Checkpoint: Phase 1
+- [ ] Tests pass, lint/formatter clean (module-specific commands)
+
+### Phase 2: Core Features
+- [ ] Task 3: ...
+- [ ] Task 4: ...
+
+### Checkpoint: Phase 2
+- [ ] End-to-end flow works
+
+### Phase 3: Polish
+- [ ] Task 5: ...
+- [ ] Task 6: ...
+
+### Checkpoint: Complete
+- [ ] All acceptance criteria met
+- [ ] Ready for review
+
+## Testing Strategy
+[How to verify: which test commands to run per module, what cases to cover,
+manual verification steps, lint/format checks. Consult each module's core
+memory for the exact commands.]
+
+## Parallelization Opportunities
+- **Safe to parallelize:** Independent feature slices across separate
+  modules, tests for already-implemented features
+- **Must be sequential:** Shared common schema changes, database migrations
+- **Needs coordination:** Features that share a contract (define the contract
+  first, then parallelize)
+
+## Open Questions
+- [Question needing human input]
+```
+
+When the plan is purely analytical (e.g. a code review or feasibility study
+with no implementation), skip the **Approach** and **Task List** sections and
+lead with **Findings** instead, keeping the rest of the structure.
+
+## Verification Checklist
+
+Before starting implementation, confirm:
+
+- [ ] Every task has acceptance criteria
+- [ ] Every task has a verification step
+- [ ] Task dependencies are identified and ordered correctly
+- [ ] No task touches more than ~5 files
+- [ ] Checkpoints exist between major phases
+- [ ] The human has reviewed and approved the plan
@@ -0,0 +1,134 @@
+---
+name: refine-prompt
+description: Refine and improve a user-supplied prompt for maximum clarity and effectiveness using prompt-engineering best practices and Penpot project context. Outputs a rewritten prompt (and brief rationale); never executes the prompt.
+---
+
+# Refine Prompt
+
+Expert prompt-engineering pass on a user-supplied prompt. Takes a draft prompt
+and returns a clearer, more effective, well-structured version — ready to be
+used with any AI model. Never executes the prompt itself.
+
+## When to Use
+
+- The user shares a prompt and asks to improve, refine, polish, or rewrite it.
+- The user asks "make this prompt better" or "can you clean this up?".
+- The user wants to add structure, constraints, examples, or output format to
+  a vague prompt.
+- The user wants a prompt adapted for a specific target model, audience, or
+  task type.
+
+Do **not** use this skill to actually answer the prompt or do the task — it
+only rewrites the prompt.
+
+## Role
+
+You are an expert Prompt Engineer with strong knowledge of Penpot. Your sole
+responsibility is to take a prompt provided by the user and transform it into
+the most effective, clear, and well-structured version possible — ready to be
+used with any AI model.
+
+You do **not** execute tasks. You do **not** write code. You only design and
+refine prompts.
+
+## Required Reading Before Refining
+
+Before rewriting, internalize the project context the prompt will likely run
+against:
+
+1. Read `AGENTS.md` (root) for the project-level rules and conventions.
+2. Read `.serena/memories/critical-info.md` (or the equivalent entry point) to
+   understand the module layout (`frontend`, `backend`, `common`,
+   `render-wasm`, `exporter`, `mcp`, `plugins`, `library`).
+3. Skim the relevant module's core memory (`mem:frontend/core`,
+   `mem:backend/core`, etc.) when the prompt targets a specific module — this
+   lets you inject precise vocabulary, file conventions, and test commands
+   into the refined prompt.
+
+This step matters most when the user is preparing a prompt *about* the
+Penpot codebase. For generic prompts, focus on prompt-engineering principles
+and only weave in Penpot context when it is clearly relevant.
+
+## Requirements
+
+- Analyze the original prompt: identify its intent, target audience,
+  ambiguities, missing context, and structural weaknesses.
+- Ask clarifying questions if the intent is unclear or if critical information
+  is missing (e.g. target model, expected output format, tone, constraints).
+  Keep questions concise and grouped. Prefer to ask 1–4 questions at once
+  rather than one at a time. **Use the `question` tool** to ask them so the
+  user gets a structured multi-choice UI; reserve a plain `## Clarifying
+  questions` markdown section for cases where the `question` tool is
+  unavailable or the question is genuinely open-ended.
+- Rewrite the prompt using prompt-engineering best practices (see below).
+- Preserve the user's original intent — do not change the underlying task.
+- When the user provides Penpot project context, weave in the relevant
+  conventions, module paths, and tooling.
+
+## Prompt Engineering Principles
+
+Apply these techniques when refining prompts:
+
+- **Be specific and explicit**: Replace vague instructions with precise ones.
+- **Set the context**: Include background information the model needs to
+  perform well.
+- **Specify the output format**: State the desired structure, length, tone,
+  or format (e.g. bullet list, JSON, step-by-step).
+- **Add constraints**: Include what the model should avoid or not do.
+- **Use examples** (few-shot): When applicable, suggest adding examples to
+  anchor the model's behaviour.
+- **Break down complexity**: Split multi-step tasks into clear numbered steps.
+- **Avoid ambiguity**: Remove pronouns and references that could be
+  misinterpreted.
+- **Chain of thought**: For reasoning tasks, include "Think step by step."
+- **Role framing**: When helpful, give the model a clear role
+  ("You are a senior backend engineer...").
+- **Tool awareness**: When the prompt targets an agentic model, mention
+  relevant tools (`grep`, `glob`, `read`, `bash`, etc.) so the model uses the
+  right surface.
+
+## Constraints
+
+- Do **not** execute the prompt yourself.
+- Do **not** answer the question inside the prompt.
+- Do **not** add unnecessary verbosity — prompts should be as short as they
+  can be while remaining complete.
+- Always preserve the user's original intent.
+- If the user provides Penpot project context, prefer Penpot-specific
+  vocabulary over generic terms (e.g. name actual modules and `mem:`
+  references instead of "the codebase").
+
+## Output Format
+
+Deliver the result in the response as two clearly separated blocks:
+
+1. **Refined prompt** — a single fenced code block (markdown ```) containing
+   the rewritten prompt, ready to copy and use.
+2. **What changed (brief)** — a short bulleted list of the most important
+   changes you made and why (3–7 bullets max). Skip the rationale if the
+   changes are trivial.
+
+If you asked clarifying questions via the `question` tool, stop and wait for
+the answers before producing a refined prompt. If the `question` tool was not
+available and you asked the questions in chat, list them in a separate
+**Clarifying questions** section above the refined prompt and stop — do not
+produce a refined prompt until the user answers. If the user explicitly told
+you to proceed without questions (e.g. "just rewrite it"), make reasonable
+assumptions and note them under **Assumptions made** in the rationale block.
+
+## File Persistence
+
+Always persist the refined prompt to disk so it can be re-used later, versioned
+in git, and shared with other agents. The response still contains the prompt
+and rationale blocks; the file is an additional artifact, not a replacement.
+
+- Save the refined prompt (the body inside the fenced code block, **without**
+  the surrounding ``` fences) to `.opencode/prompts/<descriptive-name>.md`.
+- Use a **kebab-case** filename that summarises the task, e.g.
+  `add-error-reports-management-rpc.md`, `backend-rpc-security-audit.md`. No
+  spaces, no uppercase, no version numbers or dates in the filename.
+- If `.opencode/prompts/` does not exist, create it before writing.
+- If a file with the same name already exists, overwrite it (the file is the
+  refined prompt, not a log).
+- Only skip the file write when the user explicitly opts out (e.g. "don't save
+  this one", "just show it in the chat"). When in doubt, save it.
@@ -0,0 +1,457 @@
+---
+name: security-and-hardening
+description: Hardens code against vulnerabilities. Use when handling user input, authentication, data storage, or external integrations. Use when building any feature that accepts untrusted data, manages user sessions, or interacts with third-party services.
+---
+
+# Security and Hardening
+
+## Overview
+
+Security-first development practices for web applications. Treat every external input as hostile, every secret as sacred, and every authorization check as mandatory. Security isn't a phase — it's a constraint on every line of code that touches user data, authentication, or external systems.
+
+## When to Use
+
+- Building anything that accepts user input
+- Implementing authentication or authorization
+- Storing or transmitting sensitive data
+- Integrating with external APIs or services
+- Adding file uploads, webhooks, or callbacks
+- Handling payment or PII data
+
+## Process: Threat Model First
+
+Controls bolted on without a threat model are guesses. Before hardening, spend five minutes thinking like an attacker:
+
+1. **Map the trust boundaries.** Where does untrusted data cross into your system? HTTP requests, form fields, file uploads, webhooks, third-party APIs, message queues, and **LLM output**. Every boundary is attack surface.
+2. **Name the assets.** What's worth stealing or breaking? Credentials, PII, payment data, admin actions, money movement.
+3. **Run STRIDE over each boundary** — a quick lens, not a ceremony:
+
+| Threat | Ask | Typical mitigation |
+|---|---|---|
+| **S**poofing | Can someone impersonate a user/service? | Authentication, signature verification |
+| **T**ampering | Can data be altered in transit or at rest? | Integrity checks, parameterized queries, HTTPS |
+| **R**epudiation | Can an action be denied later? | Audit logging of security events |
+| **I**nformation disclosure | Can data leak? | Encryption, field allowlists, generic errors |
+| **D**enial of service | Can it be overwhelmed? | Rate limiting, input size caps, timeouts |
+| **E**levation of privilege | Can a user gain rights they shouldn't? | Authorization checks, least privilege |
+
+4. **Write abuse cases next to use cases.** For each feature, ask "how would I misuse this?" — then make that your first test.
+
+If you can't name the trust boundaries for a feature, you're not ready to secure it. This is OWASP **A04: Insecure Design** — most breaches begin in design, not code.
+
+## The Three-Tier Boundary System
+
+### Always Do (No Exceptions)
+
+- **Validate all external input** at the system boundary (API routes, form handlers)
+- **Parameterize all database queries** — never concatenate user input into SQL
+- **Encode output** to prevent XSS (use framework auto-escaping, don't bypass it)
+- **Use HTTPS** for all external communication
+- **Hash passwords** with bcrypt/scrypt/argon2 (never store plaintext)
+- **Set security headers** (CSP, HSTS, X-Frame-Options, X-Content-Type-Options)
+- **Use httpOnly, secure, sameSite cookies** for sessions
+- **Run `npm audit`** (or equivalent) before every release
+
+### Ask First (Requires Human Approval)
+
+- Adding new authentication flows or changing auth logic
+- Storing new categories of sensitive data (PII, payment info)
+- Adding new external service integrations
+- Changing CORS configuration
+- Adding file upload handlers
+- Modifying rate limiting or throttling
+- Granting elevated permissions or roles
+
+### Never Do
+
+- **Never commit secrets** to version control (API keys, passwords, tokens)
+- **Never log sensitive data** (passwords, tokens, full credit card numbers)
+- **Never trust client-side validation** as a security boundary
+- **Never disable security headers** for convenience
+- **Never use `eval()` or `innerHTML`** with user-provided data
+- **Never store sessions in client-accessible storage** (localStorage for auth tokens)
+- **Never expose stack traces** or internal error details to users
+
+## OWASP Top 10 Prevention Patterns
+
+These are prevention patterns, not a ranking. For the 2021 ordering, see the quick-reference table in `references/security-checklist.md`.
+
+### Injection (SQL, NoSQL, OS Command)
+
+```typescript
+// BAD: SQL injection via string concatenation
+const query = `SELECT * FROM users WHERE id = '${userId}'`;
+
+// GOOD: Parameterized query
+const user = await db.query('SELECT * FROM users WHERE id = $1', [userId]);
+
+// GOOD: ORM with parameterized input
+const user = await prisma.user.findUnique({ where: { id: userId } });
+```
+
+### Broken Authentication
+
+```typescript
+// Password hashing
+import { hash, compare } from 'bcrypt';
+
+const SALT_ROUNDS = 12;
+const hashedPassword = await hash(plaintext, SALT_ROUNDS);
+const isValid = await compare(plaintext, hashedPassword);
+
+// Session management
+app.use(session({
+  secret: process.env.SESSION_SECRET,  // From environment, not code
+  resave: false,
+  saveUninitialized: false,
+  cookie: {
+    httpOnly: true,     // Not accessible via JavaScript
+    secure: true,       // HTTPS only
+    sameSite: 'lax',    // CSRF protection
+    maxAge: 24 * 60 * 60 * 1000,  // 24 hours
+  },
+}));
+```
+
+### Cross-Site Scripting (XSS)
+
+```typescript
+// BAD: Rendering user input as HTML
+element.innerHTML = userInput;
+
+// GOOD: Use framework auto-escaping (React does this by default)
+return <div>{userInput}</div>;
+
+// If you MUST render HTML, sanitize first
+import DOMPurify from 'dompurify';
+const clean = DOMPurify.sanitize(userInput);
+```
+
+### Broken Access Control
+
+```typescript
+// Always check authorization, not just authentication
+app.patch('/api/tasks/:id', authenticate, async (req, res) => {
+  const task = await taskService.findById(req.params.id);
+
+  // Check that the authenticated user owns this resource
+  if (task.ownerId !== req.user.id) {
+    return res.status(403).json({
+      error: { code: 'FORBIDDEN', message: 'Not authorized to modify this task' }
+    });
+  }
+
+  // Proceed with update
+  const updated = await taskService.update(req.params.id, req.body);
+  return res.json(updated);
+});
+```
+
+### Security Misconfiguration
+
+```typescript
+// Security headers (use helmet for Express)
+import helmet from 'helmet';
+app.use(helmet());
+
+// Content Security Policy
+app.use(helmet.contentSecurityPolicy({
+  directives: {
+    defaultSrc: ["'self'"],
+    scriptSrc: ["'self'"],
+    styleSrc: ["'self'", "'unsafe-inline'"],  // Tighten if possible
+    imgSrc: ["'self'", 'data:', 'https:'],
+    connectSrc: ["'self'"],
+  },
+}));
+
+// CORS — restrict to known origins
+app.use(cors({
+  origin: process.env.ALLOWED_ORIGINS?.split(',') || 'http://localhost:3000',
+  credentials: true,
+}));
+```
+
+### Sensitive Data Exposure
+
+```typescript
+// Never return sensitive fields in API responses
+function sanitizeUser(user: UserRecord): PublicUser {
+  const { passwordHash, resetToken, ...publicFields } = user;
+  return publicFields;
+}
+
+// Use environment variables for secrets
+const API_KEY = process.env.STRIPE_API_KEY;
+if (!API_KEY) throw new Error('STRIPE_API_KEY not configured');
+```
+
+### Server-Side Request Forgery (SSRF)
+
+Any time the server fetches a URL the user influenced — webhooks, "import from URL", image proxies, link previews — an attacker can aim it at internal services (cloud metadata, `localhost`, private IPs).
+
+```typescript
+// BAD: fetch whatever the user gives you
+await fetch(req.body.webhookUrl);
+
+// GOOD: allowlist scheme + host, reject if ANY resolved IP is private, forbid redirects
+import { lookup } from 'node:dns/promises';
+import ipaddr from 'ipaddr.js';
+
+const ALLOWED_HOSTS = new Set(['hooks.example.com']);
+
+async function assertSafeUrl(raw: string): Promise<URL> {
+  const url = new URL(raw);
+  if (url.protocol !== 'https:') throw new Error('https only');
+  if (!ALLOWED_HOSTS.has(url.hostname)) throw new Error('host not allowed');
+  // Resolve ALL records; a single private/reserved address fails the check.
+  const addrs = await lookup(url.hostname, { all: true });
+  if (addrs.some((a) => ipaddr.parse(a.address).range() !== 'unicast')) {
+    throw new Error('private/reserved IP');
+  }
+  return url;
+}
+
+await fetch(await assertSafeUrl(req.body.webhookUrl), { redirect: 'error' });
+```
+
+The `range() !== 'unicast'` check covers loopback, link-local `169.254.169.254` (cloud metadata, the #1 SSRF target), private, and unique-local ranges across IPv4 and IPv6.
+
+**Caveat — this still has a TOCTOU gap.** `fetch` resolves DNS again after the check, so an attacker using a short-TTL record can rebind to an internal IP between validation and connection. For high-risk surfaces, resolve once and connect to the pinned IP, or put a filtering agent in front (`request-filtering-agent` / `ssrf-req-filter`).
+
+## Input Validation Patterns
+
+### Schema Validation at Boundaries
+
+```typescript
+import { z } from 'zod';
+
+const CreateTaskSchema = z.object({
+  title: z.string().min(1).max(200).trim(),
+  description: z.string().max(2000).optional(),
+  priority: z.enum(['low', 'medium', 'high']).default('medium'),
+  dueDate: z.string().datetime().optional(),
+});
+
+// Validate at the route handler
+app.post('/api/tasks', async (req, res) => {
+  const result = CreateTaskSchema.safeParse(req.body);
+  if (!result.success) {
+    return res.status(422).json({
+      error: {
+        code: 'VALIDATION_ERROR',
+        message: 'Invalid input',
+        details: result.error.flatten(),
+      },
+    });
+  }
+  // result.data is now typed and validated
+  const task = await taskService.create(result.data);
+  return res.status(201).json(task);
+});
+```
+
+### File Upload Safety
+
+```typescript
+// Restrict file types and sizes
+const ALLOWED_TYPES = ['image/jpeg', 'image/png', 'image/webp'];
+const MAX_SIZE = 5 * 1024 * 1024; // 5MB
+
+function validateUpload(file: UploadedFile) {
+  if (!ALLOWED_TYPES.includes(file.mimetype)) {
+    throw new ValidationError('File type not allowed');
+  }
+  if (file.size > MAX_SIZE) {
+    throw new ValidationError('File too large (max 5MB)');
+  }
+  // Don't trust the file extension — check magic bytes if critical
+}
+```
+
+## Triaging npm audit Results
+
+Not all audit findings require immediate action. Use this decision tree:
+
+```
+npm audit reports a vulnerability
+├── Severity: critical or high
+│   ├── Is the vulnerable code reachable in your app?
+│   │   ├── YES --> Fix immediately (update, patch, or replace the dependency)
+│   │   └── NO (dev-only dep, unused code path) --> Fix soon, but not a blocker
+│   └── Is a fix available?
+│       ├── YES --> Update to the patched version
+│       └── NO --> Check for workarounds, consider replacing the dependency, or add to allowlist with a review date
+├── Severity: moderate
+│   ├── Reachable in production? --> Fix in the next release cycle
+│   └── Dev-only? --> Fix when convenient, track in backlog
+└── Severity: low
+    └── Track and fix during regular dependency updates
+```
+
+**Key questions:**
+- Is the vulnerable function actually called in your code path?
+- Is the dependency a runtime dependency or dev-only?
+- Is the vulnerability exploitable given your deployment context (e.g., a server-side vulnerability in a client-only app)?
+
+When you defer a fix, document the reason and set a review date.
+
+### Supply-Chain Hygiene
+
+`npm audit` catches known CVEs; it won't catch a malicious or typosquatted package. Also:
+
+- **Commit the lockfile** and install with `npm ci` (not `npm install`) in CI — reproducible builds, no silent version drift.
+- **Review new dependencies before adding them** — maintenance, download counts, and whether they truly earn their place. Every dependency is attack surface (OWASP **A06: Vulnerable Components**, **LLM03: Supply Chain**).
+- **Be wary of `postinstall` scripts** in unfamiliar packages — they run arbitrary code at install time.
+- **Watch for typosquats** — `cross-env` vs `crossenv`, `react-dom` vs `reactdom`.
+
+## Rate Limiting
+
+```typescript
+import rateLimit from 'express-rate-limit';
+
+// General API rate limit
+app.use('/api/', rateLimit({
+  windowMs: 15 * 60 * 1000, // 15 minutes
+  max: 100,                   // 100 requests per window
+  standardHeaders: true,
+  legacyHeaders: false,
+}));
+
+// Stricter limit for auth endpoints
+app.use('/api/auth/', rateLimit({
+  windowMs: 15 * 60 * 1000,
+  max: 10,  // 10 attempts per 15 minutes
+}));
+```
+
+## Secrets Management
+
+```
+.env files:
+  ├── .env.example  → Committed (template with placeholder values)
+  ├── .env          → NOT committed (contains real secrets)
+  └── .env.local    → NOT committed (local overrides)
+
+.gitignore must include:
+  .env
+  .env.local
+  .env.*.local
+  *.pem
+  *.key
+```
+
+**Always check before committing:**
+```bash
+# Check for accidentally staged secrets
+git diff --cached | grep -i "password\|secret\|api_key\|token"
+```
+
+**If a secret is ever committed, rotate it.** Deleting the line or rewriting history is not enough — assume it's compromised the moment it reaches a remote. Revoke and reissue the key first, then purge it from history.
+
+## Securing AI / LLM Features
+
+If your app calls an LLM — chatbots, summarizers, agents, RAG — it inherits a new attack surface. Map it to the [OWASP Top 10 for LLM Applications (2025)](https://genai.owasp.org/llm-top-10/):
+
+- **Treat all model output as untrusted input (LLM05: Improper Output Handling).** Never pass LLM output straight into `eval`, SQL, a shell, `innerHTML`, or a file path. Validate and encode it exactly as you would raw user input.
+- **Assume prompts can be hijacked (LLM01: Prompt Injection).** Untrusted text in the context window — a user message, a fetched web page, a PDF — can carry instructions. The system prompt is not a security boundary; enforce permissions in code, not in the prompt.
+- **Keep secrets and other users' data out of prompts (LLM02 / LLM07).** Anything in the context can be echoed back. Don't put API keys, cross-tenant data, or the full system prompt where the model can repeat it.
+- **Constrain tool and agent permissions (LLM06: Excessive Agency).** Scope tools to the minimum, require confirmation for destructive or irreversible actions, and validate every tool argument.
+- **Bound consumption (LLM10: Unbounded Consumption).** Cap tokens, request rate, and loop/recursion depth so a crafted input can't run up cost or hang the system.
+- **Isolate retrieval data (LLM08: Vector and Embedding Weaknesses).** In RAG, treat the vector store as a trust boundary: partition embeddings per tenant so one user can't retrieve another's data, and validate documents before indexing so poisoned content can't steer answers.
+
+```typescript
+// BAD: trusting model output as a command or as markup
+const sql = await llm.generate(`Write SQL for: ${userQuestion}`);
+await db.query(sql);                                   // arbitrary query execution
+container.innerHTML = await llm.reply(userMessage);   // stored XSS, via the model
+
+// GOOD: model output is data — parse defensively, then validate, then encode
+let intent;
+try {
+  intent = CommandSchema.parse(JSON.parse(await llm.replyJson(userMessage)));
+} catch {
+  throw new ValidationError('unexpected model output'); // JSON.parse or schema failed
+}
+await runAllowlistedAction(intent.action, intent.params);
+container.textContent = await llm.reply(userMessage);
+```
+
+## Security Review Checklist
+
+```markdown
+### Authentication
+- [ ] Passwords hashed with bcrypt/scrypt/argon2 (salt rounds ≥ 12)
+- [ ] Session tokens are httpOnly, secure, sameSite
+- [ ] Login has rate limiting
+- [ ] Password reset tokens expire
+
+### Authorization
+- [ ] Every endpoint checks user permissions
+- [ ] Users can only access their own resources
+- [ ] Admin actions require admin role verification
+
+### Input
+- [ ] All user input validated at the boundary
+- [ ] SQL queries are parameterized
+- [ ] HTML output is encoded/escaped
+- [ ] Server-side URL fetches are allowlisted (no SSRF to internal services)
+
+### Data
+- [ ] No secrets in code or version control
+- [ ] Sensitive fields excluded from API responses
+- [ ] PII encrypted at rest (if applicable)
+
+### Infrastructure
+- [ ] Security headers configured (CSP, HSTS, etc.)
+- [ ] CORS restricted to known origins
+- [ ] Dependencies audited for vulnerabilities
+- [ ] Error messages don't expose internals
+
+### Supply Chain
+- [ ] Lockfile committed; CI installs with `npm ci`
+- [ ] New dependencies reviewed (maintenance, downloads, postinstall scripts)
+
+### AI / LLM (if used)
+- [ ] Model output treated as untrusted (no eval/SQL/innerHTML/shell)
+- [ ] Secrets and other users' data kept out of prompts
+- [ ] Tool/agent permissions scoped; destructive actions require confirmation
+```
+## Common Rationalizations
+
+| Rationalization | Reality |
+|---|---|
+| "This is an internal tool, security doesn't matter" | Internal tools get compromised. Attackers target the weakest link. |
+| "We'll add security later" | Security retrofitting is 10x harder than building it in. Add it now. |
+| "No one would try to exploit this" | Automated scanners will find it. Security by obscurity is not security. |
+| "The framework handles security" | Frameworks provide tools, not guarantees. You still need to use them correctly. |
+| "It's just a prototype" | Prototypes become production. Security habits from day one. |
+| "Threat modeling is overkill here" | Five minutes of "how would I attack this?" prevents the design flaws no control can patch later. |
+| "It's just LLM output, it's only text" | That "text" can be a SQL statement, a script tag, or a shell command. Treat it like any untrusted input. |
+
+## Red Flags
+
+- User input passed directly to database queries, shell commands, or HTML rendering
+- Secrets in source code or commit history
+- API endpoints without authentication or authorization checks
+- Missing CORS configuration or wildcard (`*`) origins
+- No rate limiting on authentication endpoints
+- Stack traces or internal errors exposed to users
+- Dependencies with known critical vulnerabilities
+- Server fetches user-supplied URLs without an allowlist (SSRF)
+- LLM/model output passed into a query, the DOM, a shell, or `eval`
+- Secrets, PII, or the full system prompt placed inside an LLM context window
+
+## Verification
+
+After implementing security-relevant code:
+
+- [ ] `npm audit` shows no critical or high vulnerabilities
+- [ ] No secrets in source code or git history
+- [ ] All user input validated at system boundaries
+- [ ] Authentication and authorization checked on every protected endpoint
+- [ ] Security headers present in response (check with browser DevTools)
+- [ ] Error responses don't expose internal details
+- [ ] Rate limiting active on auth endpoints
+- [ ] Server-side URL fetches validated against an allowlist (no SSRF)
+- [ ] LLM/model output validated and encoded before use (if AI features present)
@@ -13,7 +13,7 @@ Fetch information from Taiga public API for the **Penpot** project

 ## Prerequisites

- `python3` — the `tools/taiga.py` CLI script is self-contained (stdlib only)
+- `python3` — the `scripts/taiga.py` CLI script is self-contained (stdlib only)

 ## Quick Start

@@ -21,17 +21,17 @@ The easiest way is to use the bundled Python script:

 ```bash
 # Pass a Taiga URL directly
-python3 tools/taiga.py https://tree.taiga.io/project/penpot/issue/13714
+python3 scripts/taiga.py https://tree.taiga.io/project/penpot/issue/13714

 # Or use "<type> <ref>" syntax
-python3 tools/taiga.py us 14128
-python3 tools/taiga.py task 13648
+python3 scripts/taiga.py us 14128
+python3 scripts/taiga.py task 13648

 # Add --json for raw output
-python3 tools/taiga.py --json issue 13714
+python3 scripts/taiga.py --json issue 13714

 # See full usage
-python3 tools/taiga.py --help
+python3 scripts/taiga.py --help
 ```

 ## URL Pattern Reference
@@ -51,30 +51,30 @@ To extract the **type** and **ref** from a URL:

 ## Python Script Reference

-The `tools/taiga.py` script wraps the Taiga API into a single convenient CLI
+The `scripts/taiga.py` script wraps the Taiga API into a single convenient CLI
 with sensible defaults.

 ### Usage

 ```
-python3 tools/taiga.py <taiga-url>
-python3 tools/taiga.py <type> <ref>
-python3 tools/taiga.py [--json] <taiga-url>
-python3 tools/taiga.py [--json] <type> <ref>
+python3 scripts/taiga.py <taiga-url>
+python3 scripts/taiga.py <type> <ref>
+python3 scripts/taiga.py [--json] <taiga-url>
+python3 scripts/taiga.py [--json] <type> <ref>
 ```

 ### Examples

 ```bash
 # By URL (recommended — no need to think about type/ref)
-python3 tools/taiga.py https://tree.taiga.io/project/penpot/issue/13714
+python3 scripts/taiga.py https://tree.taiga.io/project/penpot/issue/13714

 # By type and ref
-python3 tools/taiga.py us 14128
-python3 tools/taiga.py task 13648
+python3 scripts/taiga.py us 14128
+python3 scripts/taiga.py task 13648

 # Raw JSON output
-python3 tools/taiga.py --json issue 13714
+python3 scripts/taiga.py --json issue 13714
 ```

 ### Output
@@ -20,7 +20,7 @@ primary link, with the fix PR inline on the same line.

 - `gh` CLI authenticated (`gh auth status`)
 - Python 3.8+
- `tools/gh.py` helper script available
+- `scripts/gh.py` helper script available

 ## Workflow

@@ -36,21 +36,31 @@ Use the helper script. It uses GraphQL for efficient single-pass fetching

 ```bash
 # All closed issues (default)
-python3 tools/gh.py issues "2.16.0"
+python3 scripts/gh.py issues "2.16.0"

 # Include open issues too
-python3 tools/gh.py issues "2.16.0" --state all
+python3 scripts/gh.py issues "2.16.0" --state all

 # Exclude entries that should not go in the changelog
-python3 tools/gh.py issues "2.16.0" --exclude "release blocker,no changelog"
+python3 scripts/gh.py issues "2.16.0" --exclude "release blocker,no changelog"
 ```

-**Exclusion rules:**
+**Exclusion rules (issue-level):**
 - `no changelog` label — Chore/refactor work that doesn't need a changelog entry
- `Task` issue type — Internal chores are not user-facing; filter these out after fetching
+- `release blocker` label — Blocked issues not yet ready for changelog
+- `Task` issue type — Internal chores are not user-facing; automatically excluded by `gh.py`. Use `--include-tasks` to override.
+- **Rejected project status** — Issues with a "Rejected" status in the "Main" project board are automatically excluded by `gh.py`. This project-level status (independent of the GitHub issue `state`) indicates the issue was rejected from the release. Use `--include-rejected` to override.
+
+**Exclusion rules (PR-level):**
+In addition to issue-level exclusions, PRs with these labels should be
+excluded regardless of their linked issue's labels:
+- `release blocker` — PR is part of a pending release blocker batch
+- `no issue required` — Trivial fix not tracked as an issue

 The script outputs JSON with each entry containing `number`, `title`, `state`,
-`issue_type`, `labels`, and `closing_prs` (the PRs that fix each issue).
+`issue_type`, `labels`, `closing_prs` (the PRs that fix each issue), and
+`project_status` (the "Main" project board status, e.g. "Done", "Rejected",
+or `null` if not tracked in a project).

 ### 3. Identify missing entries (optional)

@@ -58,11 +68,15 @@ If updating from an existing `CHANGES.md`, find issues in the milestone that
 are NOT yet referenced in the changelog:

 ```bash
-python3 tools/gh.py issues "2.16.0" --exclude "release blocker,no changelog" --compare CHANGES.md
+python3 scripts/gh.py issues "2.16.0" --exclude "release blocker,no changelog" --compare CHANGES.md
 ```

 This returns a filtered JSON array with only the missing issues.

+> **Note:** The `--compare` flag checks **issues** only (via issue number
+> references in the changelog). To find merged **PRs** not yet referenced,
+> use the milestone PR cross-reference described in step 10 below.
+
 ### 4. Fetch additional PR details when needed

 When you need more context for specific PRs (e.g. to find the PR author for
@@ -71,18 +85,45 @@ community contribution attribution, or to read the PR body for

 ```bash
 # One or more PR numbers
-python3 tools/gh.py prs 9179 9204 9311
+python3 scripts/gh.py prs 9179 9204 9311

 # From a file
-python3 tools/gh.py prs --file prs.txt
+python3 scripts/gh.py prs --file prs.txt

 # From stdin
-cat prs.txt | python3 tools/gh.py prs --stdin
+cat prs.txt | python3 scripts/gh.py prs --stdin
+```
+
+The `prs` command also supports listing all PRs in a milestone in one call:
+
+```bash
+# All merged PRs in a milestone (default)
+python3 scripts/gh.py prs --milestone "2.16.0"
+
+# All states (merged, open, closed)
+python3 scripts/gh.py prs --milestone "2.16.0" --state all
 ```

 The `prs` command returns JSON with `number`, `title`, `body`, `state`,
 `merged_at`, `author`, `labels`, and `closing_issues`. PRs are fetched in
-batches of 50 via GraphQL to stay within API limits.
+batches of 50 via GraphQL to stay within API limits (milestone mode uses
+paginated GraphQL on the milestone's `pullRequests` connection).
+
+You can also list all PRs in a milestone in a single call:
+
+```bash
+# All merged PRs in a milestone (default)
+python3 scripts/gh.py prs --milestone "2.16.0"
+
+# All states (merged, open, closed)
+python3 scripts/gh.py prs --milestone "2.16.0" --state all
+
+# Open PRs only
+python3 scripts/gh.py prs --milestone "2.16.0" --state open
+```
+
+The milestone path uses paginated GraphQL on the milestone's `pullRequests`
+connection (100 per page), avoiding one-by-one fetches.

 ### 5. Categorize entries — strictly by issue type, never by labels or emoji

@@ -106,6 +147,12 @@ belongs to.
 The `gh.py` issues command already includes `issue_type` in every entry's
 output. **No separate GraphQL query is needed.**

+**Preserve highlighted entries:** If an entry is already featured in
+`### :rocket: Epics and highlights`, keep it in that section when refreshing a
+changelog version. Do not remove a highlighted entry just because issue type
+categorization would otherwise place it under `### :sparkles: New features &
+Enhancements`.
+
 **Community contribution attribution:** If the issue or its fix PR has the
 `community contribution` label, add an attribution `(by @<github_username>)`
 on the changelog entry line, **before** the GitHub issue/PR references.
@@ -114,7 +161,7 @@ The attribution should reference the **PR author**, not the issue author.
 The `prs` subcommand includes the `author` field — use that:

 ```bash
-python3 tools/gh.py prs <PR_NUMBER> | python3 -c "import sys,json; print(json.load(sys.stdin)[0]['author'])"
+python3 scripts/gh.py prs <PR_NUMBER> | python3 -c "import sys,json; print(json.load(sys.stdin)[0]['author'])"
 ```

 Placement in the entry line:
@@ -134,6 +181,37 @@ tracked in the milestone.
 | PR exists with no linked issue | If a corresponding closed issue exists in the same milestone, link the issue. Otherwise, skip the entry (the issue must be the changelog unit). |
 | Closed issue with no fix PR in milestone | Link the issue directly, without a PR reference. |

+> **False-positive associations:** A PR may incorrectly claim to close an issue
+> from a different context (e.g., a very old PR referencing a modern issue, or a
+> cross-project reference). If the PR title and issue title are clearly unrelated,
+> or the PR was created years before the issue, treat it as a data glitch and
+> skip it. PR [#3](https://github.com/penpot/penpot/pull/3) (ancient License PR
+> claiming to close a plugin API issue) is a known example.
+
+### 5a. ⚠️ Verify PR merge status before writing
+
+A closed issue may list closing PRs that were **closed without merging**
+(e.g., a community PR that was superseded by another). The changelog must
+only reference **merged** PRs. Verify before writing:
+
+```bash
+# Collect all PR numbers from the candidate entries and check them
+python3 scripts/gh.py prs <ALL_PR_NUMBERS> | python3 -c "
+import json, sys
+for pr in json.load(sys.stdin):
+    if pr['state'] != 'MERGED':
+        print(f'WARNING: #{pr[\"number\"]} is {pr[\"state\"]} (not merged)')
+"
+```
+
+If a closing PR is closed-unmerged, find the actual merged PR that
+superseded it:
+1. Check the issue's closing PRs list for other PRs (there may be multiple)
+2. Look for other PRs with similar titles or descriptions referencing the same issue
+3. Inspect the closed PR's conversation timeline for a pointer to the replacement
+
+Replace the reference in the changelog entry with the correct merged PR number.
+
 ### 6. Read the current CHANGES.md

 Read the top of `CHANGES.md` to understand the existing format and find the
@@ -169,6 +247,65 @@ Format details:
 - When an entry already exists in an earlier version section, it must be removed
  from the current version to avoid duplicates

+### 6a. Pre-flight checks — fix rule violations in the changelog
+
+**The LLM must apply these checks during the workflow and fix any
+violations directly in `CHANGES.md`. They are not anomalies — they are
+process errors that should be corrected before writing the new section.**
+
+The changelog is a *snapshot* of the milestone at a point in time, but
+milestones and changelog entries can drift. The LLM must reconcile the
+existing changelog against the current state of the milestone and the
+existing changelog entries.
+
+For each entry that already exists in `CHANGES.md` (in any version
+section) or in the candidate set for the current milestone, check:
+
+1. **Duplicate across versions.** Is the same issue already documented
+   in another (older) version section? If yes, this is a *backport*:
+   - The user-facing fix was already released. Remove the duplicate
+     from the current section. The earlier version is the canonical
+     reference.
+
+2. **Stale milestone assignment.** Has the issue been moved out of the
+   current milestone since the changelog was last updated (e.g., a fix
+   arrived late and the issue was reassigned to a future milestone)?
+   - Verify the issue is still in the current milestone via
+     `python3 scripts/gh.py issues <MILESTONE> --state all`. If it's no
+     longer there, remove the entry from the current section. (If the
+     target section doesn't exist yet, the entry is simply dropped.)
+
+3. **Exclusion labels newly applied.** Did the issue acquire a
+   `no changelog` or `release blocker` label since the changelog was
+   last updated? If yes, remove the entry from the current section.
+
+4. **Issue state changed.** Is the issue still closed? Has it been
+   reopened, deleted, or moved to a `Rejected` project status? If yes,
+   remove the entry.
+
+5. **Unmerged or removed PR references.** For every PR referenced in
+   the entry, is the PR still merged? Was the PR closed without
+   merging (superseded)? Was the PR moved to a different milestone?
+   If the only referenced PR is no longer merged, fix the reference
+   (find the actual merged fix PR) or remove the entry. A PR that is
+   merged in a *different* milestone is reported as an anomaly in
+   step 11 — do not silently remove it.
+
+6. **Issue type changed.** Did the issue type change (e.g., from Bug to
+   Task)? If the new type is `Task`, the issue is internal and should
+   be removed.
+
+7. **Cross-section completeness.** For every closed, non-excluded
+   milestone issue that is *not* referenced in any version section of
+   the changelog, add it to the current section (per the categorization
+   rules in step 5).
+
+After these checks, the changelog should be internally consistent with
+the milestone. **Do not defer these fixes to step 11 — they are
+workflow errors, not anomalies.** Step 11 only reports milestone
+mismatches that require human judgment about the team's release
+intent.
+
 ### 7. Build the description text

 Derive the description from the issue title, not the PR title. Strip leading
@@ -198,6 +335,72 @@ Read the top of `CHANGES.md` and confirm:
 - The section ordering is correct (newest first)
 - Formatting matches the surrounding entries

+### 10. Cross-reference milestone PRs against the changelog
+
+Issues can be fixed by PRs that aren't in the milestone, and merged PRs in
+the milestone may not close any tracked issue. After writing, run a full
+cross-reference to catch gaps:
+
+```bash
+# List all merged PRs in the milestone
+python3 scripts/gh.py prs --milestone "<MILESTONE>" --state merged > /tmp/milestone-prs.json
+
+# Extract PR numbers from the changelog section
+python3 -c "
+import json, re
+
+with open('CHANGES.md') as f:
+    content = f.read()
+
+# Extract the version section (adjust regex to match the actual version)
+match = re.search(r'## <MILESTONE> \(Unreleased\)\n(.*?)(?:\n## |\Z)', content, re.DOTALL)
+section = match.group(1)
+
+# Collect all PR numbers referenced
+changelog_prs = set()
+for m in re.findall(r'\[#(\d+)\]\(https://github\.com/penpot/penpot/pull/\d+\)', section):
+    changelog_prs.add(int(m))
+
+# Collect all milestone PRs (filtered)
+with open('/tmp/milestone-prs.json') as f:
+    milestone_prs = json.load(f)
+
+milestone_merged = {pr['number'] for pr in milestone_prs}
+
+# PRs in milestone but not in changelog
+missing = sorted(milestone_merged - changelog_prs)
+print(f'Milestone merged PRs: {len(milestone_merged)}')
+print(f'Changelog referenced PRs: {len(changelog_prs)}')
+print(f'PRs in milestone but NOT in changelog: {len(missing)}')
+for num in missing:
+    pr = next(p for p in milestone_prs if p['number'] == num)
+    print(f'  #{num} {pr[\"title\"][:80]}')
+"
+```
+
+For each missing PR found, decide whether it should be added to the
+changelog or is legitimately excluded (check its labels).
+
+Also verify that no closed-unmerged PRs remain in the changelog:
+
+```bash
+python3 scripts/gh.py prs --milestone "<MILESTONE>" --state all | python3 -c "
+import json, sys
+data = json.load(sys.stdin)
+closed = [p for p in data if p['state'] == 'CLOSED']
+if closed:
+    print('WARNING: CLOSED (unmerged) PRs in milestone:')
+    for p in closed:
+        print(f'  #{p[\"number\"]} {p[\"title\"][:80]}')
+"
+```
+
+**Post-edit audit checklist:**
+- ✅ All referenced PRs are merged (no closed-unmerged artifacts)
+- ✅ Every merged milestone PR is either in the changelog or excluded by label
+- ✅ PR and issue counts are internally consistent
+- ✅ No false-positive PR-to-issue associations
+
 ## Version section template

 ```markdown
@@ -209,6 +412,292 @@ Read the top of `CHANGES.md` and confirm:
 - <fix description> (by @contributor) [#<ISSUE>](https://github.com/penpot/penpot/issues/<ISSUE>) (PR: [#<PR>](https://github.com/penpot/penpot/pull/<PR>))
 ```

+### 11. Generate anomaly report and save to CHANGES-ISSUES.md
+
+After all edits and cross-referencing are complete, generate a structured
+report and save it to `CHANGES-ISSUES.md` (overwriting if exists).
+This provides a persistent record of any discrepancies between the milestone
+and the changelog.
+
+**Every issue and PR number in the report must be rendered as a full GitHub
+Markdown link** using the same URL format as `CHANGES.md`:
+- Issue N → `[#N](https://github.com/penpot/penpot/issues/N)`
+- PR N → `[#N](https://github.com/penpot/penpot/pull/N)`
+
+The titles and notes should also link to the corresponding issue/PR page
+where applicable, so the report is self-contained and clickable from any
+Markdown viewer.
+
+## What is an anomaly
+
+**An anomaly is a milestone-mismatch between an issue and its referenced
+PR.** It indicates that the changelog claim "this issue is fixed by this PR,
+all in milestone M" is inconsistent with the actual milestone assignments.
+There are exactly two types:
+
+1. **Issue is in the milestone, but its referenced PR is in a different
+   milestone (or has no milestone).** The changelog claims a fix in this
+   release, but the PR is being released elsewhere — the fix may not
+   actually ship here.
+2. **PR is in the milestone, but the issue it closes is in a different
+   milestone (or has no milestone).** The PR is being released here, but
+   the issue it fixes is being released in a different version (or never
+   tracked in a milestone) — the changelog pairing is misleading.
+
+**Anything else is not an anomaly.** Other discrepancies (exclusion
+labels on in-changelog issues, missing valid issues, unmerged PR
+references, duplicates across versions, stale milestone assignments)
+are **rule violations** that the LLM must fix directly in `CHANGES.md`
+during step 6a (pre-flight checks). They should not appear in this
+report — if they do, the LLM has skipped the pre-flight step and
+needs to re-run the workflow.
+
+The changelog's primary unit is the **issue**, not the PR, so a missing or
+mismatched PR only matters when its issue is part of this milestone.
+
+Run this self-contained script:
+
+```bash
+python3 << 'PYEOF'
+import json, re, subprocess, sys
+from datetime import datetime, timezone
+
+MILESTONE = "<MILESTONE>"
+CHANGES_MD = "CHANGES.md"
+OUTPUT = "CHANGES-ISSUES.md"
+REPO = "penpot/penpot"
+
+# --- URL helpers (match CHANGES.md format exactly) ---
+def issue_url(n):  return f"https://github.com/{REPO}/issues/{n}"
+def pr_url(n):     return f"https://github.com/{REPO}/pull/{n}"
+def issue_link(n): return f"[#{n}]({issue_url(n)})"
+def pr_link(n):    return f"[#{n}]({pr_url(n)})"
+def issue_link_title(n, title):
+    url = issue_url(n)
+    if title:
+        return f"[#{n}]({url}) — [{title}]({url})"
+    return f"[#{n}]({url})"
+def pr_link_title(n, title):
+    url = pr_url(n)
+    if title:
+        return f"[#{n}]({url}) — [{title}]({url})"
+    return f"[#{n}]({url})"
+def fmt_pr_list(nums):
+    return ", ".join(pr_link(n) for n in nums)
+def fmt_issue_list(nums):
+    return ", ".join(issue_link(n) for n in nums)
+
+# --- Fetch milestone data ---
+result = subprocess.run(
+    ["python3", "scripts/gh.py", "issues", MILESTONE, "--state", "all"],
+    capture_output=True, text=True)
+all_issues = json.loads(result.stdout)
+issue_by_num = {i['number']: i for i in all_issues}
+
+result = subprocess.run(
+    ["python3", "scripts/gh.py", "prs", "--milestone", MILESTONE, "--state", "all"],
+    capture_output=True, text=True)
+all_prs = json.loads(result.stdout)
+pr_by_num = {p['number']: p for p in all_prs}
+
+# --- Read changelog section ---
+with open(CHANGES_MD) as f:
+    content = f.read()
+
+m = re.search(rf'## {re.escape(MILESTONE)}(?:\s*\([^)]*\))?\n(.*?)(?:\n## |\Z)', content, re.DOTALL)
+section = m.group(1) if m else ""
+
+changelog_issues = set()
+for num in re.findall(r'\[#(\d+)\]\(https://github\.com/penpot/penpot/issues/\d+\)', section):
+    changelog_issues.add(int(num))
+for num in re.findall(r'\[Github #(\d+)\]', section):
+    changelog_issues.add(int(num))
+
+changelog_prs = set()
+for num in re.findall(r'\[#(\d+)\]\(https://github\.com/penpot/penpot/pull/\d+\)', section):
+    changelog_prs.add(int(num))
+for num in re.findall(r'PR:\[(\d+)\]', section):
+    changelog_prs.add(int(num))
+
+# --- Milestone lookup caches ---
+# PRs and issues returned by milestone queries are KNOWN to be in MILESTONE.
+# For everything else, fall back to `gh` per-item lookups.
+pr_milestone_cache = {p['number']: MILESTONE for p in all_prs}
+issue_milestone_cache = {i['number']: MILESTONE for i in all_issues}
+
+def get_pr_milestone(pr_num):
+    """Return the milestone title for a PR, or None if unassigned / unknown."""
+    if pr_num in pr_milestone_cache:
+        return pr_milestone_cache[pr_num]
+    try:
+        r = subprocess.run(
+            ["gh", "pr", "view", str(pr_num), "--json", "milestone"],
+            capture_output=True, text=True, check=True)
+        data = json.loads(r.stdout)
+        ms = data.get('milestone')
+        pr_milestone_cache[pr_num] = (ms or {}).get('title')
+    except (subprocess.CalledProcessError, json.JSONDecodeError):
+        pr_milestone_cache[pr_num] = None
+    return pr_milestone_cache[pr_num]
+
+def get_issue_milestone(issue_num):
+    """Return the milestone title for an issue, or None if unassigned / unknown."""
+    if issue_num in issue_milestone_cache:
+        return issue_milestone_cache[issue_num]
+    try:
+        r = subprocess.run(
+            ["gh", "issue", "view", str(issue_num), "--json", "milestone"],
+            capture_output=True, text=True, check=True)
+        data = json.loads(r.stdout)
+        ms = data.get('milestone')
+        issue_milestone_cache[issue_num] = (ms or {}).get('title')
+    except (subprocess.CalledProcessError, json.JSONDecodeError):
+        issue_milestone_cache[issue_num] = None
+    return issue_milestone_cache[issue_num]
+
+# --- Exclusion rules (shared) ---
+EXCLUDED_LABELS = {'release blocker', 'no changelog'}
+EXCLUDED_ISSUE_TYPES = {'Task'}
+EXCLUDED_PROJECT_STATUS = {'Rejected'}
+
+def issue_excluded(issue):
+    if not issue: return True
+    if issue.get('state') != 'CLOSED': return True
+    if issue.get('issue_type') in EXCLUDED_ISSUE_TYPES: return True
+    if issue.get('project_status') in EXCLUDED_PROJECT_STATUS: return True
+    if EXCLUDED_LABELS & set(issue.get('labels', [])): return True
+    return False
+
+# --- ANOMALIES: milestone mismatches between issues and their referenced PRs ---
+# These are the ONLY items that should appear in the report. All other
+# discrepancies (exclusion labels, missing valid issues, unmerged PRs,
+# duplicates, stale milestone assignments) are workflow errors that the
+# LLM must fix in step 6a (pre-flight checks) — they are not anomalies.
+
+# Type A: issue in MILESTONE, referenced PR in different milestone or no milestone
+anomalies_a = []  # list of dicts: {issue, issue_title, pr, pr_milestone}
+for issue_num in sorted(changelog_issues):
+    issue = issue_by_num.get(issue_num)
+    if not issue: continue
+    if get_issue_milestone(issue_num) != MILESTONE: continue
+    for pr_num in issue.get('closing_prs', []):
+        pr_ms = get_pr_milestone(pr_num)
+        if pr_ms != MILESTONE:
+            anomalies_a.append({
+                'issue': issue_num,
+                'issue_title': issue.get('title', ''),
+                'pr': pr_num,
+                'pr_milestone': pr_ms,  # may be None
+            })
+
+# Type B: PR in MILESTONE, the issue it closes is in different milestone or no milestone
+anomalies_b = []  # list of dicts: {pr, pr_title, issue, issue_milestone}
+for pr_num in sorted(changelog_prs):
+    pr = pr_by_num.get(pr_num)
+    if not pr: continue
+    if get_pr_milestone(pr_num) != MILESTONE: continue
+    for issue_num in pr.get('closing_issues', []):
+        issue_ms = get_issue_milestone(issue_num)
+        if issue_ms != MILESTONE:
+            anomalies_b.append({
+                'pr': pr_num,
+                'pr_title': pr.get('title', ''),
+                'issue': issue_num,
+                'issue_milestone': issue_ms,  # may be None
+            })
+
+# --- Write report ---
+def fmt_ms(ms):
+    return ms if ms else "_none_"
+
+with open(OUTPUT, 'w') as f:
+    f.write(f'# Changelog Anomaly Report — {MILESTONE}\n\n')
+    f.write(f'Generated: {datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M UTC")}\n\n')
+    f.write('---\n\n')
+
+    n_a = len(anomalies_a)
+    n_b = len(anomalies_b)
+
+    f.write('## Summary\n\n')
+    f.write(f'- **Issue in {MILESTONE}, referenced PR in different milestone or no milestone:** {n_a}\n')
+    f.write(f'- **PR in {MILESTONE}, closing issue in different milestone or no milestone:** {n_b}\n')
+    f.write(f'- **Total anomalies:** {n_a + n_b}\n\n')
+
+    # --- Anomalies section ---
+    if n_a or n_b:
+        f.write('## Anomalies\n\n')
+        f.write('These are milestone mismatches between an issue in the changelog '
+                'and its referenced PR (or vice-versa). The changelog claim '
+                '"this issue is fixed by this PR, all in this milestone" is '
+                'inconsistent with the actual milestone assignments. '
+                'Resolve by either updating the milestone on the issue/PR or '
+                'removing the misleading entry from the changelog.\n\n')
+
+        if n_a:
+            f.write(f'### Issue in {MILESTONE}, PR in different milestone or no milestone\n\n')
+            by_issue = {}
+            for a in anomalies_a:
+                by_issue.setdefault(a['issue'], []).append(a)
+            for issue_num in sorted(by_issue):
+                entries = by_issue[issue_num]
+                title = entries[0]['issue_title']
+                f.write(f'- {issue_link_title(issue_num, title[:80])}\n')
+                for e in entries:
+                    ms_label = fmt_ms(e['pr_milestone'])
+                    badge = '🔴' if e['pr_milestone'] is None else '⚠️'
+                    f.write(f'  - {badge} Referenced {pr_link(e["pr"])} is in milestone **{ms_label}** (expected: {MILESTONE})\n')
+                f.write('\n')
+
+        if n_b:
+            f.write(f'\n### PR in {MILESTONE}, closing issue in different milestone or no milestone\n\n')
+            by_pr = {}
+            for b in anomalies_b:
+                by_pr.setdefault(b['pr'], []).append(b)
+            for pr_num in sorted(by_pr):
+                entries = by_pr[pr_num]
+                title = entries[0]['pr_title']
+                f.write(f'- {pr_link_title(pr_num, title[:80])}\n')
+                for e in entries:
+                    ms_label = fmt_ms(e['issue_milestone'])
+                    badge = '🔴' if e['issue_milestone'] is None else '⚠️'
+                    f.write(f'  - {badge} Closing {issue_link(e["issue"])} is in milestone **{ms_label}** (expected: {MILESTONE})\n')
+                f.write('\n')
+    else:
+        f.write('✅ No anomalies found. All (issue, PR) pairs in the changelog have aligned milestone assignments.\n\n')
+
+    # --- Context ---
+    f.write('---\n\n')
+    f.write('## Context\n\n')
+    f.write(f'- Milestone: **{MILESTONE}**\n')
+    f.write(f'- Milestone total issues (all states): {len(all_issues)}\n')
+    f.write(f'- Closed issues in milestone: {sum(1 for i in all_issues if i.get("state") == "CLOSED")}\n')
+    f.write(f'- Valid issues after exclusions (after step 5/6a): {len([i for i in all_issues if not issue_excluded(i)])}\n')
+    f.write(f'- Issues referenced in changelog: {len(changelog_issues)}\n')
+    f.write(f'- PRs referenced in changelog: {len(changelog_prs)}\n')
+
+print(f"Anomaly report written to {OUTPUT}")
+PYEOF
+```
+
+This generates `CHANGES-ISSUES.md` containing **only the anomalies** —
+milestone mismatches between issues and their referenced PRs:
+
+1. **Issue in milestone, referenced PR in different milestone or no milestone** —
+   the changelog claims a fix here, but the PR is released elsewhere.
+2. **PR in milestone, closing issue in different milestone or no milestone** —
+   the PR is released here, but the issue it fixes belongs to another version.
+
+**Rule violations are not in the report** — they are workflow errors the
+LLM must fix directly in `CHANGES.md` during step 6a (pre-flight checks).
+If the report contains a rule violation, the LLM has skipped the pre-flight
+step and needs to re-run the workflow before re-generating the report.
+
+The report is overwritten each time it's generated, reflecting the current
+state of the milestone and changelog. Every number is rendered as a full
+`[#N](https://github.com/penpot/penpot/issues/N)` or
+`[#N](https://github.com/penpot/penpot/pull/N)` link so the report is
+self-contained and clickable in any Markdown viewer.
+
 ## Key Principles

 - **Issue = changelog unit.** The primary link always points to the
@@ -225,7 +714,11 @@ Read the top of `CHANGES.md` and confirm:
  between the description and the issue link. Use the **PR author** (not the
  issue author) for the attribution.
 - **Only closed issues.** An issue must have `state: "closed"` to appear in
-  the changelog. Open unresolved issues are omitted.
+  the changelog. Open/unresolved issues are omitted.
+- **Rejected project status.** Issues marked as "Rejected" in the "Main"
+  project board are automatically excluded by `gh.py`, even if they are
+  closed. The project status is distinct from the GitHub issue state.
+  Use `--include-rejected` to override this behavior.
 - **Excluded issues.** Issues with `no changelog` label must be excluded.
  Issues with `issue_type: "Task"` must also be excluded — they are internal
  chores, not user-facing changes.
@@ -241,6 +734,31 @@ Read the top of `CHANGES.md` and confirm:
  reference if applicable.
 - **Re-fetch before editing.** Milestones can change — always re-fetch issues
  before making edits, don't rely on cached data.
- **Use `tools/gh.py`.** Prefer the helper script over raw `gh api` calls for
+- **Use `scripts/gh.py`.** Prefer the helper script over raw `gh api` calls for
  milestone issue listing and PR detail fetching. It handles GraphQL
  pagination, batching, and label filtering automatically.
+- **Verify PR merge status.** Not all closing PRs are merged — community PRs
+  can be superseded and closed without merging. Always check that every PR
+  referenced in the changelog has `state: MERGED`.
+- **PR-level exclusions apply.** A PR can carry its own exclusion labels
+  (`release blocker`, `no issue required`) independent of its linked issue's
+  labels. Check both.
+- **Cross-reference milestone PRs, not just issues.** The `--compare` flag on
+  the `issues` command only compares issue numbers. Merged PRs not linked to
+  any milestone issue can be missed. Use `python3 scripts/gh.py prs --milestone`
+  for a full PR cross-reference.
+- **False-positive PR-to-issue associations.** A PR may claim to close an
+  issue from a different project or context. If the PR title and issue title
+  are clearly unrelated, or the PR predates the issue by years, treat it as a
+  data glitch and skip it.
+- **Anomaly = milestone mismatch only.** The report contains only milestone
+  mismatches: (1) the issue is in this milestone but the referenced PR is
+  in a different milestone (or unassigned), and (2) the PR is in this
+  milestone but the issue it closes is in a different milestone (or
+  unassigned). These are anomalies because the changelog pairing is
+  *misleading* — the human needs to decide whether the milestone or the
+  changelog is wrong. All other discrepancies (exclusion labels, missing
+  valid issues, unmerged PR references, duplicates, stale milestone
+  assignments) are **rule violations** that the LLM must fix directly in
+  `CHANGES.md` during step 6a (pre-flight checks). They never appear in
+  the report — if they do, the pre-flight step was skipped.
@@ -0,0 +1,40 @@
+# Backend Auth, Permissions, and Product Domain Subtleties
+
+## Auth and sessions
+
+- Main auth RPC commands live in `app.rpc.commands.auth`; LDAP and OIDC provider logic live in `app.auth.ldap` and `app.auth.oidc`, with LDAP-specific RPC checks in `app.rpc.commands.ldap`.
+- Public auth endpoints must explicitly set `::rpc/auth false`; RPC auth defaults to enabled. Session cookie creation/deletion is usually attached as an RPC response transform.
+- Basic Penpot registration is token staged: prepare/register creates or verifies temporary tokens, then profile creation/session setup is reused by other auth backends. The frontend `/auth/verify-token` flow is a hub for registration confirmation, email change, and invitation tokens.
+- OIDC-compatible providers share a generic flow: redirect to provider, validate callback/request token, fetch identity data, then login an existing profile or register a new one. Known providers may have hardcoded endpoints; generic OIDC can use discovery/configured endpoints.
+- LDAP login validates credentials against the external directory, fetches identity data, then logs in or registers a matching Penpot profile. LDAP registration is not a separate Penpot signup flow.
+- Logout may return an OIDC provider redirect URI when the session claims include provider/session data and the provider has a logout URI.
+- Invitation tokens are verified through token issuers and only accepted when the token member id/email matches the authenticated profile; otherwise login proceeds without consuming the invitation.
+- HTTP/session parsing details such as cookie/header precedence, JWT session token versions, and SameSite behavior are in `mem:backend/http-storage-filedata-subtleties`.
+
+## Permission model
+
+- `app.rpc.permissions` provides predicate/check factories. Failed permission checks intentionally raise `:not-found` / `:object-not-found`, not an authorization-specific error, to avoid leaking object existence.
+- Team role flags are normalized as owner > admin > editor > viewer. Owner/admin imply edit; any membership row implies read.
+- File/project/comment checks are implemented in the owning command namespaces, often via helpers imported from `files`, `teams`, or `projects`; do not bypass those helpers with direct DB lookups unless preserving their not-found semantics.
+- Comment permission includes both logged-in state and the file/team comment policy. Shared viewer paths may pass `share-id`; preserve that path when changing comment queries.
+
+## Teams, projects, and invitations
+
+- Team/project commands mix DB changes, email, message bus notifications, media/storage cleanup, feature flags, quotas, and audit metadata. Keep mutations transactional when the existing command does so.
+- Invitation flows validate muted/bounced emails before sending and use tokenized invitation state. Accepting an invitation is tied to the invited member identity, not just possession of a token.
+- Logical deletion is used for many product objects; prefer existing logical-deletion helpers over hard deletes unless the command already performs permanent cleanup.
+- Bounced/spam-complaint emails can mute/block a profile for login/registration and email sending. Devenv MailCatcher is the normal local path for registration/email-flow testing.
+
+## Comments, webhooks, and audit
+
+- Comment thread queries join file/project/profile state and exclude deleted files/projects. Unread comment counts depend on `comment_thread_status.modified-at` and profile notification preferences.
+- Webhook edits are allowed for team editors/admins or the webhook creator. Webhook validation performs a synchronous HEAD request with a short timeout; validation errors are mapped through `app.loggers.webhooks`.
+- Audit events are prepared from RPC metadata, result metadata, params, request context, and selected auth identifiers. Webhook event batching can be controlled through audit/webhook metadata on commands or results.
+- Webhook and audit logging are cross-cutting side effects of product commands; when adding a command, check nearby command metadata and result metadata patterns before inventing a new event shape.
+
+## Local testing notes
+
+- Enable LDAP login locally with frontend flag `enable-login-with-ldap`; the devenv includes a configured test LDAP service.
+- OIDC testing requires external provider app credentials plus matching backend/frontend config.
+- Backend domain tests usually live under `backend/test/backend_tests/rpc/commands/*_test.clj` or nearby backend test namespaces. Use focused `clojure -M:dev:test --focus ...` from `backend/` when possible.
+- For auth/session or HTTP behavior, combine backend tests with the HTTP/session notes in `mem:backend/http-storage-filedata-subtleties` because RPC-level tests may not exercise cookie/header transforms.
@@ -0,0 +1,110 @@
+# Backend Architecture and Workflow
+
+Backend: JVM Clojure; Integrant; PostgreSQL; Redis/Valkey; RPC; HTTP; storage; mail; audit/logging; workers.
+
+## Focused memories
+
+- RPC, DB helpers, workers, cron: `mem:backend/rpc-db-worker-subtleties`
+- HTTP sessions, config, storage, media, file data persistence: `mem:backend/http-storage-filedata-subtleties`
+- Auth flows, permission model, teams, projects, invitations, comments, webhooks, audit: `mem:backend/auth-permissions-product-domains`
+- Services, task-queue/Pub-Sub topology constraints -> `mem:prod-infra/core`.
+
+## Stable namespace map
+
+- `app.rpc.commands.*`: RPC command implementations exposed under `/api/rpc/command/<cmd-name>`.
+- `app.rpc.permissions`: permission predicate/check helper factories.
+- `app.http.*`: HTTP routes and middleware.
+- `app.auth.*`: provider-specific authentication helpers such as LDAP/OIDC.
+- `app.loggers.*`: audit, webhook, database, and external log integrations.
+- `app.db.*` / `app.db`: next.jdbc wrapper and SQL helpers.
+- `app.tasks.*`: background task handlers.
+- `app.worker`: task execution/cron plumbing.
+- `app.main`: Integrant system map and component wiring.
+- `app.config`: `PENPOT_*` env config and feature flags.
+- `app.srepl.*`: development REPL helpers for manual backend operations (data inspection, migration helpers, one-off admin tasks).
+- `app.nitrate`, `app.rpc.commands.nitrate`, and `app.rpc.management.nitrate`: external Nitrate subscription/organization integration, gated by the `:nitrate` feature flag and shared-key HTTP calls.
+
+## RPC conventions
+
+RPC commands are defined with `app.util.services/defmethod` and schemas. Use `get-` prefixes for read operations. Command metadata usually includes auth, docs version, params schema, and result schema. Return plain maps/vectors or raise structured exceptions from `app.common.exceptions`.
+
+Backend RPC command areas without focused memories include access tokens, binfile, demo, feedback, file snapshots, fonts, management, Nitrate, and webhooks beyond the notes in `mem:backend/auth-permissions-product-domains`; inspect nearby command tests and command metadata before changing them.
+
+## DB conventions
+
+`app.db` helpers accept cfg, pool, or conn in most places and convert kebab-case to snake_case:
+- `db/get`, `db/get*`, `db/query`, `db/insert!`, `db/update!`, `db/delete!`.
+- Use `db/run!` for multiple operations on one connection.
+- Use `db/tx-run!` for transactions.
+
+Database migrations live in `backend/src/app/migrations/`; pure SQL migrations are under `backend/src/app/migrations/sql/`. SQL filenames conventionally start with a sequence and verb/table description, e.g. `0026-mod-profile-table-add-is-active-field`. Applied migrations are tracked in the `migrations` table.
+
+For interactive PostgreSQL access with correct dev defaults, use `scripts/psql`; to dump
+the current DDL schema, use `scripts/db-schema` (see `mem:scripts/psql`).
+
+For deeper details on transaction semantics, advisory locks, Transit vs JSON helpers, and dev/test DB URLs: `mem:backend/rpc-db-worker-subtleties`.
+
+## Background tasks
+
+A task handler is an Integrant component with `ig/assert-key`, `ig/expand-key`, and `ig/init-key`, returning the function run by the worker. New tasks also need wiring in `app.main`: handler config, worker registry entry, and cron entry if scheduled.
+
+For worker dispatch, cron, retry semantics, deduplication, and queue internals: `mem:backend/rpc-db-worker-subtleties`.
+
+## REPL
+
+In devenv, backend nREPL is exposed on port 6064.
+
+### Non-interactive eval (preferred for agents)
+
+`./scripts/nrepl-eval.mjs` connects to an already-running nREPL server and evaluates code. Session state (defs, `in-ns`) persists across invocations via a stored session ID in `/tmp/penpot-nrepl-session-<host>-<port>`.
+
+```bash
+./scripts/nrepl-eval.mjs '(+ 1 2)'                            # single expression
+./scripts/nrepl-eval.mjs "(require '[my.ns :as ns] :reload)"  # reload after edits
+./scripts/nrepl-eval.mjs -e                                   # inspect last exception (*e)
+./scripts/nrepl-eval.mjs --reset-session '(def x 0)'          # discard session, start fresh
+./scripts/nrepl-eval.mjs <<'EOF'                              # multi-expression heredoc
+(def x 10)
+(+ x 20)
+EOF
+```
+
+Default port is 6064. Use `-p <PORT>` for a different port. Use `-t <MS>` to override the 120s timeout. Do not start the nREPL server — assume it is already running.
+
+### Interactive REPL
+
+`backend/scripts/nrepl` starts a REPLy client connected to the running nREPL.
+
+For an in-process backend REPL (where you control the JVM lifecycle), stop the running backend first so port 9090 is free, then run `backend/scripts/repl`. Useful top-level helpers include `(start)`, `(stop)`, `(restart)`, `(run-tests)`, and `(repl/refresh-all)`. Many `app.srepl.main` helpers accept the global `system` var, e.g. manual email or maintenance operations.
+
+## Fixtures
+
+Fixtures can populate local data for manual testing/perf work. From the backend REPL, run `(app.cli.fixtures/run {:preset :small})`; fixture users conventionally look like `profileN@example.com` with password `123123`. Standalone fixture aliases may exist, but check current `backend/deps.edn` before relying on old command names.
+
+## Performance
+
+* **Type Hinting:** Use explicit JVM type hints (e.g. `^String`, `^long`) in performance-critical paths to avoid reflection overhead.
+* **Batch inserts:** Use `db/insert-many!` for bulk row inserts — generates a single SQL with multiple parameter tuples. Avoid on very large datasets (SQL length / parameter count limits).
+* **Server-side cursors:** Use `db/plan` (fetch-size 1000, forward-only, read-only) or `db/cursor` for large result sets. Never fetch large collections into memory at once.
+* **Transaction discipline:** Use `tx-run!` for writes (opens a transaction), `run!` for reads (single connection, no transaction). Set `:read-only` on `tx-run!` when applicable to let PostgreSQL optimize.
+
+## Lint and Format
+
+IMPORTANT: all CLI commands must be executed from the `backend/` subdirectory.
+
+* **Linting:** `clj-kondo --lint ../common/src/ src/`.
+* **Formatting:** `cljfmt check src/ test/` to check, `cljfmt fix src/ test/` to fix. Avoid unrelated whitespace diffs.
+
+**Before linting:** if delimiter errors are suspected (after LLM edits), run
+`scripts/paren-repair` on the affected files first. Delimiter errors produce
+misleading linter/compiler output. See `mem:scripts/paren-repair`.
+
+## Testing
+
+IMPORTANT: all CLI commands must be executed from the `backend/` subdirectory. JVM tests are invoked directly via `clojure -M:dev:test` — there is no pnpm wrapper. If you need to filter output, tee to a temp file first: `clojure -M:dev:test 2>&1 | tee /tmp/penpot-test-output.txt`. See `mem:testing` for execution discipline.
+
+* **Coverage:** If code is added or modified in `src/`, corresponding tests in `test/backend_tests/` must be added or updated.
+* **Isolated run:** `clojure -M:dev:test --focus backend-tests.my-ns-test` for a specific test namespace.
+* **Regression run:** `clojure -M:dev:test` to ensure no regressions in related functional areas.
+* **Principles:** Cross-cutting testing principles, anti-patterns, and verification checklist: `mem:testing`.
+
@@ -0,0 +1,31 @@
+# Backend HTTP, Storage, Media, and File Data Subtleties
+
+## Config and HTTP/session middleware
+
+- `app.config/config` and `flags` are dynamic `defonce` vars populated from `PENPOT_*` env vars through the shared schema string transformer. Tests and tooling can bind them.
+- `parse-flags` automatically adds `:disable-secure-session-cookies` when `public-uri` is plain HTTP and not localhost. This changes cookie defaults without an explicit env flag.
+- The backend sets Clojure `*assert*` globally from the `:backend-asserts` feature flag. Assertion-dependent checks can therefore differ by runtime flags.
+- Request body parsing is mostly POST-oriented and supports Transit JSON plus plain JSON. Plain JSON request keys are kebab-decoded before being merged into `:params`.
+- Response formatting negotiates with `Accept` or `_fmt=json`. Transit is the default for collection/boolean bodies; JSON encoding has special pointer-map handling.
+- Auth prefers the session cookie token before the `Authorization` header. Headers may be `Token` or `Bearer`; JWTs with `kid=1` and `ver=1` are decoded as v1 session tokens, otherwise they are treated as legacy tokens.
+- Shared-key auth requires `x-shared-key` as `<key-id> <key>` and stores the lowercased key id on the request. If no shared keys are configured it always rejects.
+- Session management uses DB storage unless the DB pool is read-only, then falls back to the in-memory manager. DB sessions support both legacy string ids and v2 UUID session ids.
+- Session cookies are renewed when using a legacy string id or when `modified-at` is older than the renewal interval. SameSite is `none` for CORS, otherwise strict/lax based on config.
+
+## Storage and media
+
+- Storage has a fixed valid bucket set. Backends are `:fs` and `:s3`; default backend comes from deprecated `assets-storage-backend` only when present, otherwise `objects-storage-backend`, defaulting to `:fs`.
+- `put-object!` creates the DB `storage_object` row before writing backend content. Backend writes happen only for newly created rows, so deduplication can skip object writes.
+- Deduplication only applies when requested, when the content can provide a hash, and when bucket metadata is present. Reads exclude soft-deleted storage rows.
+- `sto/resolve` can reuse the current DB connection via `::db/reuse-conn true`; preserve this in transaction-sensitive code.
+- SVG validation strips DOCTYPE and uses secure SAX parsing. Basic SVG info falls back to 100x100 dimensions when width/height/viewBox are missing.
+- Raster metadata is shell-derived with ImageMagick `identify`, verifies detected MIME against the supplied MIME, and swaps dimensions for EXIF orientations 6/8.
+- Remote image download requires 2xx status, `content-length`, a known MIME, and size under the configured maximum before writing the temp file; mismatched byte count is an internal error.
+- Font processing shells out to FontForge and WOFF conversion tools and can derive TTF/OTF/WOFF variants from uploaded fonts.
+
+## File data persistence
+
+- File data backends are `legacy-db`, `db`, and `storage`. The storage backend keeps encoded file data in storage bucket `file-data`; the DB row stores metadata with `storage-ref-id` and nil data.
+- `fdata/upsert!` touches any storage object referenced by incoming metadata before storing the new row/blob.
+- Pointer-map fragments are persisted separately as type `fragment`, and only modified pointer maps are written.
+- `fdata/realize` combines pointer realization and object-map realization. Use it before operations that need complete in-memory file data instead of pointer placeholders.
@@ -0,0 +1,26 @@
+# Backend RPC/DB/Worker Subtleties
+
+## RPC exposure and wrappers
+
+- RPC commands are discovered from vars created by `app.util.services/defmethod`; adding a command namespace is not enough unless `backend/src/app/rpc.clj` includes it in `resolve-methods`.
+- `GET`/`HEAD` RPC calls are only allowed for method names starting with `get-`. Other methods are method-not-allowed even if they are read-only internally.
+- RPC auth defaults to enabled. Public endpoints must set `::auth false` metadata explicitly.
+- The wrapper stack does auth before params validation, then auditing/rate/concurrency/metrics/retry/condition handling, with DB transaction handling inside that stack. `::db/transaction` metadata controls transaction wrapping.
+- Params with `::sm/params` are decoded/conformed through the JSON transformer and successful IObj results get `:encode/json` metadata. Legacy spec conforming only applies when no Malli params schema exists.
+- Nil RPC bodies become HTTP 204 unless explicit status metadata is present. Stream bodies default to `application/octet-stream` when no content type is set.
+
+## DB helpers
+
+- Most `app.db` helpers accept a pool, connection, or map containing `::db/pool` / `::db/conn`; preserve that convention in shared code.
+- `db/tx-run!` uses `next.jdbc.transaction/*nested-tx* :ignore`: nested transaction calls reuse the outer transaction, not a savepoint. Use explicit savepoints when nested rollback semantics matter.
+- `db/run!` opens/reuses one connection but does not create a transaction.
+- `db/tjson` is Transit JSON for jsonb storage; `db/json` is plain JSON. Worker task props use Transit and are decoded with `decode-transit-pgobject`.
+- Advisory transaction locks accept UUIDs or ints. UUID locks are hashed using a zero-UUID seeded siphash.
+
+## Workers and cron
+
+- Task queues are tenant-prefixed. Submit dedupe only removes not-yet-due `new` tasks with the same name/queue/label; it does not dedupe due, scheduled, retry, running, or completed work.
+- The dispatcher selects `new`/`retry` tasks with `FOR UPDATE SKIP LOCKED`, marks them `scheduled`, and publishes Redis payload `[id scheduled-at]`. The runner skips Redis messages whose scheduled timestamp no longer matches DB state.
+- Lost `scheduled` tasks are rescheduled after about 5 minutes; `running` tasks older than about 24 hours are marked failed as orphans.
+- A task handler that is missing or returns an invalid result currently defaults to completed after warning. Throwing with `ex-data :type ::retry` controls retry behavior; `:strategy ::noop` retries without incrementing retry count.
+- Cron jobs lock their `scheduled_task` row with `FOR UPDATE SKIP LOCKED`, disable statement/idle-in-transaction timeouts locally, and reschedule themselves in `finally` unless interrupted. Worker, dispatcher, and cron components do not start when the DB pool is read-only.
@@ -0,0 +1,39 @@
+# File Mutations: Changes and Undo Architecture
+
+Penpot mutates file data through change records. A change set is both the persistence payload and the basis for undo/redo, so UI actions, tests, backend file updates, and library/file tooling should drive the production change pipeline instead of ad hoc object-map mutation.
+
+## Change shape
+
+Each change is a map such as `{:type ... :id ... :page-id ...}`. Common families:
+
+- `:add-obj`, `:mod-obj`, `:del-obj`: shape lifecycle. `:mod-obj` contains `:operations`, commonly `{:type :set :attr ... :val ... :ignore-geometry ... :ignore-touched ...}` or `{:type :set-touched ...}`.
+- `:add-component`, `:mod-component`, `:del-component`: component/library metadata.
+- `:add-children`, `:remove-children`, `:reg-objects`: tree and object-map edits.
+- `:set-option`, `:add-page`, `:mov-page`, and related file/page metadata changes.
+
+Each transaction carries `:redo-changes` and inverse `:undo-changes`. The undo stack stores transactions and can move its index backward/forward.
+
+## changes-builder API
+
+`common/src/app/common/files/changes_builder.cljc` (usually alias `pcb`) is the fluent builder. Start from `(pcb/empty-changes <it> <page-id>)` or `(pcb/empty-changes nil <page-id>)` for tests.
+
+High-value builder operations:
+- `pcb/with-page-id`, `pcb/with-objects`, `pcb/with-library-data`: set context for following operations.
+- `pcb/update-shapes ids update-fn`: emits `:mod-obj` with diff-derived `:set` ops. Options include `{:with-objects? true}`, `{:ignore-touched true}`, and `{:attrs #{...}}`.
+- `pcb/add-objects`, `pcb/change-parent`, `pcb/remove-objects`, `pcb/resize-parents`: shape/tree edits.
+- `pcb/add-component`, `pcb/update-component`, `pcb/mod-component`: component/library edits.
+- `pcb/set-translation? true`: marks the whole change set as translation-only, which lets component sync skip expensive work.
+
+## Applying changes in tests
+
+`thf/apply-changes` in `app.common.test-helpers.files` is the test analog of the production applier. It validates by default; pass `:validate? false` only for intentionally-invalid intermediate states.
+
+The applier uses the same `process-operation` multimethod as production (`common/src/app/common/files/changes.cljc`), so tests that use it exercise production behavior.
+
+## :touched and geometry
+
+For component touched semantics and sync groups, read `mem:common/component-data-model`. For the exact `set-shape-attr` / second-pass behavior during change application, read `mem:common/file-change-validation-migration-subtleties`. For transform-specific ignore-geometry behavior, read `mem:frontend/workspace-transform-subtleties`.
+
+## Inspection
+
+To inspect what a UI action emitted, use `mem:frontend/cljs-repl` with the snippets in `mem:common/component-debugging-recipes` rather than adding temporary source instrumentation.
@@ -0,0 +1,41 @@
+# Component and Variant Data Model
+
+## Shape roles relative to components
+
+A shape can occupy multiple roles at once:
+
+1. Master/main instance: defines a component and has `:main-instance true` plus `:component-id`.
+2. Copy/non-main instance: produced by instantiating a component and carries `:shape-ref` pointing at the master shape. `(ctk/in-component-copy? shape)` is essentially `(some? (:shape-ref shape))`.
+3. Component root: topmost shape of an instance, marked `:component-root true` and carrying surface attrs such as `:component-id` and `:component-file`.
+
+Variant masters are main instances and component roots. Their descendants may themselves be component copies, so master/copy logic must handle nested instances rather than assuming those roles are exclusive.
+
+## :shape-ref chains
+
+`:shape-ref` walks up the inheritance hierarchy and can cross files for remote libraries. `find-ref-shape` and `get-ref-chain-until-target-ref` in `app.common.types.file` follow this chain.
+
+`find-shape-ref-child-of` in `app.common.logic.variants` walks the chain looking for the first ref-shape whose ancestors include a specific parent. Variant switch uses this to locate the equivalent master child in the target variant.
+
+## :touched flags
+
+`:touched` is a set of override-group keywords such as `:geometry-group`, `:fill-group`, and `:text-content-group`. It means a copy diverged from its master for attrs in that sync group.
+
+`sync-attrs` in `app.common.types.component` maps attrs to groups. `set-touched-group` is the legitimate setter; the central `set-shape-attr` path calls it only for copies and only when ignore flags allow it.
+
+Masters are not normally touched through `set-shape-attr`, but touched flags can appear on master shapes through cloning/duplication paths. `add-touched-from-ref-chain` in `app.common.logic.variants` unions touched flags from ancestors into the copy being processed, so upstream/master touched state can affect downstream switch behavior.
+
+## Cloning paths
+
+`make-component-instance` in `app.common.types.container` produces a clean component copy through `update-new-shape`, dissociating attrs such as `:touched`, `:variant-id`, and `:variant-name` on cloned shapes.
+
+`duplicate-component` in `app.common.logic.libraries` creates a new component master by cloning existing component shapes, setting component metadata, and applying a position delta. It does not have the same clean-copy semantics as `make-component-instance`, so inherited attrs on the source can matter.
+
+When a bug depends on touched state, identify which cloning path produced the shape before changing sync logic.
+
+## Variant containers
+
+A variant container is a frame with `:is-variant-container true`. Its children are variant masters with `:variant-id` pointing at the container and `:variant-name` naming the variant value. Component records in the library carry `:variant-properties`.
+
+Predicates are broad: `ctk/is-variant?` checks `:variant-id` and applies to both variant master shapes and component rows; `ctk/is-variant-container?` checks the container shape flag.
+
+Moving/dropping a shape into a variant container through the move-to-frame path can auto-convert it into a variant via `generate-make-shapes-variant`, which may duplicate the underlying component. Treat drag/drop into variant containers as a component/variant operation, not a plain reparent.
@@ -0,0 +1,58 @@
+# Common Component and Change Debugging Recipes
+
+Keep source changes out of these recipes unless the task requires a durable fix.
+
+## Inspect recent workspace changes
+
+From `cljs_repl` after triggering an action:
+
+```clojure
+(let [items (get-in @app.main.store/state [:workspace-undo :items])
+      n (count items)]
+  (->> items
+       (drop (max 0 (- n 5)))
+       (map-indexed (fn [i it]
+                      {:idx (+ i (max 0 (- n 5)))
+                       :tags (:tags it)
+                       :n (count (:redo-changes it))
+                       :types (frequencies (map :type (:redo-changes it)))
+                       :ids (mapv :id (:redo-changes it))}))))
+```
+
+To inspect operations within the latest `:mod-obj`:
+
+```clojure
+(let [items (get-in @app.main.store/state [:workspace-undo :items])
+      mod-obj (->> (:redo-changes (last items))
+                   (filter #(= :mod-obj (:type %)))
+                   first)]
+  (:operations mod-obj))
+```
+
+## Trace variant switch attribute copying
+
+To capture what `update-attrs-on-switch` saw during a real UI swap, patch it temporarily in `cljs_repl`:
+
+```clojure
+(def orig (deref #'app.common.logic.libraries/update-attrs-on-switch))
+(def trace-buf (atom []))
+(set! app.common.logic.libraries/update-attrs-on-switch
+      (fn [& args]
+        (swap! trace-buf conj
+               (let [[_ curr prev _ _ origin _] args]
+                 {:curr (select-keys curr [:name :x :y :selrect :points :touched])
+                  :prev (select-keys prev [:name :x :y :selrect :points :touched])
+                  :origin-ref (select-keys origin [:id :name :x :y :width :height :selrect])}))
+        (apply orig args)))
+;; trigger UI action, then inspect @trace-buf
+(set! app.common.logic.libraries/update-attrs-on-switch orig)
+```
+
+Runtime patching is faster than adding temporary source instrumentation and avoids recompilation cleanup. Restore the var or reload the frontend when finished.
+
+## Test-side helpers
+
+- Use `thf/dump-file file :keys [...]` to print a shape tree with selected keys during common tests.
+- Prefer production-path helpers such as `cls/generate-update-shapes` plus `thf/apply-changes` for shape mutations.
+- For component swaps with keep-touched behavior, use `tho/swap-component-in-shape` with `{:keep-touched? true}`.
+- Temporary `prn` calls in production code are acceptable while investigating but should be removed before committing.
@@ -0,0 +1,42 @@
+# Component Swap and Variant Switch Pipeline
+
+## Entry points
+
+Frontend entry points under `frontend/src/app/main/data/workspace/`:
+- `variants.cljs`: `variants-switch` and `variant-switch` events feed property-toggle UI and Plugin API `switchVariant` behavior into `dwl/component-swap`.
+- `libraries.cljs`: `component-swap` is the single-swap workhorse; `component-multi-swap` batches swaps and calls `component-swap` with `keep-touched? = false`.
+
+`keep-touched? = true` is the discriminator for preserving user overrides during variant switch. Batch/multi-swap paths intentionally bypass that logic.
+
+## Common-side pipeline
+
+For a single swap with `keep-touched? = true`:
+
+1. `cll/generate-component-swap` in `common/src/app/common/logic/libraries.cljc` builds the base changes: remove old shape and instantiate the target component in its place through `generate-new-shape-for-swap`, `generate-instantiate-component`, and `make-component-instance`.
+2. `clv/generate-keep-touched` in `common/src/app/common/logic/variants.cljc` walks pre-swap children, augments each with chain-derived touched flags through `add-touched-from-ref-chain`, finds the equivalent target shape through `find-shape-ref-child-of`, then calls `update-attrs-on-switch`.
+3. `update-attrs-on-switch` in `app.common.logic.libraries` decides which touched attrs from the previous shape should be copied onto the freshly instantiated target shape.
+
+## update-attrs-on-switch hazards
+
+The routine compares `current-shape` (fresh target copy), `previous-shape` (pre-swap shape with chain-derived touched), and `origin-ref-shape` (source variant master's equivalent shape). It loops over sync attrs except `swap-keep-attrs` and copies only attrs that pass several guards:
+
+- skip equal previous/current values;
+- skip equal composite geometry for selected attrs;
+- require the corresponding touched group;
+- for most attrs, require source and target masters to agree;
+- for fixed-size selrect/points/width/height, use dedicated fixed-layout geometry handling;
+- text and path shapes have specialized value conversion paths.
+
+The generic fallback branch copies from `previous-shape`. It represents the intended "carry user override through switch" behavior, but bugs usually appear when guards fail to reject incompatible geometry or master differences before reaching that branch.
+
+## Known sharp edges
+
+- Composite `:selrect` and `:points` bypass the simple different-master skip; width/height checks catch some but not all positional differences.
+- `previous-shape` may be repositioned by destination-root minus origin-root before copying. For normal variant switch this is often zero, but do not assume it for all swap entry points.
+- Touched flags can be inherited through ref chains, so a shape that looks untouched locally may still behave as touched after `add-touched-from-ref-chain`.
+
+## Test harness
+
+`common/src/app/common/test_helpers/compositions.cljc` has `swap-component-in-shape`, which drives `generate-component-swap` plus `generate-keep-touched` with the production `keep-touched?` flag. Use it for focused common tests of variant-switch behavior.
+
+`common/test/common_tests/logic/variants_switch_test.cljc` is the canonical reference suite for swap+touched scenarios. Read nearby tests before adding another case.
@@ -0,0 +1,56 @@
+# Common Architecture and Workflow
+
+`common/` intro: shared CLJC for frontend, backend, exporter, library/file tooling, tests. Small semantic changes can affect multiple runtimes.
+
+## Stable namespace map
+
+- `app.common.data` and `app.common.data.macros`: generic data helpers and performance macros that do not depend on Penpot domain entities.
+- `app.common.types.*`: shared shape/file/page/component/token data types, schemas, predicates, and entity-local operations. `app.common.types.nitrate-permissions` contains shared fail-closed Nitrate organization/team permission rules.
+- `app.common.files.*`: file-level operations, shape tree helpers, change application, migrations, validation, and undo/redo-related logic.
+- `app.common.logic.*`: higher-level workflows/algorithms over files, shapes, components, variants, libraries, tokens, etc.
+- `app.common.geom.*`: geometry helpers and transformations.
+- `app.common.schema` / `app.common.schema.*`: Malli abstraction layer.
+- `app.common.math`, `app.common.time`, `app.common.uuid`, `app.common.json`, etc.: cross-runtime utilities.
+- `app.common.test_helpers.*`: test builders and production-path helpers.
+
+## Layering and cross-runtime rules
+
+Use reader conditionals for platform-specific code. Because CLJC runs on JVM and CLJS targets, avoid assuming browser-only or JVM-only behavior unless the reader conditional isolates it.
+
+Respect the intended abstraction direction in new/refactored code:
+- generic data utilities should not know Penpot domain concepts;
+- `types.*` should preserve invariants for a single domain entity or ADT;
+- `files.*` can coordinate several entities inside a file and preserve referential integrity;
+- `changes*` should adapt serializable change records to lower-level operations and avoid embedding broad business algorithms;
+- `logic.*` and frontend/backend event layers own higher workflow/business behavior.
+
+Some legacy code violates this layering; do not copy those violations into new code when a focused refactor is practical.
+
+## Focused memory routing
+
+Model, schema, and persistence shape:
+- File/page/shape/component attr changes, import/export surfaces, inspector/codegen, and cross-module checklist: `mem:common/data-model-change-checklist`.
+- Token data structures, token import/export, active theme/set semantics, and schema/coercion behavior: `mem:common/tokens-schema-subtleties`.
+
+Geometry and layout:
+- Shape geometry invariants, redundant geometry fields, and geometry-sensitive tests: `mem:common/geometry-invariants`.
+- Coordinate drift and approximate float comparisons: `mem:common/decimals-and-coordinates`.
+- Layout/grid assignment, deassignment, metadata cleanup, and auto-positioning: `mem:common/layout-grid-subtleties`.
+
+Change pipeline, validation, and migrations:
+- Change records, undo/redo architecture, changes-builder API, and production-path mutation guidance: `mem:common/changes-architecture`.
+- Change application, shape-tree edits, validation/repair, migrations, and second-pass touched behavior: `mem:common/file-change-validation-migration-subtleties`.
+
+Components, variants, and debugging:
+- Component/variant data model, ref chains, touched override semantics, and cloning paths: `mem:common/component-data-model`.
+- Component swap, variant switch, and keep-touched pipeline: `mem:common/component-swap-pipeline`.
+- Live inspection snippets, temporary runtime patching, and test-side debugging helpers for common change/component behavior: `mem:common/component-debugging-recipes`.
+
+Text and tests:
+- Shared text data conversion, DraftJS compatibility, modern text content, and derived position data: `mem:common/text-subtleties`.
+- Common test commands, helper conventions, production-path test mutations, and runtime coverage choices: `mem:common/testing`.
+- Cross-cutting testing principles, anti-patterns, and verification checklist: `mem:testing`.
+
+## Areas without focused memories
+
+Common areas with little or no dedicated memory include colors, media/SVG helpers, path operations, thumbnail helpers, generic pools, weak refs, and some utility namespaces. Treat work there as source/test-led unless a focused memory exists.
@@ -0,0 +1,25 @@
+# Common Data Model Change Checklist
+
+## Attribute conventions
+
+- Prefer optional page/shape attrs with default behavior when absent. Reverting to default should usually remove the attr instead of storing nil.
+- Do not treat nil as a distinct persisted state from absence. Import/export and cleanup paths may filter nil attrs away.
+- Avoid Clojure-special naming in exported object attrs, especially boolean names ending in `?`; exported/imported data must survive JSON/SVG/Transit and external tooling.
+- Any new shape attr that participates in component sync must be listed in `app.common.types.component/sync-attrs` with the correct touched group. Attrs absent from `sync-attrs` are ignored by component synchronization.
+
+## Cross-module update checklist
+
+When changing the file data model, check the relevant paths:
+- Schema/type definitions under `common/src/app/common/types*` and helpers under `common/src/app/common/files*` / `logic*`.
+- File migrations in `common/src/app/common/files/migrations.cljc` when old files cannot safely use absence/default behavior.
+- Frontend edit forms under `frontend/src/app/main/ui/workspace/sidebar/options/`; multi-selection behavior is usually in `multiple.cljs` and must handle `:multiple` values.
+- SVG/file render and export metadata under `frontend/src/app/main/ui/shapes/*`, especially `export.cljs` when an attr is not a native SVG property.
+- SVG import/parser paths under `frontend/src/app/worker/import/parser.cljs`; attrs not exported and imported will be lost on reimport.
+- Viewer inspect and code generation under `frontend/src/app/main/ui/viewer/inspect/*` and `frontend/src/app/util/code_gen.cljs` / markup/style helpers when handoff output should expose the attr.
+- Exporter/library consumers when the change affects file construction, rendering, or packaged `.penpot` archives.
+
+## Migrations
+
+Existing files should keep working unchanged when possible. If absence cannot preserve old behavior, add a migration and preserve append/order semantics described in `mem:common/file-change-validation-migration-subtleties`.
+
+Model changes can also require file feature flags or migration metadata updates; check nearby migrations and `common/src/app/common/features.cljc` before inventing a new pattern.
@@ -0,0 +1,54 @@
+# Decimals and Coordinates in Penpot
+
+Penpot stores all geometry as JS numbers (doubles in CLJS, doubles in
+CLJ for the JVM-side common code). Several Penpot-specific facts
+about how this plays out are not obvious from reading the code.
+
+## Sub-pixel drift is routine
+
+Coordinate values that "should" be integers are routinely off by ~1e-5
+in production data. A `:width` of 107 will frequently appear as
+`107.00001275539398` after the value has passed through:
+
+- the modifier propagation pipeline (`apply-wasm-modifiers` and the
+  Rust WASM transform engine)
+- any rotation/scale composition
+- repeated translations
+
+This drift is invisible in the UI (the renderer rounds at draw time)
+but defeats exact equality comparisons in business logic. It does NOT
+appear in JVM-only test setups because the WASM pipeline isn't
+involved — tests that build shapes via `setup-shape` and `add-sample-shape`
+get clean integer values. Bugs that depend on drift will pass tests
+but fire in production unless tests explicitly inject drift.
+
+## Use the close? helpers, not `=`
+
+For comparing coordinate-like floats, the established convention is:
+
+- `app.common.math/close?` — scalar tolerance comparison.
+  Default precision 0.001 (sub-pixel; tight enough to keep distinct
+  shapes distinct, loose enough to absorb arithmetic noise).
+  Two-arity uses default precision; three-arity takes a custom one.
+- `app.common.geom.point/close?` — element-wise close on `gpt/Point`
+  records. Compares :x and :y via `mth/close?`.
+- `app.common.geom.matrix/close?` — close on transform matrices.
+- `app.common.geom.shapes/close-attrs?` — used inside `set-shape-attr`
+  to decide whether a re-assigned `:width`/`:height` should be treated
+  as a no-op (suppresses spurious touched marking from drift).
+
+Treat `=` on `:x`, `:y`, `:width`, `:height`, `:selrect`, or `:points`
+fields as a code smell when the inputs may have flowed through any
+transform. The `set-shape-attr`-style precedent (already using
+`close-attrs?`) is the right model.
+
+## The redundancy multiplies failure modes
+
+A shape's position lives in `:x/:y`, `:selrect`, AND `:points` (see
+`mem:common/geometry-invariants` memory). Each is a separate set of float
+values. After any operation that touches geometry, all three should
+agree, but each is computed by a different path and accumulates
+its own drift. Comparing `:selrect.width` from shape A to
+`:selrect.width` from shape B is comparing two values that
+"semantically" should be equal but were computed via different
+operation chains — exact equality will often be false.
@@ -0,0 +1,28 @@
+# Common File Change, Validation, and Migration Subtleties
+
+## Change application
+
+- `process-changes` validates the whole change vector once by default, reduces changes, then performs a second pass for collected touched changes. Callers that already validated can pass `verify? false`.
+- `process-operation :set` delegates to `ctn/set-shape-attr`; `:assign` first decodes attrs with the shape-attrs JSON transformer and then emits per-attr set operations.
+- `set-shape-attr` treats `:position-data` as derived and never touched. Geometry/content-path changes use approximate equality; geometry differences under about 1px can be ignored for touched purposes.
+- Width/height are excluded from the `is-geometry?` branch in `set-shape-attr`; do not assume all geometry-group attrs follow identical ignore-geometry behavior.
+- `process-touched-change` marks the owning component modified when a touched shape belongs to a main instance; component-data changes can come from shape ops through this second pass.
+
+## Shape tree edits
+
+- `shape-tree/add-shape` falls back invalid/missing parent or frame ids to root (`uuid/zero`), ensures parent `:shapes` is a vector, avoids duplicate child ids, and clears `:remote-synced` on copy parents unless `ignore-touched` is true.
+- `shape-tree/delete-shape` removes the shape and all descendants from the objects map and removes the id from its parent. This is different from render-wasm deletion, which may keep deleted children for undo/redo internals.
+- Page object maps can carry metadata indexes such as cached frame lists. `start-page-index` / `update-page-index` rebuild those metadata indexes; `frontend` commit application calls `ctst/update-object-indices` after page changes.
+
+## Validation and repair
+
+- Full referential/semantic validation currently runs only when file features contain `"components/v2"`.
+- Validation starts at root plus orphan shapes, then validates component records. `validate-file!` raises `:validation :referential-integrity` with collected details.
+- `repair-file` does not mutate data directly; it reduces validation errors into redo changes using `changes-builder`. Callers must apply or persist those changes.
+
+## Migrations
+
+- Prefer optional attrs/default behavior so old files continue working without migration. If absence cannot preserve old behavior, add a migration.
+- Migrations are an ordered set mixing legacy version-derived ids and newer named ids. Keep append order stable; `migrate` applies the set difference between available migrations and file migrations.
+- `migrate-file` synthesizes legacy migration ids from old numeric versions when `:migrations` is absent, migrates legacy features, and records feature flags created through `cfeat/*new*`.
+- When a file had no previous `:migrations`, `migrate-file` marks all migrations as migrated in metadata so callers persist the complete migration set, not only transformations that changed data.
@@ -0,0 +1,48 @@
+# Geometry Invariants in Penpot Shapes
+
+Core invariant: shape position is stored redundantly, and all geometry fields must stay coherent.
+
+## Redundant fields
+
+For a shape at `(x, y)` with width `w` and height `h`:
+
+- `:x`, `:y`, `:width`, `:height`: top-left and dimensions.
+- `:selrect`: `{:x :y :width :height :x1 :y1 :x2 :y2}`, where `x2 = x + w` and `y2 = y + h`.
+- `:points`: four corners for an axis-aligned rect, clockwise from top-left.
+- `:transform` and `:transform-inverse`: identity for axis-aligned shapes; populated for transformed shapes.
+
+After a geometric mutation, equivalent fields such as `:y`, `(:y :selrect)`, and the first point's `:y` should agree. The renderer and hit-testing read `:selrect` / `:points`, so a shape can render or select incorrectly even when `:x` / `:y` look right.
+
+## Helpers that preserve the invariant
+
+- `gsh/move`: translates by delta and updates geometry consistently.
+- `gsh/absolute-move`: moves to an absolute position by computing a delta from the current selrect.
+- `gsh/transform-shape`: applies a full transform.
+- `cts/setup-shape`: initializes geometry for new shapes; variant test helpers such as `thv/add-variant-with-child` use it.
+
+## Edits that break the invariant
+
+- `(assoc shape :x ...)` or `(assoc shape :y ...)`: updates only one field and leaves `:selrect` / `:points` stale.
+- `ths/update-shape file label :y val`: goes through `set-shape-attr`, but does not repair all position fields for `:y` alone.
+- Direct `update-in` edits to `:selrect`, `:points`, or dimensions.
+
+## Test setup warning
+
+When positioning test shapes, use `gsh/absolute-move`, `gsh/move`, or production change helpers. Do not set only `:x` / `:y`.
+
+```clojure
+(cls/generate-update-shapes
+  (pcb/empty-changes nil page-id)
+  #{(:id child)}
+  #(gsh/absolute-move % (gpt/point (:x %) 101))
+  (:objects page)
+  {})
+```
+
+Using `(ths/update-shape file label :y 101)` leaves `:selrect.y` stale. Downstream code that reads `:selrect` can then fail in ways that look like product bugs but are only invalid test setup.
+
+## :touched and geometry mutation
+
+When a copy shape changes geometry through the proper pipeline (`set-shape-attr` via `process-operation :set`), `:touched` gains `:geometry-group` unless ignored. Tests can either drive the production update with `cls/generate-update-shapes`, or inject `(assoc shape :touched #{:geometry-group})` when only touched state matters.
+
+If a test needs both a new position and touched state, move the shape first with geometry-preserving helpers, then inject or assert touched state.
@@ -0,0 +1,13 @@
+# Common Layout and Grid Subtleties
+
+## Layout metadata
+
+- Layout container data and child layout-item data are removed by different helpers. Do not assume clearing a layout frame also clears all child layout metadata.
+- Layout data can affect both container attrs and immediate child attrs; validate behavior for both sides when changing cleanup or propagation.
+
+## Grid assignment
+
+- Grid `assign-cells` ensures at least one column and row, skips absolute-position children, creates non-tracked rows/cols when children exceed tracked cells, and asserts that assigned cells do not overlap.
+- Grid deassignment removes cells for shapes that are no longer direct children or have become absolute-positioned.
+- Auto-positioning is not just sorting: some auto cells are converted to manual when empty/manual/span state would break the auto sequence, then auto single-span items can be compacted.
+- `fix-overlaps` is marked dev-only and removes one overlapping cell, preferring empty cells first. Avoid depending on it as normal production repair.
@@ -0,0 +1,53 @@
+# Common Testing and Verification
+
+`common/` is CLJC shared code. Tests should cover the relevant runtime(s): JVM for backend/common logic and JS for frontend/exporter behavior. For geometry, component, and file-model changes, JVM tests are common and fast, but JS/browser behavior can differ when WASM modifier math or CLJS-specific state is involved.
+
+## Unit tests
+
+READ `mem:testing` FIRST — it defines the execution discipline (no piping, tee to file, preferred commands) that applies to all CLJS/JS and JVM test runs.
+
+Common tests live under `common/test/common_tests/` and use `clojure.test`.
+They are CLJC and run on both JVM and JS.
+
+From `common/`:
+- Full JVM test run: `clojure -M:dev:test`
+- Full JS test run (always builds, suppressed output): `pnpm run test:quiet`
+- Full JS test run (always builds, build output visible): `pnpm run test`
+- Focus a JVM test namespace: `clojure -M:dev:test --focus common-tests.logic.variants-switch-test`
+- Focus a JVM test var: `clojure -M:dev:test --focus common-tests.logic.variants-switch-test/test-basic-switch`
+- Focus a JS test namespace: `pnpm run test:quiet -- --focus common-tests.logic.comp-sync-test`
+- Focus a JS test var: `pnpm run test:quiet -- --focus common-tests.logic.comp-sync-test/test-sync-when-changing-attribute`
+- Quiet logging during a JS run: append `--log-level warn` (or `trace|debug|info|warn|error`)
+- Build JS test target only (no run): `pnpm run build:test`
+- After `build:test` has been run, run the compiled runner directly: `node target/tests/test.js [--focus ...] [--log-level ...]`.
+
+New common JS test namespaces must be required/listed in `common_tests/runner.cljc`;
+new vars in existing namespaces need no runner change. Multiple JVM `--focus` flags
+compose as a union.
+
+## Test helpers
+
+Helpers live under `common/src/app/common/test_helpers/` and are usually aliased with short `th*` prefixes. Test namespaces using label->uuid helpers should start with `(t/use-fixtures :each thi/test-fixture)` so labels reset between tests.
+
+Useful builders:
+- `thf/sample-file` creates a base file.
+- `tho/add-simple-component` creates a simple component.
+- `thc/instantiate-component` instantiates a component copy.
+- `thv/add-variant-with-child` creates a variant container with two child variants.
+- `thv/add-variant-with-copy` creates variants whose children are component instances.
+
+`add-variant-with-copy` does not accept position params for children; use `gsh/absolute-move` after creation if positions matter.
+
+## Driving production paths
+
+For shape mutations, prefer production-path helpers such as `cls/generate-update-shapes` plus `thf/apply-changes`. For component swaps with keep-touched behavior, use `tho/swap-component-in-shape` with `{:keep-touched? true}`.
+
+`thf/apply-changes` validates by default and usually gives the most useful invariant failure. Pass `:validate? false` only for intentionally malformed intermediate state.
+
+## Geometry setup caution
+
+For geometry-sensitive tests, read `mem:common/geometry-invariants` before positioning shapes. Use geometry-preserving helpers or production change helpers rather than direct single-field edits.
+
+## Debugging
+
+Use `mem:common/component-debugging-recipes` for shape-tree dumps, undo/change inspection, and temporary live instrumentation recipes.
@@ -0,0 +1,14 @@
+# Common Text Subtleties
+
+## DraftJS compatibility
+
+- `app.common.text` is legacy DraftJS conversion support. New text work should prefer the newer text type namespaces unless specifically touching DraftJS conversion.
+- DraftJS style values are encoded as Transit strings under `PENPOT$$$<key>$$$<encoded>` style names. `PENPOT_SELECTION` is a special marker.
+- Text conversion uses Unicode code points on both CLJ and CLJS paths, not UTF-16 code units. This matters for offsets around emoji and astral characters.
+- Draft conversion fixes gradient type strings back to keywords.
+
+## Modern text content
+
+- Modern text content schema is narrow: root -> paragraph-set -> paragraph -> text nodes.
+- `position-data` is derived layout/geometry/font fragment data and should be treated as generated state, not source-of-truth file data.
+- Token propagation and some non-current-page text updates drop `:position-data` so it can be regenerated in the right runtime context.
@@ -0,0 +1,17 @@
+# Common Tokens and Schema Subtleties
+
+## Tokens
+
+- `TokensLib` always ensures an internal hidden theme exists and defaults active themes to that hidden theme path. That hidden theme represents the UI state where active sets are controlled without modifying a user-created theme.
+- `get-tokens-in-active-sets` merges tokens from sets selected by active themes in set order, so later active sets with the same token name override earlier ones.
+- Activating a real theme in `common.logic.tokens` removes the hidden theme from the active-theme set unless it is the only active theme. Toggling active sets directly copies current active sets into the hidden theme.
+- DTCG import/export deliberately hides the internal hidden theme: exports omit it from `$themes` and activeThemes, while `activeSets` records the hidden/current active sets.
+- Single-set DTCG/legacy imports throw if no supported tokens are found. Multi-set import normalizes set names, keeps `tokenSetOrder`, rejects conflicting token path names, discards unsupported token types, and validates theme sets against existing sets.
+- Token values stored on shapes are token names in `:applied-tokens`, not token ids. Renames and group renames must update those name paths.
+- Token serialization has both Transit handlers for frontend/backend transport and Fressian handlers for internal file-data storage, with migrations for older token-lib internal versions.
+
+## Schema
+
+- `app.common.schema/json-transformer` has custom map-of key decoding/encoding, so map keys can be transformed based on the key schema instead of only the value schema.
+- `check-fn` throws `ex-info` with default `:type :assertion`, `:code :data-validation`, and `::explain`. Prefer reusable `check-fn`/lazy validators in hot or repeated paths; `sm/check` creates a checker every call.
+- `coercer` decodes with the JSON transformer and then checks. This is the common pattern for accepting external JSON-shaped data into internal types.
@@ -1,32 +0,0 @@
-# Creating Commits
-
-## Message Format
-
-```
-:emoji: Subject line (imperative, capitalized, no period, ≤70 chars)
-
-Body (clear, concise description)
-
-Co-authored-by: <You (the LLM)>
-```
-
-## Commit Type Emojis
-
-`:bug:` bug fix · `:sparkles:` enhancement · `:tada:` new feature · `:recycle:` refactor · `:lipstick:` cosmetic · `:ambulance:` critical fix · `:books:` docs · `:construction:` WIP · `:boom:` breaking · `:wrench:` config · `:zap:` perf · `:whale:` docker · `:paperclip:` other · `:arrow_up:` dep upgrade · `:arrow_down:` dep downgrade · `:fire:` removal · `:globe_with_meridians:` translations · `:rocket:` epic/highlight
-
-## Changelog (CHANGES.md)
-
-Update `CHANGES.md` for user-facing or notable changes. Add entry under the current unreleased version in the matching section (`### :boom:`, `### :sparkles:`, `### :bug:`, etc.).
-
-Entry format:
-```
- Description of change [Taiga #NNNN](https://tree.taiga.io/project/penpot/us/NNNN)
-```
-or for GitHub issues/PRs:
-```
- Description of change [Github #NNNN](https://github.com/penpot/penpot/issues/NNNN)
-```
-
-Changes that affect the JavaScript plugin API must additionally be documented in `plugins/CHANGELOG.md`:
-  * Add an entry at the top of the file (unreleased section)
-  * Prefix entries that change the types/signatures in the API with `**plugin-types:**` and changes affecting the runtime with `**plugin-runtime:**`.
@@ -1,30 +0,0 @@
-# Creating Pull Requests
-
-Important: Before creating a PR, ensure that you are on a branch that is specific to the 
-issue or feature you are working on. If necessary, create a new branch.
-
-## Title Format
-
-PR titles follow the same convention as commit titles:
-
-```
-:emoji: Subject line (imperative, capitalized, no period, ≤70 chars)
-```
-
-See the `creating-commits` memory for the list of emoji codes.
-
-## Description Format
-
-The PR description must start with the following notice:
-
-> **Note:** This PR was created with AI assistance as part of the Penpot MCP self-improvement initiative.
-
- **Related Issues** section with a bullet list of linked issues:
-
-```
-In addition to sections summarising and explaining the changes in the PR, it should contain a section 'Relevant Issues' with a bullet list:
-
- Fixes #NNNN
- Resolves #NNNN
- Relates to #NNNN
-```
@@ -1,27 +1,90 @@
-You are working on the GitHub project penpot/penpot.
+You are working on the GitHub project `penpot/penpot`, a monorepo.

-# Working with Penpot Designs via the JavaScript API
+# Memory system

-Before working with Penpot designs, call the `high_level_overview` tool of the Penpot MCP server.
-It explains the API, which you can use to automate tasks via the `execute_code` tool.
+- Memories are the primary project guidance (not docs or other readme files).
+- A section's top-level memory is `<section>/core`. When a section is relevant, read the core memory
+   before focused memories.
+- Edits/stale refs/duplication cleanup: `mem:memory-maintenance`.
+- Cross-cutting testing principles, TDD workflow, and anti-patterns: `mem:testing`.

-# Dev Workflow
+# Development workflow

-Memories:
-  - before creating a commit, read `creating-commits`.
-  - before creating a PR, read `creating-prs`.
+- Commit/PR/issue creation is **on explicit request only**. Before any of these actions, read the relevant memory — don't infer format from prior examples:
+  - Before `git commit` → `mem:workflow/creating-commits` (subject format, body, `AI-assisted-by: model-name` trailer)
+  - Before `gh issue create` → `mem:workflow/creating-issues` (title derivation, body template, labels, Issue Type)
+  - Before `gh pr create` / `gh pr edit` → `mem:workflow/creating-prs` (title format, body structure, "Note:" line)
+- **Never `git push`, force-push, or modify `git origin`** (or any other remote). The user pushes from their own shell; if a push is required, say so and wait. Never amend a commit that the user has already pushed unless explicitly asked.
+- You have access to the GitHub CLI `gh` or corresponding MCP tools.
+- Issues are also managed on Taiga. Read issues using the `read_taiga_issue` tool.
+- Before writing code, analyze the task in depth and describe your plan. If the task is complex, break it down into atomic steps.
+  *After making changes, run the applicable lint and format checks for the affected module before considering the work done (per example `mem:backend/core` or `mem:frontend/core`).
+- Align `let` binding values: when a `let` form has multiple bindings spanning
+  several lines, align the value forms to the same column with spaces.
+- If you introduce delimiter errors (mismatched parens/brackets) in Clojure/CLJS files,
+  fix them with `scripts/paren-repair` BEFORE running lint/format checks.
+  See `mem:scripts/paren-repair` for usage.
+- Never run anything that destroys data without explicit permission, including `drop-devenv`, `docker compose down -v`, `docker volume rm ...`. The user's real work lives in the volumes of the shared infra.

-# Frontend
+# Project modules

-Read the file `frontend/AGENTS.md` for an overview.
-Memories:
-  - connection between the JavaScript API and the ClojureScript code: `frontend/js-api-to-cljs-binding`.
-  - executing ClojureScript code in the frontend: `frontend/cljs-repl`.
-  - programmatically navigating to a file in the workspace: `frontend/navigation`.
-  - handling Clojure compiler errors, runtime patching and debug helpers: `frontend/handling-errors-and-debugging`.
+This is a monorepo. Principles that apply to one module do *not* generally apply to others. Do not make assumptions.
+
+- `frontend/`: ClojureScript + SCSS SPA/design editor; core conventions: `mem:frontend/core`.
+- `backend/`: JVM Clojure HTTP/RPC server with PostgreSQL, Redis, storage, mail, and workers; core conventions: `mem:backend/core`. Runtime services and the task-queue vs Pub/Sub topology that constrains horizontal scaling: `mem:prod-infra/core`.
+- `common/`: shared CLJC data types, geometry, schemas, file/change logic, and utilities; core conventions: `mem:common/core`.
+- `render-wasm/`: Rust -> WebAssembly Skia renderer consumed by frontend; core conventions: `mem:render-wasm/core`.
+- `exporter/`: ClojureScript/Node headless Playwright SVG/PDF export; core conventions: `mem:exporter/core`.
+- `mcp/`: TypeScript Model Context Protocol integration; core conventions: `mem:mcp/core`.
+- `plugins/`: TypeScript plugin runtime/examples and Plugin API types; core conventions: `mem:plugins/core`.
+- `library/`: design library workflows; core conventions: `mem:library/core`.
+- `docs/`: documentation site; core workflow and conventions: `mem:docs/core`.
+
+The memory is structured in a way that you can get the critical information about the
+module. You can read it from `mem:<MODULE>/core`
+
+# Low-centrality project paths
+
+- `docker/` contains devenv related code, not needed unless specifically instructed.
+   When working on devenv startup, compose layout, instance config (`defaults.env`),
+   tmux session lifecycle, MinIO provisioning, or anything in `manage.sh`'s
+   `*-devenv` commands, read `mem:devenv/core`.
+- `experiments/` contains standalone experimental HTML/JS/scripts; treat it as non-core unless the user explicitly asks about it.
+- `sample_media/` contains sample image/icon media and config used as fixtures/demo material; do not infer app behavior from it.
+
+# Dev Scripts (scripts/)
+
+- `scripts/nrepl-eval.mjs` — Evaluate Clojure/ClojureScript code via nREPL.
+  Supports `--backend` (port 6064) and `--frontend` (port 3447) aliases.
+  See `mem:scripts/nrepl-eval`.
+- `scripts/paren-repair` — Fix mismatched delimiters in Clojure/CLJS files
+  and reformat with cljfmt. Run before lint checks when LLM edits break parens.
+  See `mem:scripts/paren-repair`.
+- `scripts/psql` — PostgreSQL client wrapper with devenv defaults.
+  Companion: `scripts/db-schema` for DDL dumps. See `mem:scripts/psql`.
+- `scripts/taiga.py` — Fetch public issues, user stories, and tasks from the
+  Penpot Taiga project without authentication. See `mem:scripts/taiga`.
+- `scripts/gh.py` — GitHub operations helper: list milestone issues, fetch PR
+  details, compare against CHANGES.md. Requires `gh` CLI. See `mem:scripts/gh`.
+- `scripts/error-reports.mjs` — Query error reports via RPC API with token
+  authentication. Supports list/get operations with filtering and pagination.
+  See `mem:scripts/error-reports`.
+
+# Dependency graph
+
+`frontend -> common`, `backend -> common`, `exporter -> common`, and `frontend -> render-wasm`. Changes in `common` can
+affect frontend, backend, exporter, file migrations, and design-library behavior; validate across consumers when
+semantics change.
+
+# Working with Penpot designs
+
+- Before automating or inspecting Penpot designs through the Plugin API, call the Penpot MCP `high_level_overview` tool.
+- connection between the JavaScript plugin API and the ClojureScript code: `mem:frontend/plugin-api-to-cljs-binding`.
+- executing ClojureScript code in the frontend: `mem:frontend/cljs-repl`.
+- handling Clojure compiler errors, runtime patching and debug helpers: `mem:frontend/handling-errors-and-debugging`.

 ## Detecting Crashes

-The Penpot frontend can crash silently from the JS API's perspective: `execute_code` calls return successfully, but 1-2s later the workspace becomes unusable (Internal Error page). 
+The Penpot frontend can crash silently from the JS API's perspective: `execute_code` calls return successfully, but 1-2s later the workspace becomes unusable (Internal Error page).
 The `execute_code` tool then stops working, but `cljs_repl` still works. Use it to detect a crash via `(some? (:exception @app.main.store/state))`.
-For details on handling crashes, read memory `frontend/handling-crashes`.
+For details on handling crashes, read memory `mem:frontend/handling-crashes`.
@@ -0,0 +1,73 @@
+# Devenv startup and configuration
+
+Compose-based dev environment under `docker/devenv/`, driven by `manage.sh`. Parallel instances share infra + Postgres + MinIO; each instance has its own `main` container, Valkey, source checkout, tmux session.
+
+## Compose project layout
+
+- `penpotdev-infra`: shared `postgres`, `minio`, `minio-setup`, `mailer`, `ldap`. File: `docker-compose.infra.yml`.
+- `penpotdev-wsN` (N=0,1,…): per-instance `main` + `redis` (Valkey). File: `docker-compose.main.yml`. ws0 (a.k.a. `main`) binds `$PWD`; ws1+ bind clones at `${PENPOT_WORKSPACES_DIR}/wsN/` (default `~/.penpot/penpot_workspaces/`), maintained by the developer.
+- All projects join external network `penpot_shared`. Created idempotently by `ensure-devenv-network`, never removed by lifecycle commands.
+
+## Source-of-truth files
+
+- `docker/devenv/defaults.env`: ws0 baseline — container/volume names, runtime env, published host ports, tmux defaults. `manage.sh` aborts if unreadable.
+- For ws1+, `instance-env-overrides` computes the per-instance overrides (container/volume names, host ports offset `10000·N`, `PENPOT_PUBLIC_URI`, `PENPOT_REDIS_URI`, `PENPOT_BACKEND_WORKER=false`) and `instance-compose` injects them as env vars at compose time — never written to disk, recomputed each call so they can't drift. ws0 uses `defaults.env` as-is.
+- `backend/scripts/_env`: backend-internal only — secret keys, `PENPOT_FLAGS` (with `enable-backend-worker` gated on `PENPOT_BACKEND_WORKER`), `JAVA_OPTS`, `setup_minio()`. Never duplicates `defaults.env`.
+- Compose files use pure `${VAR}` substitution; missing var = compose fails.
+
+## Invariants
+
+- `infra-compose` / `instance-compose` wrap `docker compose` with `env -i`, then re-inject what compose needs. Stripping is required because `defaults.env` is sourced into manage.sh's shell at startup (stale values would leak); the ws1+ overrides are deliberately re-injected as shell env vars precisely because Compose gives shell precedence over `--env-file`, so they override the `defaults.env` baseline.
+- Volume names pinned via `name:` (PENPOT_*_VOLUME), decoupled from the compose project name. ws1+ inject distinct per-instance volume names; ws0 keeps the historical `penpotdev_*` physical names so project renames never require data migration.
+- Network aliases (`- main`, `- redis`) are not declared in main.yml. Compose's auto-service-alias still registers `redis` on the shared network, so DNS for `redis` is non-deterministic with multiple instances. Backend uses `PENPOT_REDIS_URI=redis://penpot-devenv-wsN-valkey/0` (container_name) instead.
+- No cross-project `depends_on`. `manage.sh ensure-infra-up` `docker wait`s on the `minio-setup` one-shot.
+- `JAVA_OPTS` in `manage.sh` is shadowed inside the container by `_env`. The `-e JAVA_OPTS=...` flag only matters for processes that don't source `_env`.
+
+## Worker policy
+
+Backend workers run only on ws0. `_env` gates `enable-backend-worker` on `PENPOT_BACKEND_WORKER`; ws1+ inject it as false. ws0 must be running whenever any ws1+ is running, and is the last instance to stop — `run-devenv --agentic --ws N` (N≥1) auto-starts ws0 first; `stop-devenv` refuses to stop ws0 while any ws1+ is up. Workers are pure fire-and-forget: `wrk/submit!` inserts a row into the shared Postgres `task` table and returns; RPC handlers never wait on completion and workers never publish to msgbus. The reason for "ws0 only" is avoiding multi-instance worker races (cron dedup is best-effort across instances, `wrk/submit!` `dedupe` is racy across submitters); details in `mem:prod-infra/core`.
+
+## Port layout
+
+Container-internal ports fixed; host side offset `10000·N`.
+
+| ws0 | ws1 | wsN | container | role |
+|---|---|---|---|---|
+| 3449 | 13449 | 3449+10000·N | 3449 | public HTTPS (Caddy; `/mcp/ws` same-origin) |
+| 3449/udp | 13449/udp | … | 3449/udp | HTTP/3 |
+| 4401 | 14401 | … | 4401 | MCP HTTP stream |
+| 4403 | 14403 | … | 4403 | MCP REPL |
+| 14181 | 24181 | … | 14281 | Serena MCP |
+| 14182 | 24182 | … | 24282 | Serena dashboard |
+
+Everything else (frontend dev, backend API, exporter, storybook, REPLs, plugin dev, MCP inspector/WebSocket) is in-process or same-origin via Caddy/nginx. Infra publishes: mailer 1080, ldap 10389/10636 (singletons, not offset).
+
+## Tmux + MCP routing
+
+`docker/devenv/files/start-tmux.sh` is session-level idempotent. Reads `PENPOT_TMUX_ATTACH`. If the session exists it attaches or exits; otherwise creates 4 base windows (frontend watch / storybook / exporter / backend) plus `mcp` (when `enable-mcp` in `PENPOT_FLAGS`) and `serena` (when `SERENA_ENABLED=true`). `run-devenv --agentic` always sets both env vars. The legacy `run-devenv` alias doesn't, hence its 4-window-only session. To switch from a legacy session to agentic, `stop-devenv` then `run-devenv --agentic` — the conditional windows are only added at session create time.
+
+MCP plugin routing is same-origin: frontend uses `<public-uri>/mcp/ws`, per-instance nginx proxies to MCP port 4401 in-container. For the plugin↔MCP server wiring (how the browser plugin discovers the URL, the in-memory connection registry, why DB-mediated routing isn't needed), see `mem:mcp/core`.
+
+## Workspace orchestration (ws1+)
+
+Workspace directories are user-maintained at `${PENPOT_WORKSPACES_DIR}/wsN`. `run-devenv --agentic --ws i` syncs only when `--sync` is passed, with one exception: if the workspace directory is missing on first use, sync runs implicitly to seed it.
+
+`sync-workspace wsN`:
+1. `assert-clean-git-state` — refuses on `.git/{rebase-apply,rebase-merge,MERGE_HEAD,CHERRY_PICK_HEAD,index.lock}`. No `--sync-force` escape.
+2. `rsync -a --delete $PWD/.git/ $workspace/.git/`.
+3. `git ls-files -z --cached --others --exclude-standard` → `rsync --files-from` (Git is the authority on tracked files; rsync's gitignore filter would drop committed files under gitignored parents like `.clj-kondo/config.edn`).
+4. Initial-only copy of `frontend/resources/public/js/config.js` (gitignored, but agentic mode needs it). After the first sync the workspace's copy belongs to the developer — subsequent syncs leave it alone.
+5. `git switch -C "wsN/<current-branch>"` inside the workspace.
+
+No `--delete` on the working-tree pass: gitignored caches in the workspace survive. Workspace dir + named volumes survive `compose down`.
+
+## CLI surface
+
+- `run-devenv --agentic [--ws main|0|wsN|N] [--sync] [--serena-context CTX]`: bring one instance up. Agentic only — MCP and Serena windows are always created. Default target main. Errors out if the target is already running. `--sync` is rejected on main; on ws1+ it's optional (forced only when the workspace dir does not exist yet). Auto-starts ws0 first when the target is ws1+ and ws0 is not yet up.
+- `stop-devenv [--ws main|0|wsN|N] [--all]`: stop instances. Flags mutually exclusive. `--ws N` (N≥1) stops just that workspace. `--ws 0` or no flag stops ws0 + shared infra, refused while any ws1+ is running. `--all` stops every ws highest-first then ws0, then infra.
+- `run-devenv`: legacy alias, ws0 non-agentic attached.
+- `attach-devenv [--ws main|0|wsN|N]`: pure attach. Fails fast if instance/session missing.
+- `run-devenv-shell [--instance 0|wsN|N] [cmd...]`: bash in target instance. (`--instance` flag not yet renamed to `--ws`.)
+- `start-devenv` / `log-devenv` / `drop-devenv`: legacy paths around ws0 + shared infra. `drop-devenv` never removes volumes.
+
+`exporter/scripts/run` and `wait-and-start.sh` source `backend/scripts/_env` then `_env.local` if present.
@@ -0,0 +1,19 @@
+# Docs
+
+`docs/`: Penpot documentation site; Eleventy.
+
+## Layout and Tooling
+
+- `docs/package.json`: Eleventy commands and documentation-site dependencies.
+- `docs/README.md`: local setup and tooling notes.
+- Diagrams may use PlantUML, svgbob, mermaid, and C4/arc42 conventions depending on the existing page.
+
+## Commands
+
+From `docs/`:
+- Install deps if needed: `pnpm install`.
+- Local server: `pnpm start` or `pnpm run serve` (serves on `http://localhost:8080`).
+- Build: `pnpm run build`.
+- Watch: `pnpm run watch`.
+
+Documentation changes should follow the existing page structure and rendered Help Center conventions rather than inventing a new style locally.
@@ -0,0 +1,34 @@
+# Exporter Architecture and Workflow
+
+`exporter/`: CLJS/Node headless export service. Depends on `common/`; uses Playwright plus export JS/CLJS deps for SVG/PDF/assets.
+
+## Layout and commands
+
+- Source: `exporter/src/`; config: `deps.edn`, `shadow-cljs.edn`, `package.json`; runtime helpers/assets: `vendor/`, `scripts/`.
+- From `exporter/`: setup `./scripts/setup`; watch `pnpm run watch` or `pnpm run watch:app`; production build `pnpm run build`; lint `pnpm run lint`; format check/fix `pnpm run check-fmt` / `pnpm run fmt`.
+- Because exporter consumes `common/`, shared file/shape/model changes may need exporter verification even when the immediate change is not under `exporter/`.
+- Cross-cutting testing principles and anti-patterns: `mem:testing`.
+
+## HTTP and browser pool
+
+- POST body limit is about 60 MB. Exporter supports `application/transit+json`; request params merge query params and body params.
+- Map response bodies are Transit JSON and force HTTP 200; nil 200 bodies become 204.
+- Auth token comes from cookie `auth-token`, then uploads use Bearer auth plus the management shared key.
+- Each export job gets a fresh Playwright browser context. On success, the context closes and the browser returns to the pool; on error, the browser is destroyed instead of reused.
+- Borrow validates browser connection. Pool acquire timeout is about 10s; font loading timeout logs a warning and continues after about 15s.
+
+## Export batching and async behavior
+
+- `prepare-exports` groups entries by `[scale type]` and partitions groups into chunks of 50. Each partition uses file/page/share/name from its first item, so be careful if entries might cross those boundaries.
+- Single-export response is used only when multiple export is not forced and there is exactly one prepared export containing exactly one object.
+- Multi-object export can run async: when `wait` is false it returns a resource immediately and publishes progress/end/error to Redis by profile topic; when `wait` is true it waits for upload and returns the uploaded resource.
+- Frame export returns a resource immediately and publishes Redis updates; it does not follow the same `wait` option path.
+- ZIP entry names are sanitized and duplicates receive numeric suffixes.
+
+## Render details
+
+- Bitmap export differs for WASM vs non-WASM render paths: WASM forces Playwright `deviceScaleFactor` to 1 and passes scale through the render URL; non-WASM uses `deviceScaleFactor = scale`.
+- WebP is produced by taking a PNG screenshot and converting it with ImageMagick.
+- SVG export rasterizes text foreignObjects to PNG, converts through PPM/color masks/potrace, and reassembles SVG paths. It also replaces non-breaking spaces for SVG compatibility and drops empty defs/paths.
+- PDF export injects `@page` sizing through raw browser `evaluate` JavaScript; that code cannot rely on CLJS runtime helpers.
+- Temporary resources schedule local deletion, then uploads POST to `/api/management/methods/upload-tempfile` with `X-Shared-Key: exporter <management-key>` and Bearer auth.
@@ -1,76 +1,94 @@
-# ClojureScript REPL Access via shadow-cljs
+# ClojureScript REPL and Frontend Debugging

-Execute code in the REPL via the Penpot MCP's `cljs_repl` tool.
+Execute code in the live frontend via the Penpot MCP `cljs_repl` tool. For browser-console debugging, the frontend also exports a `debug` JS namespace in development builds.

-## Accessing App State
+## Accessing app state

-The main store is `app.main.store/state`. It contains workspace metadata, selection, UI state, etc.
-However, **page objects are NOT in the main store atom**. They live behind derived refs.
+The main store is `app.main.store/state`. It contains workspace metadata, selection, UI state, profile, route, etc. Page objects are not under a `:workspace-data` key; use derived refs.

-### Top-level store keys (subset)
-`:current-page-id`, `:current-file-id`, `:workspace-local`, `:workspace-global`,
-`:workspace-trimmed-page`, `:workspace-undo`, `:workspace-guides`, `:workspace-layout`,
-`:workspace-presence`, `:workspace-ready`, `:profile`, `:route`, etc.
-
-**Notable absence:** There is no `:workspace-data` key in the store. The old path
-`(get-in state [:workspace-data :pages-index page-id :objects])` does NOT work.
-
-### Getting page objects — use `app.main.refs/workspace-page-objects`
 ```clojure
-;; This is a derived ref (reactive lens). Deref it directly:
+;; Current selection
+(mapv str (get-in @app.main.store/state [:workspace-local :selected]))
+
+;; Current page objects
 (let [objects @app.main.refs/workspace-page-objects
      shape (get objects (parse-uuid "some-uuid-here"))]
  (select-keys shape [:name :type :x :y :width :height :fills :strokes :rotation :opacity :frame-id :parent-id]))
 ```

-### Getting the current selection
+Shape keys use kebab-case keywords. Internal `:rect` corresponds to "rectangle" in the JS Plugin API, and `:frame` corresponds to "board".
+
+Component instance shapes carry `:component-id` and `:component-file` directly; `:component-root` flags the root of an instance. Use `app.common.types.container/get-head-shape` for nearest head and `get-instance-root` for outermost root; they differ for nested instances.
+
+## Navigation recipe
+
+To programmatically open a workspace file, all three ids are required:
+
 ```clojure
-;; Selection is in the main store under :workspace-local :selected
-(let [state @app.main.store/state
-      selected (get-in state [:workspace-local :selected])]
-  (mapv str selected))
-;; Returns vector of UUID strings for selected shapes
+(do (require '[app.main.data.common :as dcm])
+    (app.main.store/emit! (dcm/go-to-workspace
+      :team-id (parse-uuid "<team-id>")
+      :file-id (parse-uuid "<file-id>")
+      :page-id (parse-uuid "<page-id>"))))
 ```

-### Other useful store access
+Get `team-id` from `(:current-team-id @app.main.store/state)`. Get file ids from `(vals (:files @app.main.store/state))`. Get page ids by fetching file data, e.g. through `rp/cmd! :get-file` with current features.
+
+## Reload the live runtime
+
+`(.reload js/location)` (alias `app.util.dom/reload-current-window`) from `cljs_repl` reloads the browser page: clears `set!` runtime patches, re-fetches file state, and is the simplest crash recovery while the repl is live (`mem:frontend/handling-crashes`). To re-fetch only the current file's data without a full page reload, emit `(app.main.store/emit! (potok.v2.core/event :app.main.data.workspace/reload-current-file))`.
+
+## Useful lookup helpers
+
+`app.plugins.utils` contains state lookup helpers that are useful from any CLJS, despite living under `plugins/`:
+
+- `locate-shape`, `locate-objects`, `locate-file`.
+- `locate-component` resolves through the outermost instance root.
+- `locate-head-component` resolves through the nearest component head.
+- `locate-library-component` does direct file-id/component-id lookup.
+
+## Runtime patching with `set!`
+
+Some frontend vars are deliberately mutable escape hatches for runtime instrumentation or circular-dependency patching. From `cljs_repl`, use `set!` for temporary debugging of CLJS vars such as `app.main.store/on-event`, `app.main.errors/reload-file`, `app.main.errors/is-plugin-error?`, `app.main.errors/last-report`, or `app.main.errors/last-exception`. These patches affect only the live browser runtime and disappear on reload or recompilation.
+
 ```clojure
-;; Current page id
-(:current-page-id @app.main.store/state)
-
-;; Verify state is accessible
-(some? @app.main.store/state)  ;; should be true
-
-;; workspace-local keys: :zoom :selected :hide-toolbar :last-selected :vbox
-;;   :highlighted :vport :expanded :selrect :zoom-inverse
+;; Log non-noisy Potok events temporarily.
+(set! app.main.store/on-event
+      (fn [event]
+        (when (potok.v2.core/event? event)
+          (.log js/console (potok.v2.core/repr-event event)))))
 ```

-### Shape data structure (internal ClojureScript representation)
-Shape keys use kebab-case keywords (`:fill-color`, `:fill-opacity`, `:parent-id`, `:frame-id`).
-The shape `:type` is a keyword like `:rect`, `:path`, `:text`, `:ellipse`, `:image`, `:bool`, `:svg-raw`, `:frame`, `:group`.
-Note `:rect` in CLJS corresponds to "rectangle" in the JS Plugin API, and `:frame` corresponds to "board".
+Restore mutable hooks after debugging, or reload the frontend. Use JVM `alter-var-root` only for JVM Clojure; it is not the normal way to patch live CLJS browser vars.

-Component instance shapes additionally carry `:component-id` and `:component-file` directly, and `:component-root` flags the root of an instance. To navigate from a shape to its component, use `app.common.types.container/get-head-shape` (nearest head) or `get-instance-root` (outermost root) — these differ when instances are nested.
+## Browser-console debug namespace

-### Helper utilities (`app.plugins.utils`)
-Despite living under `plugins/`, these are general-purpose lookup helpers usable from any CLJS:
- `locate-shape` — find a shape by file-id, page-id, id
- `locate-objects` — get the object tree for a page
- `locate-component` — resolve the component for a shape (walks to **outermost** instance root, not nearest head — beware when instances are nested)
- `locate-library-component` — direct lookup by file-id and component-id
- `locate-file` — look up a file by id from state
+In development, the JS console exposes `debug` helpers from `frontend/src/debug.cljs`:

-## Notes
- The `:main` build has multiple modules: shared, main, main-workspace, rasterizer, etc.
- `app.main.store/state` is a potok store (wrapping an okulary atom) created via `defonce`
- Use `timeout` to avoid hanging if the browser is disconnected
+```javascript
+debug.set_logging("namespace", "debug");
+debug.dump_state();
+debug.dump_buffer();
+debug.get_state(":workspace-local :selected");
+debug.dump_objects();
+debug.dump_object("Rect-1");
+debug.dump_selected();
+debug.dump_tree(true, true);
+```

-## Troubleshooting
+Visual workspace debug overlays can be toggled with `debug.toggle_debug("bounding-boxes")`, `"group"`, `"events"`, or `"rotation-handler"`; `debug.debug_all()` and `debug.debug_none()` toggle all visual aids.

-`cljs_repl` may not connect to the right runtime when several are attached (e.g. workspace tab + rasterizer). Verify with `(.-title js/document)` — it should show your file name, not "Penpot - Rasterizer".
+For temporary source traces, prefer existing logging (`app.common.logging` / `app.util.logging`) or short-lived `prn`, `app.common.pprint/pprint`, `js/console.log`, or `js-debugger` calls. Remove temporary source instrumentation before committing.

-To list runtimes or target one by client-id, use `npx shadow-cljs clj-eval` from `/home/penpot/penpot/frontend`. It talks to the shadow-cljs JVM process, so unlike `cljs_repl` it has access to `shadow.cljs.devtools.api`:
+## Runtime targeting
+
+`cljs_repl` may connect to the wrong runtime when several are attached, such as workspace plus rasterizer. Verify with `(.-title js/document)`; it should show the workspace file name, not "Penpot - Rasterizer".
+
+To list or target shadow-cljs runtimes, run from `/home/penpot/penpot/frontend`:

 ```bash
 printf '(shadow.cljs.devtools.api/repl-runtimes :main)\n' | timeout 10 npx shadow-cljs clj-eval --stdin
 printf '(shadow.cljs.devtools.api/cljs-eval :main "<cljs-code>" {:client-id 5})\n' | timeout 10 npx shadow-cljs clj-eval --stdin
-```
+```
+
+Use command timeouts so a disconnected browser does not hang the session.
@@ -0,0 +1,22 @@
+# Frontend Compile Diagnostics
+
+Separate from runtime crash recovery.
+
+## First check the shadow-cljs build
+
+Use the Penpot MCP `cljs_compiler_output` tool to inspect the latest shadow-cljs `:main` build status. This is the fastest way to distinguish a bad build from a runtime error in the browser.
+
+Recommended order after CLJ/CLJC/CLJS source edits:
+1. Run `cljs_compiler_output`.
+2. If the compiler reports a Clojure syntax problem, especially unmatched delimiters or a confusing location, run `clj_check_parentheses` on the absolute path of the suspect `.clj`, `.cljc`, or `.cljs` file.
+3. After the build is healthy, use `mem:frontend/cljs-repl`, browser tools, or runtime crash checks for behavior.
+
+## Parentheses checker
+
+`clj_check_parentheses` analyzes one Clojure/ClojureScript source file and reports the area likely responsible for unclosed parentheses/brackets/braces. Use it when compiler output points near EOF, points at a misleading later form, or says delimiter-related syntax errors.
+
+## Hot reload notes
+
+When the frontend shadow-cljs watch process is running, edits to CLJC files in `common/` are normally recompiled into the browser automatically. Do not restart the frontend before checking `cljs_compiler_output`; stale behavior is often a failed build.
+
+For production/minified stack traces, build the production bundle from `frontend/` with `pnpm run build:app`. Output and source maps are generated under `frontend/resources/public/js`; inspect source maps or shadow-cljs reports using build ids from `shadow-cljs.edn`.
@@ -0,0 +1,60 @@
+# Frontend Architecture and Workflow
+
+Frontend: CLJS SPA; React/Rumext; Potok; RxJS; okulary refs; SCSS modules; shared `common/`; JS/TS workspace packages.
+
+## Stable namespace map
+
+- `app.main.ui.*`: Rumext/React UI components for workspace, dashboard, viewer, settings, auth, nitrate, etc.
+- `app.main.data.*`: Potok event handlers and side effects.
+- `app.main.refs`: reactive refs/lenses over store and derived workspace data.
+- `app.main.store`: Potok store and `emit!`.
+- `app.plugins.*` and `app.plugins`: CLJS implementation of Plugin JS API proxies.
+- `app.render_wasm.*`: frontend bridge to Rust/WASM renderer.
+- `app.util.*`: DOM, HTTP, i18n, keyboard, codegen, and general frontend utilities.
+- `frontend/packages/*` and `frontend/text-editor`: JS/TS workspace packages consumed by the app.
+- Nitrate subscription/organization UI and flows live under `app.main.data.nitrate` and `app.main.ui.nitrate*`; backend/API behavior is covered by backend memories, and shared permission rules are in `common/src/app/common/types/nitrate_permissions.cljc`.
+
+
+## Lint and Format
+
+From `frontend/`:
+
+- CLJ/CLJS lint: `pnpm run lint:clj`.
+- JS lint currently no-ops via `pnpm run lint:js`.
+- SCSS lint: `pnpm run lint:scss`.
+- Format checks: `pnpm run check-fmt:clj`, `pnpm run check-fmt:js`, `pnpm run check-fmt:scss`.
+- Format fix: `pnpm run fmt`, or targeted `fmt:clj` / `fmt:js` / `fmt:scss`.
+- Translation formatting after i18n edits: `pnpm run translations`.
+
+**Before linting:** if delimiter errors are suspected (after LLM edits, or
+lint/compiler reports syntax errors), run `scripts/paren-repair` on the
+affected files first. Delimiter errors produce misleading linter output.
+See `mem:scripts/paren-repair`.
+
+## Focused memory routing
+
+UI and packages:
+- App UI components, SCSS modules, style-system boundaries, accessibility, i18n, and render performance: `mem:frontend/ui-conventions-and-style-system`.
+- JS/TS packages, shared UI package, text editor, Storybook, and package builds: `mem:frontend/ui-packages-text-editor-workflow`.
+
+Workspace behavior:
+- Workspace state, commits, persistence, undo, repo calls, and refs: `mem:frontend/workspace-state-persistence-subtleties`.
+- Workspace transforms, modifier previews, WASM modifier integration, and transform commits: `mem:frontend/workspace-transform-subtleties`.
+- Workspace token application/propagation: `mem:frontend/workspace-token-subtleties`; shared token data/schema: `mem:common/tokens-schema-subtleties`.
+
+App shell and product flows:
+- Routing, root app shell, websocket, and global errors: `mem:frontend/routing-app-shell-subtleties`.
+- Dashboard and viewer flows: `mem:frontend/dashboard-viewer-subtleties`.
+- Plugin JS API runtime inside the frontend app: `mem:frontend/plugin-api-to-cljs-binding`.
+
+Diagnostics and validation:
+- Runtime inspection and navigation: `mem:frontend/cljs-repl`.
+- Source-edit compile/hot-reload diagnostics: `mem:frontend/compile-diagnostics`.
+- Runtime crash recovery: `mem:frontend/handling-crashes`.
+- Tests and live verification: `mem:frontend/testing`.
+- Cross-cutting testing principles and anti-patterns: `mem:testing`.
+- Real pointer/keyboard gesture reproduction: `mem:frontend/playwright-gestures`.
+
+## Areas without focused memories
+
+These frontend areas currently have no dedicated Serena memory beyond this architecture entry and nearby source/tests: clipboard, drawing tools, boolean/path operations, interactions/prototyping, color/style asset management, grid-layout editing UI, comments UI, fonts UI, and many dashboard/settings subflows. Treat work there as less memory-covered and inspect source/tests more carefully.
@@ -0,0 +1,16 @@
+# Frontend Dashboard and Viewer Subtleties
+
+## Dashboard
+
+- Dashboard initialization fetches projects and fonts for the team, then listens to websocket messages only for global topic `uuid/zero` or the current profile id.
+- Project fetch replaces each project map completely instead of merging, so fields such as `deleted-at` can disappear cleanly.
+- Dashboard file/project mutations are often optimistic local updates with fire-and-forget RPC watchers. Bulk permanent delete/restore paths use SSE progress and progress notifications.
+- File creation/duplication strips file `:data` before putting file summaries into dashboard state.
+
+## Viewer
+
+- Viewer initialization sets `:current-file-id`, `:current-share-id`, and `:viewer-local`, then fetches the view-only bundle. Comment threads are fetched only for logged-in users.
+- Viewer bundle fetch sends the full supported feature set because anonymous shared viewers may not know team-enabled features.
+- View-only bundles can contain pointer values in `:pages-index` and file data. Viewer resolves those fragments with `:get-file-fragment` before storing the bundle.
+- `bundle-fetched` indexes pages and precomputes viewer frames/all-frames, stores libraries/users/thumbnails/permissions under `:viewer`, then navigates to frame id, query index, or auto-selected frame.
+- Viewer zoom and interaction mode changes update both `:viewer-local` and the `:viewer` route query params.
@@ -1,41 +1,38 @@
-# Handling Penpot Frontend Crashes
+# Frontend Runtime Crash Handling

-When the Penpot frontend crashes, it usually shows the **Internal Error** page (title text "Something bad happened", class `main_ui_static__download-link`).
+## Detect a runtime workspace crash

-A typical error pattern is: Changes go through (JS API, `execute_code`), but about 1-2s later, an `update-file` request hits the backend with the change and gets rejected.
-So be sure to check the status for a crash.
+Runtime crashes usually show the Internal Error page with title text "Something bad happened" and class `main_ui_static__download-link`. A common pattern is: changes go through via JS API / `execute_code`, then 1-2s later an `update-file` request reaches the backend and is rejected.

-After a crash, `execute_code` is unusable (no instances connected), and any data in `storage` is lost, but `cljs_repl` keeps working!
+After a crash, `execute_code` can become unusable because no plugin instances are connected and any data in its `storage` is lost, but `cljs_repl` usually still works.

-## 1. Detect the crash
+Check crash state:

-cljs REPL `(some? (:exception @app.main.store/state))` returns `true` when the Internal Error page is showing, 
-`false` on a healthy workspace (and after a successful reload).
+```clojure
+(some? (:exception @app.main.store/state))
+```

-## 2. Read the cause
+It returns `true` when the Internal Error page is showing and `false` on a healthy workspace or after a successful reload.
+
+## Read the runtime cause

 The exception is stored at `(:exception @app.main.store/state)`. Useful keys:

- `:type`, `:code`, `:status` — error class (e.g. `:validation` / `:referential-integrity` / `400`)
- `:hint`, `:details` — human-readable explanation; `:details` typically contains a vector of validation problems with `:shape-id`, `:page-id`, `:args`, etc.
- `:uri` — the API endpoint that returned the error (e.g. `update-file`)
- `:app.main.errors/instance` — the underlying JS Error object
- `:app.main.errors/trace` — JS stack trace string (only shows the response-handling path, not the dispatch site that produced the bad change)
+- `:type`, `:code`, `:status`: error class, e.g. `:validation`, `:referential-integrity`, `400`.
+- `:hint`, `:details`: human-readable explanation; `:details` often contains validation problems with `:shape-id`, `:page-id`, `:args`, etc.
+- `:uri`: API endpoint that returned the error, e.g. `update-file`.
+- `:app.main.errors/instance`: underlying JS Error object.
+- `:app.main.errors/trace`: JS stack trace string, usually response-handling path rather than the dispatch site that produced the bad change.

-```
+```clojure
 (let [ex (:exception @app.main.store/state)]
  (select-keys ex [:type :code :status :hint :details :uri]))
 ```

-For backend validation errors (`:type :validation`), `:details` is the most informative field — it tells you exactly which shape and which invariant was violated.
+For backend validation errors (`:type :validation`), `:details` is usually the most informative field; it identifies the shape and invariant that failed.

-## 3. Recover and continue testing
+## Recover and continue testing

-Reload steps:
-1. List tabs with `playwright:browser_tabs` (`action: list`) and find the Penpot workspace tab (URL contains `/#/workspace`, title ends in `- Penpot`).
-2. If it isn't the current tab, select it via `playwright:browser_tabs` (`action: select`, `index: <n>`). The selected tab's URL then appears as "Page URL" in the result.
-3. Reload by calling `playwright:browser_navigate` with that same URL.
-4. Confirm recovery: `(some? (:exception @app.main.store/state))` should now return `false`.
+Simplest path when `cljs_repl` is still live (usually true after a crash): reload via repl with `(.reload js/location)` — see `mem:frontend/cljs-repl`. Alternatively via Playwright: find the workspace tab (URL contains `/#/workspace`, title ends `- Penpot`), select it if not current, then `playwright:browser_navigate` to that same URL. Either way, confirm recovery with `(some? (:exception @app.main.store/state))` returning `false`.

-Whether the offending change persists depends on the crash type:
-For **backend-rejected changes** (e.g. `:type :validation`, 4xx on `update-file`), changes are NOT persisted. Reload restores the pre-crash state — safe to retry.
+For backend-rejected changes, such as validation errors on `update-file`, changes are not persisted. Reload restores the pre-crash state, so it is safe to retry after fixing the cause.
@@ -10,6 +10,12 @@ You have access to two tools for finding errors in Clojure source code (which yo
 The latter is needed because syntax errors in parentheses give an uninformative compiler error, and the second
 tool can often find the exact location of such errors.

+When delimiter errors are detected (typically from lint or compiler output),
+fix the affected files with `scripts/paren-repair`. The `clj_check_parentheses`
+MCP tool can also pinpoint the error location when available, but it is not
+required — standard build errors are usually enough.
+See `mem:scripts/paren-repair`.
+
 ## Runtime patching with `set!`

 Some frontend vars are deliberately mutable escape hatches for runtime instrumentation or circular-dependency patching. 
@@ -1,25 +0,0 @@
-# How the Plugin JS API connects to ClojureScript
-
-## Type Definitions
- `plugins/libs/plugin-types/index.d.ts` contains TypeScript type declarations (e.g. `ShapeBase`, `LibraryComponent`).
- These are **type-only** — no runtime code. The actual objects are constructed in ClojureScript.
-
-## Runtime Shape Proxy
- `frontend/src/app/plugins/shape.cljs` builds the JS shape proxy via `obj/reify`.
- Each method/property from the TS interface (e.g. `:component`, `:isComponentRoot`, `:componentHead`) is defined as a keyword entry in the `obj/reify` form, with a ClojureScript function as the implementation.
- The proxy is created by the `shape-proxy` function, which takes `plugin-id`, `file-id`, `page-id`, and shape `id`, and closes over them.
-
-## Library Proxies
- `frontend/src/app/plugins/library.cljs` defines proxies for library types like `LibraryComponentProxy` (via `lib-component-proxy`), also using `obj/reify`.
- The proxy satisfies the `LibraryComponent` TS interface, exposing `.id`, `.name`, `.path`, etc.
-
-## Circular Dependency Resolution
- `shape.cljs` and `library.cljs` have circular dependencies (shapes reference library component proxies and vice versa).
- `shape.cljs` declares forward references as mutable `def nil` vars (e.g. `(def lib-component-proxy nil)`, line 144).
- `frontend/src/app/plugins.cljs` patches them at load time: `(set! shape/lib-component-proxy library/lib-component-proxy)`.
- Same pattern for `lib-typography-proxy?` and `variant-proxy`.
-
-## Key Domain Namespaces
- `app.common.types.component` (aliased `ctk`) — component predicates: `instance-root?`, `instance-head?`, `in-component-copy?`, `is-variant?`
- `app.common.types.container` (aliased `ctn`) — container/tree operations: `in-any-component?`, `get-instance-root`, `get-head-shape`, `inside-component-main?`
- `app.common.types.file` (aliased `ctf`) — file-level operations: `resolve-component`, `get-ref-shape`
@@ -1,14 +0,0 @@
-# Navigating to a File in the Workspace
-
-To programmatically open a file in the workspace, use `cljs_repl` with:
-```clojure
-(do (require '[app.main.data.common :as dcm])
-    (app.main.store/emit! (dcm/go-to-workspace
-      :team-id (parse-uuid "<team-id>")
-      :file-id (parse-uuid "<file-id>")
-      :page-id (parse-uuid "<page-id>"))))
-```
-**All three IDs are required.** You can get:
- `team-id` from `(:current-team-id @app.main.store/state)`
- `file-id` from the dashboard files: `(vals (:files @app.main.store/state))`
- `page-id` by fetching the file: `(get-in file-data [:data :pages])` via `(rp/cmd! :get-file {:id file-id :features (get @app.main.store/state :features)})`
@@ -0,0 +1,117 @@
+# Penpot Canvas → Playwright Viewport Coordinate Mapping
+
+## Goal
+Map Penpot shape coordinates (from the JS/ClojureScript API) to browser viewport CSS pixels
+so that Playwright mouse actions (click, drag, hover) can target specific canvas objects.
+
+## Key Facts
+
+### Playwright coordinate system
+Playwright mouse coordinates are **viewport CSS pixels**: `(0, 0)` is the top-left of the
+browser's rendered content area (not the screen, not the OS window chrome).
+`getBoundingClientRect()` returns the same coordinate system — they are directly compatible.
+
+### Canvas element location
+The Penpot canvas is rendered by two co-located elements:
+- `<canvas>` — the rasterised render
+- `<svg id="render">` — vector overlay
+- `<svg class="...viewport-controls ...">` — interaction/control layer (has the `viewBox`)
+
+Get the canvas origin with:
+```js
+document.querySelector("#render").getBoundingClientRect()
+// => { left: 318, top: 0, width: 514, height: 586, ... }  (values vary with window size/panels)
+```
+The left offset (currently ~318 px) is caused by the left-side panel (layers, assets).
+
+### Zoom and pan state
+Available in two equivalent ways:
+
+**1. App state (ClojureScript):**
+```clojure
+(let [wl (get @app.main.store/state :workspace-local)]
+  {:zoom  (get wl :zoom)    ; scale factor: penpot-units → CSS pixels
+   :vbox  (get wl :vbox)})  ; Rect {:x :y :width :height} — penpot coords of visible area
+```
+
+**2. SVG viewBox attribute (DOM):**
+```js
+document.querySelector("svg.viewport-controls, [class*='viewport-controls']")
+  .getAttribute("viewBox")
+// => "670 658.31 224 255.38"  i.e. "vbox.x vbox.y vbox.width vbox.height"
+```
+Both sources are live and always in sync.
+
+### Coordinate conversion formula
+```
+viewport_x = canvas_left  + (penpot_x - vbox.x) * zoom
+viewport_y = canvas_top   + (penpot_y - vbox.y) * zoom
+```
+
+Sanity check: `vbox.width * zoom ≈ canvas CSS width` (and same for height). ✓
+
+### Device Pixel Ratio
+The canvas physical pixel size = CSS size × DPR (observed DPR = 1.25, so canvas internal
+size 642×732 vs CSS size 514×586). This does **not** affect the formula — both
+`getBoundingClientRect()` and Playwright use CSS pixels.
+
+### Ruler label offset
+The on-screen rulers show coordinates offset from absolute Penpot coordinates (they display
+frame-relative values, offset by ~the top-level frame's x/y). **Ignore for coordinate
+mapping** — use `vbox` directly.
+
+---
+
+## ClojureScript Helper (paste into cljs REPL session)
+
+```clojure
+(defn penpot->viewport-coords
+  "Convert Penpot canvas coordinates to browser viewport CSS pixel coordinates.
+   Returns {:vp-x <number> :vp-y <number>} — pass directly to Playwright mouse actions."
+  [penpot-x penpot-y]
+  (let [state       @app.main.store/state
+        wl          (get state :workspace-local)
+        vbox        (get wl :vbox)
+        zoom        (get wl :zoom)
+        canvas      (js/document.querySelector "svg.viewport-controls, #render")
+        canvas-rect (.getBoundingClientRect canvas)]
+    {:vp-x (+ (.-left canvas-rect) (* (- penpot-x (:x vbox)) zoom))
+     :vp-y (+ (.-top  canvas-rect) (* (- penpot-y (:y vbox)) zoom))}))
+```
+
+Usage example — click the center of a shape:
+```clojure
+(let [shape   (get-in @app.main.store/state [:files file-id :data :pages-index page-id :objects shape-id])
+      cx      (+ (:x shape) (/ (:width shape) 2))
+      cy      (+ (:y shape) (/ (:height shape) 2))
+      {:keys [vp-x vp-y]} (penpot->viewport-coords cx cy)]
+  ;; pass vp-x, vp-y to Playwright page.mouse.click(vp-x, vp-y)
+  {:vp-x vp-x :vp-y vp-y})
+```
+
+---
+
+## JavaScript equivalent (for use in Playwright scripts directly)
+
+```js
+function penpotToViewport(penpotX, penpotY) {
+  // Read viewBox from the controls SVG (always in sync with app state)
+  const svg = document.querySelector('[class*="viewport-controls"]');
+  const [vbX, vbY, vbW, vbH] = svg.getAttribute('viewBox').split(' ').map(Number);
+  const rect = svg.getBoundingClientRect();
+  const zoom = rect.width / vbW;  // == rect.height / vbH
+  return {
+    x: rect.left + (penpotX - vbX) * zoom,
+    y: rect.top  + (penpotY - vbY) * zoom,
+  };
+}
+```
+
+This function can be injected and called via `page.evaluate()` in Playwright:
+```js
+const {x, y} = await page.evaluate(
+  ([px, py]) => penpotToViewport(px, py),
+  [penpotX, penpotY]
+);
+await page.mouse.click(x, y);
+```
@@ -0,0 +1,47 @@
+# Driving Real User Gestures via Playwright
+
+Use Playwright when the bug or behavior depends on Penpot's real input pipeline: pointer gestures, keyboard modifiers, drag/drop targeting, modifier propagation, hover/focus behavior, or alt-drag duplication. The plugin JS API and `penpot:execute_code` can bypass these paths by dispatching store/API operations directly.
+
+## When `execute_code` Is Not Enough
+
+`execute_code` runs in the plugin sandbox. It is excellent for creating shapes, calling Plugin API methods, and querying design data, but it does not faithfully reproduce all user gestures. If the issue involves interactive transforms, frame targeting during drop, drag previews, modifier keys, or canvas hit-testing, drive the browser with Playwright and inspect results via cljs-repl.
+
+## Gesture Pattern
+
+A reliable drag gesture generally needs:
+- focus on the canvas first;
+- key modifiers held from before mouse down until after mouse up;
+- intermediate mouse move events, not just start/end;
+- short waits so Penpot's drag pipeline observes the gesture;
+- a trailing wait for the transaction to commit.
+
+Alt-drag duplication example:
+
+```javascript
+async (page) => {
+  await page.mouse.click(700, 700);
+  await page.waitForTimeout(200);
+
+  const startX = 821, startY = 565, endX = 821, endY = 815;
+  await page.keyboard.down('Alt');
+  await page.mouse.move(startX, startY);
+  await page.waitForTimeout(100);
+  await page.mouse.down();
+  await page.waitForTimeout(100);
+  for (let i = 1; i <= 10; i++) {
+    const t = i / 10;
+    await page.mouse.move(startX + (endX - startX) * t,
+                          startY + (endY - startY) * t);
+    await page.waitForTimeout(20);
+  }
+  await page.waitForTimeout(100);
+  await page.mouse.up();
+  await page.waitForTimeout(100);
+  await page.keyboard.up('Alt');
+  await page.waitForTimeout(500);
+}
+```
+
+## Coordinate Planning
+
+For reliably finding pixel positions of objects, see `mem:frontend/penpot-to-browser-coords`.
@@ -0,0 +1,34 @@
+# Frontend Plugin API Runtime Subtleties
+
+## Type declarations vs runtime
+
+- The Plugin API is a public facade over internal frontend/common data. Do not expect Plugin API property names, value shapes, or behavior boundaries to match internal CLJS attrs or helper APIs; inspect the relevant proxy and internal code path before using Plugin API observations in production internals or tests.
+- `plugins/libs/plugin-types/index.d.ts` contains TypeScript declarations only. Runtime objects are CLJS proxies built under `frontend/src/app/plugins/*.cljs` with `obj/reify`.
+- `shape.cljs` builds shape proxies with hidden ids and per-property CLJS implementations. `library.cljs` builds library proxies such as `LibraryComponentProxy`.
+- `shape.cljs`, `library.cljs`, and related namespaces break circular dependencies with mutable nil vars patched from `app.plugins` at load time. If a proxy constructor appears nil, check the patching path in `frontend/src/app/plugins.cljs`.
+
+## Key Domain Namespaces
+- `app.common.types.component` (aliased `ctk`) — component predicates: `instance-root?`, `instance-head?`, `in-component-copy?`, `is-variant?`
+- `app.common.types.container` (aliased `ctn`) — container/tree operations: `in-any-component?`, `get-instance-root`, `get-head-shape`, `inside-component-main?`
+- `app.common.types.file` (aliased `ctf`) — file-level operations: `resolve-component`, `get-ref-shape`
+
+## Runtime initialization and permissions
+
+- The frontend initializes `@penpot/plugins-runtime` only after `features/initialize` and only when feature `plugins/runtime` is active. It also installs the runtime `isPluginError` predicate into frontend error handling.
+- Manifest parsing expands write permissions to read permissions (`content:write` => `content:read`, etc.). Permission checks also allow the all-zero plugin id and the hard-coded MCP plugin id.
+- Manifest URL origin differs by manifest version: v1 clears the path; v2 joins `.` to the plugin URL. Existing plugin ids are reused by matching manifest name and host.
+- The MCP plugin id is defined in `app.plugins.register` to avoid a circular dependency with workspace MCP code.
+
+## Proxy behavior
+
+- Public Plugin API objects are lightweight handles, not durable snapshots. Most getters locate fresh state from `app.main.store/state` using hidden `$id`, `$file`, `$page`, etc.
+- `not-valid` logs by default but throws when the plugin flag `throwValidationErrors` is enabled. The MCP execute-code handler deliberately enables that flag while running code.
+- `naturalChildOrdering` and `throwValidationErrors` are stored per plugin under `[:plugins :flags plugin-id ...]`; changing default behavior affects automation and MCP diagnostics.
+- Plugin data is stored under keyword namespaces: private data uses `(keyword "plugin" plugin-id)`, shared data uses `(keyword "shared" namespace)`.
+
+## Events and history
+
+- Plugin listeners are watches on the global store and callbacks are debounced about 10ms. Callback exceptions are caught and logged so plugin code does not crash the app.
+- `selectionchange` callbacks receive arrays of shape id strings, while `filechange`, `pagechange`, and `shapechange` return proxies.
+- `contentsave` fires only when persistence status transitions to `:saved`; it calls the callback with no value.
+- Plugin history `undoBlockBegin` creates a workspace undo transaction with a JS `Symbol`; `undoBlockFinish` commits that symbol. Missing finish eventually relies on the workspace transaction timeout.
@@ -0,0 +1,17 @@
+# Frontend Routing, App Shell, Websocket, and Error Subtleties
+
+## Router, app shell, and errors
+
+- Routing uses browser-history hash tokens, but `on-navigate` rejects navigation if the current origin/path does not match `cf/public-uri`.
+- Route params are split into `:path` and `:query`; duplicate query params can become vectors, so use `rt/get-query-param` when a scalar is required.
+- Unknown/empty routes trigger an extra `get-profile`/`get-teams` check before redirecting. This avoids invitation and root-route race conditions.
+- The root app renders an exception page from `:exception` state before the normal error boundary. `rt/navigated` clears `:exception`.
+- Frontend error handling treats stale cross-build JS chunk failures specially: messages containing `$cljs$cst$` or `$cljs$core$I` plus undefined/null/not-a-function signatures trigger throttled reload.
+- Plugin-originated uncaught errors are identified through the plugin runtime hook and logged rather than turning into the global exception page.
+
+## Store and websocket
+
+For general store mechanics such as `emit!`, `last-events`, persistence, and undo, read `mem:frontend/workspace-state-persistence-subtleties`.
+
+- Websocket initialization uses `cf/public-uri` joined with `ws/notifications`, converting `http/https` to `ws/wss`, and includes the current `session-id` as query param.
+- Reinitializing or finalizing websocket stops the previous receive stream. Incoming websocket payloads become Potok data events under `app.main.data.websocket/message`.
@@ -0,0 +1,38 @@
+# Frontend Testing and Live Verification
+
+Frontend validation: CLJS + React/Rumext + RxJS/Potok; SCSS modules; shared CLJC from `common/`.
+
+## Unit tests
+
+READ `mem:testing` FIRST — it defines the execution discipline (no piping, tee to file, preferred commands) that applies to all CLJS/JS test runs.
+
+Frontend unit tests live under `frontend/test/frontend_tests/` and use `cljs.test`. They should be deterministic, avoid DOM/UI integration where possible, and mock side effects such as RPC, storage, timers, or network access.
+
+From `frontend/`:
+- Full unit test run (always builds, suppressed output): `pnpm run test:quiet`.
+- Full unit test run (always builds, build output visible): `pnpm run test`.
+- Focus a frontend CLJS test namespace: `pnpm run test:quiet -- --focus frontend-tests.logic.components-and-tokens`.
+- Focus one frontend CLJS test var: `pnpm run test:quiet -- --focus frontend-tests.logic.components-and-tokens/change-spacing-token-in-main-updates-copy-layout`.
+- Quiet `app.*` logging during a run: append `--log-level warn` (or `trace|debug|info|warn|error`).
+- Build test target only (no run): `pnpm run build:test`.
+- After `build:test` has been run, run the compiled runner directly: `node target/tests/test.js [--focus ...] [--log-level ...]`.
+- Watch tests: `pnpm run watch:test`.
+
+New frontend test namespaces must be required/listed in `frontend_tests/runner.cljs`; new vars in existing namespaces need no runner change.
+
+## Playwright integration tests
+
+Do not add, modify, or run Playwright integration tests under `frontend/playwright` unless explicitly asked. When explicitly asked, use `pnpm run test:e2e` or `pnpm run test:e2e --grep "pattern"` from `frontend/`; ensure dependencies are installed through `./scripts/setup` if the environment is not prepared.
+
+Integration tests fake backend behavior by intercepting network/websocket traffic, so every RPC or websocket the page needs must be mocked. Use existing Page Object Models:
+- `BasePage.mockRPC` intercepts RPC calls and already prefixes `/api/rpc/command/`; pass command names such as `get-profile`, not full URLs.
+- Workspace or other websocket-using pages should extend/use `BaseWebSocketPage`, initialize websocket mocks before each test, and mock `/ws/notifications` with the provided helpers.
+- Prefer common locators/actions in POMs; ad-hoc locators can stay in a single test.
+
+Locator priority should follow user-facing semantics: `getByRole`, `getByLabel`, `getByPlaceholder`, `getByText`, then semantic alternatives such as alt/title, with `getByTestId` as the last resort. Name tests from the user's perspective and prefer positive, single-purpose assertions.
+
+## Live browser verification
+
+Because CLJC compiles to both JVM and CLJS, JVM/common tests can miss frontend-only state caused by browser runtime, WASM modifier math, or real pointer events. Use `mem:frontend/cljs-repl` to inspect live app state and `mem:frontend/playwright-gestures` when real input is needed.
+
+For stale hot reload or failed CLJ/CLJC/CLJS source builds, read `mem:frontend/compile-diagnostics`. For Internal Error pages or delayed runtime crashes after automation/API actions, read `mem:frontend/handling-crashes`. Translation `.po` changes are bundled into `index.html` and require a browser refresh.
@@ -0,0 +1,56 @@
+# Frontend UI Conventions and Style System
+
+## CLJS app UI
+
+- Main app components live under `frontend/src/app/main/ui*` and normally use Rumext `mf/defc` with a `*` suffix for component vars and `[:> component* props]` call sites.
+- Components should have clear ownership. Use `children` for normal composition; use slotted props only when separate owned regions are needed. Do not style or structurally manipulate child DOM that the component did not instantiate.
+- Accept and merge a `class` prop when callers reasonably need layout/positioning customization. Use `mf/spread-props` so Rumext prop transformations such as `:class` -> `className` still apply.
+- Avoid boolean prop names ending in `?`; they do not translate cleanly to JavaScript props. Use type hints such as `^boolean` where JS truthiness/semantics matter.
+- Split large components into smaller private components when useful; `::mf/private true` is the local convention for private Rumext components.
+
+## Styling
+
+- Co-located SCSS modules are preferred. Use `app.main.style/stl` helpers from CLJS and design-system SCSS tokens/mixins instead of legacy global selectors or high-specificity nesting.
+- Keep CSS specificity low. Avoid nested selectors unless they target elements the component owns; CSS Modules already prevent class-name collisions.
+- Prefer CSS logical properties for directional spacing/layout (`padding-inline-start`, etc.). Physical `width`/`height` are still acceptable where they are clearer.
+- Use named design-system variables/tokens for spacing, borders, fixed dimensions, colors, and typography. Avoid hardcoded px/rem values and deprecated `resources/styles/common/refactor/spacing.scss` variables.
+- Use component-local CSS custom properties for variants and theming instead of one-off Sass variables when a component has multiple visual states.
+- Prefer DS typography components (`heading*`, `text*`) and typography mixins instead of plain text wrappers or deprecated typography mixins.
+
+## Accessibility
+
+- Prefer semantic HTML first: anchors for navigation/download/email links, buttons for actions, correct heading levels, and keyboard-focusable controls.
+- If native elements cannot be used, apply appropriate ARIA roles/patterns. Follow WAI-ARIA APG patterns for standard widgets.
+- Icon-only controls need an accessible name via surrounding text, `aria-label`, `alt`, or equivalent. Decorative images/icons should be hidden from assistive tech.
+
+## I18n
+
+- Translations must be resolved during render or render-time memoization, not at namespace load time. For static option lists, memoize inside render so locale changes still update labels.
+- Translation files live in `frontend/translations/*.po`. Translation changes are bundled into `index.html`; refresh the browser after changing translations because there is no hot reload for translation strings.
+- Run `pnpm run translations` from `frontend/` after adding/updating translation text.
+- Adding a new supported locale requires updates in both `frontend/src/app/util/i18n.cljs` (`supported-locales`) and `frontend/scripts/_helpers.js` (`langs`).
+
+## Performance
+
+- Keep expensive derived data in refs, memoized selectors, or pure helpers. In hot render paths, prefer existing `app.common.data.macros` helpers where local code already uses them.
+- Avoid creating new callback functions/objects inside hot renders when a named function, memoized callback, data attribute, or precomputed JS props object works.
+- Destructure props/state values used repeatedly. Avoid repeated deref/property access in render loops.
+
+## Shared React UI package
+
+- `frontend/packages/ui` is the shared React/Vite package. It should remain framework-neutral relative to the CLJS app store; reusable primitives belong here only when they do not depend on Potok/Rumext app state.
+- Package styles are emitted through the package build and copied into `frontend/resources/public/css/ui.css`; stale shared styles are often a build artifact issue.
+- Storybook is the primary visual harness for shared UI/package behavior. Use `mem:frontend/ui-packages-text-editor-workflow` for package build/test commands.
+
+## Choosing a location
+
+- Put editor/dashboard/viewer workflow logic in CLJS app namespaces close to the owning feature.
+- Put reusable presentational React primitives in `frontend/packages/ui` when they can be consumed without Penpot app state.
+- Put CLJS design-system components under `frontend/src/app/main/ui/ds`; new DS components need implementation, CSS module, Storybook story, optional MDX docs, and export from `frontend/src/app/main/ui/ds.cljs` with a JavaScript-friendly name.
+- Put text editing internals in `frontend/text-editor` when the behavior belongs to the JS editor package; use `mem:common/text-subtleties` for shared text data-model behavior.
+
+## Validation
+
+- For CLJS app UI, use `mem:frontend/testing`, `mem:frontend/compile-diagnostics`, and live browser/REPL checks when behavior depends on store or canvas state.
+- For shared UI package changes, run the package build plus Storybook/component tests when relevant.
+- For text editor changes, run `frontend/text-editor` tests and refresh/copy WASM artifacts if render-wasm output is involved.
@@ -0,0 +1,34 @@
+# Frontend UI Packages and Text Editor Workflow
+
+`frontend/packages/`, `frontend/text-editor/`, Storybook/component tests. Separate from CLJS app UI under `frontend/src/app/main/ui`.
+
+## Package boundaries
+
+- `frontend/packages/ui` builds `@penpot/ui`, a React/Vite library package. It exports ESM and type declarations from `dist/`; React and ReactDOM are peer dependencies and must stay external in the Vite library build.
+- The UI package build copies generated `dist/index.css` into `frontend/resources/public/css/ui.css`. If shared UI styles look stale in the app, rebuild the package or check this copy step before debugging CLJS style code.
+- `frontend/text-editor` builds `@penpot/text-editor` from `src/editor/TextEditor.js`. It is a Vite JS package, not CLJS, and has its own Vitest/browser-test setup.
+- The text editor consumes render-wasm artifacts copied from `frontend/resources/public/js` into `frontend/text-editor/src/wasm`. Use `pnpm run wasm:update` after rebuilding `render-wasm` if tests or local dev use stale WASM files.
+- Other packages under `frontend/packages/` such as `tokenscript`, `draft-js`, and `mousetrap` are workspace dependencies used by the frontend app; do not assume their runtime behavior lives in CLJS namespaces.
+
+## Commands
+
+From `frontend/`:
+- Build app-side JS package assets: `pnpm run build:app:libs`.
+- Watch app-side JS package assets: `pnpm run watch:app:libs`.
+- Storybook build: `pnpm run build:storybook`; local Storybook: `pnpm run watch:storybook`.
+- Storybook/component tests: `pnpm run test:storybook`.
+
+From `frontend/packages/ui`:
+- Build library and CSS artifact: `pnpm run build`.
+- Watch library build: `pnpm run watch`.
+
+From `frontend/text-editor`:
+- Local Vite dev: `pnpm run dev`.
+- Tests: `pnpm run test`; coverage: `pnpm run coverage`; browser watch: `pnpm run test:watch:e2e`.
+- Format check: `pnpm run fmt:js`.
+
+## Validation notes
+
+- Frontend root `check-fmt:js` covers stories, Playwright scripts, frontend scripts, and `text-editor/**/*.js`; it does not replace package-specific builds/tests.
+- Changes to shared UI package exports should be validated both in the package build and in the consuming app/Storybook path.
+- Changes that alter text rendering/editing can involve `frontend/text-editor`, `render-wasm`, CLJS text integration, and `mem:common/text-subtleties`; verify the runtime that actually owns the changed behavior.
@@ -0,0 +1,33 @@
+# Frontend Workspace State and Persistence Subtleties
+
+## Store and interaction streams
+
+- `app.main.store/state` is the Potok store; `emit!` always returns nil. Store errors flow through the mutable `on-error` atom.
+- `last-events` keeps a filtered rolling buffer of about 50 event type strings and commit hint origins. It intentionally omits noisy websocket/persistence/pointer events.
+- `ongoing-tasks` controls `window.onbeforeunload`: any non-empty set blocks tab unload.
+- `app.main.streams/wasm-modifiers` and `workspace-selrect` are behavior subjects used for high-frequency interactive preview state that bypasses normal store updates and lenses.
+- Keyboard modifier streams merge a window blur signal so stuck modifier-key state is cleared after focus loss.
+
+## Repo calls
+
+- `app.main.repo/send!` uses GET only when the RPC name starts with `get-`, when all params are query params, or for configured special cases. Only GET requests are retried.
+- GET retry is limited to transient `:network`, `:bad-gateway`, `:service-unavailable`, and `:offline` errors with exponential backoff. Mutations are not retried.
+- A server SSE response is only accepted when the command is configured `:stream?`; otherwise it raises an unexpected-response assertion.
+
+## Commits, undo, persistence
+
+- `commit-changes` refuses to create commits unless `:permissions :can-edit` is true. It captures file revn/vern, selected-before, features, tags, undo group, and translation flag into a `::commit` event.
+- Applying a remote commit first rolls back pending local commits, applies the remote changes, then replays pending local redo changes. Index updates are emitted for undo, remote redo, and replayed redo paths.
+- Local commits are independently consumed by undo, persistence, WASM model updates, thumbnail/library watchers, and text position-data recalculation.
+- Persistence buffers local commits: status becomes pending after about 200ms, commits are flushed after about 3s or `::force-persist`, and buffered commits are merged per file before `:update-file`.
+- Persistence sends revn as the max of the commit revn and locally tracked latest revn; remote commits update that revn tracker.
+- Persistence is skipped in version preview/read-only mode or without edit permission.
+- Undo transactions can stay open only temporarily; timed-out pending transactions are force-committed after about 20s. Undo entries are capped at 50.
+- Undo/redo are ignored while a normal editor/drawing interaction is active, except grid-layout edition handles undo through this path.
+- After local commits and when render-wasm is active, text shapes get derived `:position-data` recomputed in a separate commit tagged `#{:position-data}`; that tag is excluded from the position-data watcher to avoid loops.
+
+## Refs
+
+- `refs/libraries` is explicitly deprecated for performance; prefer derefing `refs/files` and memoizing `select-libraries` in components.
+- `refs/workspace-page-objects` uses `identical?` equality, so preserving object map identity matters for avoiding derived-ref churn.
+- Selected-shapes refs use a small `{objects selected}` wrapper with custom equality before running `process-selected`; avoid bypassing that pattern in hot UI paths.
@@ -0,0 +1,17 @@
+# Frontend Workspace Token Subtleties
+
+## Token refs and visibility
+
+- Workspace token refs intentionally hide the internal hidden theme from theme trees/lists and expose active tokens through `get-tokens-in-active-sets`.
+- Token values stored on shapes are token names under `:applied-tokens`, not token ids. Renames/group renames must update those paths in common token logic.
+
+## Token application
+
+- Token application refuses to run while a text shape is in text-editing mode and shows a warning instead.
+- Applying a token writes token names into shape `:applied-tokens`, resolves active tokens through Style Dictionary or `tokenscript` depending on feature flags, updates concrete shape attrs, and wraps the operation in an undo transaction.
+- Applying composite typography removes atomic typography token attrs; applying atomic typography removes the composite typography token attr.
+- Spacing tokens have a special split path: layout containers receive gap/padding updates, while immediate children of layouts receive margin updates.
+
+## Propagation
+
+- Token propagation resolves active tokens, buffers many `update-shapes` commits, walks the current page first then the remaining pages, clears affected frame/component thumbnails, and drops `:position-data` for text shapes on non-current pages so it can be regenerated.
@@ -0,0 +1,20 @@
+# Frontend Workspace Transform Subtleties
+
+## Preview vs committed transforms
+
+- High-frequency previews use `app.main.streams/wasm-modifiers` and `workspace-selrect` behavior subjects instead of normal store commits; components consume them through refs that wrap plain atoms.
+- `apply-modifiers*` is the lower-level commit path once object/text modifiers are ready. It updates frame guides, frame comment threads, and then emits `update-shapes` with `:reg-objects? true`.
+- Transform commits restrict diff attrs to `transform-attrs` to avoid scanning unrelated shape attrs.
+- Text transforms may carry derived `:position-data`; `assoc-position-data` attaches it while preserving the original text shape context.
+
+## Component-copy touched suppression
+
+- `calculate-ignore-tree` walks modified shapes and descendants to decide per copy-shape `ignore-geometry?`.
+- `check-delta` compares a copy's relative position/rotation to its component root before and after transform. If relative movement is under about 1px and size/rotation are effectively unchanged, geometry touching is suppressed.
+- This logic is why pure translations of component copies can avoid marking every descendant as geometry-touched, while resizes/rotations still propagate touched state.
+
+## WASM bridge details
+
+- WASM modifier updates set plugin/local props with parsed geometry/structure modifiers rather than directly mutating file data.
+- The position-data recomputation watcher ignores commits tagged `:position-data`; keep that tag when adding derived position-data commits.
+- Rotation has separate WASM and non-WASM event paths. Check both when changing rotation modifier semantics.
@@ -0,0 +1,31 @@
+# Library Architecture and Workflow
+
+`library/`: builds `@penpot/library`; JS-facing in-memory Penpot file builder and `.penpot` ZIP exporter. Separate from main app runtime.
+
+## Layout and commands
+
+- Source: `library/src/`; tests: `library/test/`; experimentation/docs: `playground/`, `docs/`; config: `shadow-cljs.edn`, `deps.edn`, `package.json`.
+- From `library/`: build `pnpm run build`; bundle helper `pnpm run build:bundle` or `./scripts/build`; tests `pnpm run test`; watch `pnpm run watch` / `pnpm run watch:test`; lint `pnpm run lint`; format check/fix `pnpm run check-fmt` / `pnpm run fmt`.
+- When changing file-format construction or export behavior in `common/`, consider whether `@penpot/library` should be tested because it constructs Penpot files outside the app UI.
+- Cross-cutting testing principles and anti-patterns: `mem:testing`.
+
+## JS API and builder state
+
+- The JS build context wraps an atom and implements `IDeref`; `getInternalState` exposes the CLJ state converted to JS.
+- Public methods decode JS objects through the JSON transformer before calling common builder functions. Exceptions become JS `BuilderError` objects with enumerable `cause` and an `explain` getter for Malli explain data.
+- `create-build-context` can store an optional `referer`, later written into the export manifest.
+- The builder is stateful: call `addFile` before `addPage`. `addPage` resets the frame/group stack to the root and clears page-local naming state when the page closes.
+- `addBoard` and `addGroup` push onto the parent stack; matching close calls pop it. `closeGroup` requires at least one child and recalculates group geometry. Masked groups use the first child as mask and copy its geometry.
+- `commit-shape` emits `:add-obj` with `:ignore-touched true`, using the current parent, frame, and page from the stack.
+- Layer names are uniqued per current page; duplicate names get generated suffixes.
+- `addBool` converts an existing group into a bool shape and updates style/content/geometry via `:mod-obj` operations rather than adding a new object.
+- Media blobs are stored separately from file-media metadata; `add-file-media` requires a `BlobWrapper`.
+
+## Export package
+
+- `.penpot` ZIPs include `manifest.json`, file/page/shape JSON, components/colors/typographies/tokens, media metadata, and media object blobs.
+- Path/bool shape `:content` is converted to vectors before JSON encoding.
+- File export intentionally includes only selected top-level attrs plus data options; color export removes `:file-id` and drops empty paths.
+- Manifest type is `penpot/export-files`, version 1, generated by `penpot-library/%version%`, with optional referer and file relations.
+- Export generation is sequential and lazy: delayed JSON/blob work is computed only as each zip entry is written, and the progress callback receives `{total,item,path}` after each entry.
+- The library has compatibility defaults for features/migrations in the common builder; do not assume it always exports with the newest app-default migrations/features.
@@ -0,0 +1,96 @@
+# Penpot MCP
+
+This subproject provides an MCP server for Penpot integration.
+The MCP server communicates with a Penpot plugin via WebSockets, allowing
+the MCP server to send tasks to the plugin and receive results,
+enabling advanced AI-driven features in Penpot.
+
+## Tech Stack
+
+- Language: TypeScript
+- Runtime: Node.js
+- Framework: MCP SDK (@modelcontextprotocol/sdk)
+- Build Tool: TypeScript Compiler (tsc) + esbuild
+- Package Manager: pnpm
+
+## General Principles
+
+IMPORTANT: Use an idiomatic, object-oriented style.
+In particular, this implies that, for any non-trivial interfaces, you use interfaces that expect explicitly typed abstractions
+rather than mere functions (i.e. use the strategy pattern, for example).
+
+Comments:
+When describing parameters, methods/functions and classes, you use a precise style, where the initial (elliptical) phrase
+clearly defines *what* it is. Any details then follow in subsequent sentences.
+
+When describing what blocks of code do, you also use an elliptical style and start with a lower-case letter unless
+the comment is a lengthy explanation with at least two sentences (in which case you start with a capital letter, as is
+required for sentences).
+
+## Project Structure (Excerpt)
+
+```
+mcp/
+├── packages/common/           # Shared type definitions
+│   ├── src/
+│   │   ├── index.ts           # exports for shared types
+│   │   └── types.ts           # PluginTaskResult, request/response interfaces
+│   └── package.json           # @penpot-mcp/common package
+├── packages/server/           # MCP server subproject
+│   ├── src/
+│   │   ├── index.ts           # entry point
+│   │   ├── PenpotMcpServer.ts # MCP server implementation (connection handling, tool registration, etc.)
+│   │   ├── Tool.ts            # base class for tools
+│   │   ├── PluginTask.ts      # base class for plugin tasks
+│   │   ├── tasks/             # PluginTask implementations
+│   │   └── tools/             # Tool implementations
+|   ├── data/                  # contains resources, such as API info and prompts
+│   └── package.json           
+├── packages/plugin/           # Penpot plugin subproject
+│   ├── src/
+│   │   ├── main.ts            # handles communication
+│   │   └── plugin.ts          # plugin implementation
+│   └── package.json           # Includes @penpot-mcp/common dependency
+└── prepare-api-docs           # Python project for the generation of API docs
+```
+
+## Key Development Tasks
+
+### Adjusting the Prompts
+
+The system prompt file (aka Penpot High-Level Overview) is located in 
+`packages/server/data/initial_instructions.md`.
+
+### Adding a new Tool
+
+1. Implement the tool class in `packages/server/src/tools/` following the `Tool` interface.
+   IMPORTANT: Do not catch any exceptions in the `executeCore` method. Let them propagate to be handled centrally.
+2. Register the tool in `PenpotMcpServer`.
+
+Tools can be associated with a `PluginTask` that is executed in the plugin.
+Many tools build on `ExecuteCodePluginTask`, as many operations can be reduced to code execution.
+
+### Adding a new PluginTask
+
+1. Implement the input data interface for the task in `packages/common/src/types.ts`.
+2. Implement the `PluginTask` class in `packages/server/src/tasks/`.
+3. Implement the corresponding task handler class in the plugin (`packages/plugin/src/task-handlers/`).
+    * In the success case, call `task.sendSuccess`.
+    * In the failure case, just throw an exception, which will be handled centrally!
+4. Register the task handler in `packages/plugin/src/plugin.ts` in the `taskHandlers` list.
+
+## Dev Tooling
+
+From the `mcp/` directory, run
+
+* `pnpm run build` to test the build of all packages
+* `pnpm run fmt` to apply the auto-formatter
+* Cross-cutting testing principles and anti-patterns: `mem:testing`.
+
+## Devenv plugin/server wiring
+
+In the normal Penpot devenv MCP path, the browser plugin does not discover or route through Postgres. The frontend provides the plugin extension API with `mcp.getServerUrl()`, currently derived from `frontend/src/app/config.cljs` as `penpotMcpServerURI` if set, otherwise `<public-uri>/mcp/ws`. The MCP plugin opens a direct WebSocket to that URL and appends the current MCP access token as a query parameter.
+
+The live plugin connection registry is in-memory inside each MCP server process (`PluginBridge.connectedClients` / `clientsByToken`). The database only stores MCP access tokens and profile props such as `mcp-enabled`; it does not manage which plugin is connected to which MCP server.
+
+For parallel devenvs, prefer same-origin MCP routing: each Penpot instance should expose `/mcp/ws` through its own nginx/Caddy path to the MCP server running inside the same main container. Keep container-internal ports fixed (MCP defaults `4401/4402/4403`, backend/exporter/frontend defaults, etc.) and only offset host-side published ports per instance. If internal ports are offset, hardcoded local proxy config such as `docker/devenv/files/nginx.conf` will misroute unless templated too.
@@ -0,0 +1,33 @@
+# Memory Maintenance
+
+## Discovery Model
+
+- Core principle: progressive discovery through references, building a graph of memories.
+- Initially, agents are provided with the list of all memories (names only).
+- Agents should read `mem:critical-info` as the top-level entry point (graph root).
+  This memory should contain references to other memories covering major project domains.
+  The referenced memories shall, in turn, shall contain references to even more specific memories, and so on.
+  The depth of the graph shall depend on the project complexity.
+- Use topics/folders to group related memories in order to make the content structure explicit.
+  Folders can mirror project structure (e.g. modules like frontend/backend) or topics like debugging, architecture, etc.
+- Memory references must use a mem: prefix inside backticks, e.g. `mem:frontend/core`.
+  The surrounding text should clearly indicate when to read the memory/which content to expect.
+  The text should provide more precise guidance than the memory name alone,
+  i.e. avoid a reference like "frontend debugging and error handling: `mem:frontend/handling-errors-and-debugging` and instead make clear which concrete aspects are covered in the memory.
+- Memories themselves should not contain information about when to read them; this is the responsibility of the referring memory.
+
+## Style
+
+Dense agent notes, not prose docs. Prefer invariants, terse bullets.
+Avoid obvious context, rationale, and examples unless they prevent likely mistakes.
+Keep guidance durable and generalizable, not task-local.
+
+## Add/update threshold
+
+Add or update memories only with stable, non-obvious project conventions that avoid complex rediscovery in the future.
+Do not add: quick-read facts; generic language/framework knowledge; one-off task notes; volatile line-level details; behavior likely to change soon.
+
+## Maintenance Actions
+
+- Renaming memories: References are updated automatically if handled via Serena's memory rename tool.
+- Checking for stale memories (e.g. after deletion): Call `serena memories check` for a report.
@@ -0,0 +1,38 @@
+# Plugins Architecture and Workflow
+
+`plugins/`: standalone TypeScript/pnpm workspace for Plugin API packages and sample plugins. Related to, distinct from, frontend CLJS Plugin API runtime.
+
+## Layout
+
+- `libs/plugin-types`: TypeScript declarations for the public Penpot Plugin API. Type-only package; runtime behavior is implemented elsewhere.
+- `libs/plugins-runtime`: runtime that loads plugins and exposes/generated API behavior to plugin code.
+- `libs/plugins-styles`: reusable styling package for plugins.
+- `apps/*-plugin`: sample/development plugins. `apps/e2e`: plugin e2e tests.
+
+## Dev Workflow
+
+- From `plugins/`: install `pnpm -r install`; runtime dev server `pnpm run start` or `pnpm run start:app:runtime`; sample plugin `pnpm run start:plugin:<name>`; build runtime `pnpm run build:runtime`; build plugins `pnpm run build:plugins`; lint `pnpm run lint`; format `pnpm run format:check` / `pnpm run format`; tests `pnpm run test`; e2e `pnpm run test:e2e`.
+- If a change affects public Plugin API types or runtime, update `plugins/CHANGELOG.md`. Prefix type/signature entries with `**plugin-types:**`; runtime behavior entries with `**plugin-runtime:**`.
+- Cross-cutting testing principles and anti-patterns: `mem:testing`.
+- JS Plugin API behavior inside Penpot app: `mem:frontend/plugin-api-to-cljs-binding`; TS declarations are not runtime code; many API objects are CLJS proxies in `frontend/src/app/plugins/*.cljs`.
+
+## Sandbox and global cleanup
+
+- The runtime uses SES compartments. Public API return values are passed through `ses.safeReturn` before crossing back to plugin code.
+- Plugin `fetch` is sanitized: credentials are omitted and Authorization is blanked. The exposed response only includes ok/status/statusText/url/text/json.
+- Timer callbacks are wrapped to mark plugin-originated errors, and timeout/interval IDs are tracked so plugin close can clear them.
+- Plugin-originated errors are tracked in a WeakMap instead of mutating error objects, because SES can freeze errors.
+- Closing a plugin removes public API keys from the compartment globalThis.
+
+## Lifecycle
+
+- Loading a plugin closes existing non-background plugins and resets the runtime registry. Be careful around `allowBackground` semantics when changing load/close behavior.
+- If sandbox evaluation fails, the runtime marks the error as plugin-originated, closes the plugin, and rethrows.
+- `plugin-manager` removes event listeners, timers, intervals, and modal state on close, and marks the plugin destroyed. Listener callbacks check that flag because Penpot events can fire after close.
+
+## Modal/UI behavior
+
+- Modal URL preparation differs by manifest version: v1 uses query string parameters, v2 puts parameters in the URL hash.
+- `openModal` is idempotent for the same iframe source and avoids reopening when the target URL is already displayed.
+- Modal permissions are derived from manifest permissions (`allow:downloads`, `clipboard:read`, `clipboard:write`).
+- `resizeModal` clamps to at least 200x200 and at most the window minus margins, adjusting transform so the modal remains in the viewport.
@@ -0,0 +1,33 @@
+# Production infrastructure (services Penpot depends on)
+
+Backend (`app.config`, `PENPOT_*` env vars) is parameterized; deployments choose providers.
+
+## Services
+
+- **PostgreSQL**: durable store. Profiles, teams, files, sessions, audit, `storage_object` metadata, the `task` queue, `scheduled_task` cron registry, migrations. File-data also lives here when the file-data backend is `legacy-db`/`db`. One shared DB across all backends.
+- **Redis (Valkey-compatible)**: per-backend message bus and cache. Concrete uses: msgbus Pub/Sub for collaborative-editing broadcasts and team/profile-org notifications fired by RPC handlers (`app.rpc.notifications`, `files_update`, `teams`, `websocket`); file-summary cache gated by `enable-redis-cache`; rate-limit counters; and the dispatcher→runner work hand-off list `penpot.worker.queue:<tenant>:<queue>`. `PENPOT_REDIS_URI`.
+- **Object storage**: backends `:s3` and `:fs`. S3 in prod; devenv uses MinIO. Holds uploaded media, file-data when the file-data backend is `storage`, exports. Backend-side details (resolve, dedup, bucket set, file-data backends): `mem:backend/http-storage-filedata-subtleties`.
+- **SMTP mailer**: invitations, password resets, email verification (sent via the `:sendmail` worker task).
+- **LDAP** (optional auth provider): helpers in `app.auth.*`, gated by `enable-login-with-ldap`.
+
+## Task queue and worker model
+
+Async tasks are enqueued via `wrk/submit!` (`app.worker`), which inserts a row into the shared Postgres `task` table tagged with `queue = "<tenant>:<queue-name>"`. Submission is **fire-and-forget** — RPC handlers never poll, never wait, and workers never publish to msgbus. The only completion signal is the `task` row's `status` / `completed_at` columns, which nothing in `rpc/` reads. Soft-delete RPCs return immediately after marking the top-level row, leaving the cascade and reaping to workers.
+
+Workers run on backends with `enable-backend-worker` in `PENPOT_FLAGS`. Each worker-enabled backend has a `dispatcher` (polls `task` with `FOR UPDATE SKIP LOCKED`, marks status='scheduled', RPUSHes claimed task IDs into **its own** Redis list) and one or more `runner`s per queue (BLPOP from that same local list, execute, update the Postgres row). The Redis hand-off list is purely intra-backend — cross-backend coordination happens at the Postgres row level.
+
+## Cross-backend safety
+
+Postgres row locking is the only correctness primitive: `task` claims via `FOR UPDATE SKIP LOCKED`, cron firing via `FOR UPDATE SKIP LOCKED` on the `scheduled_task` row, plus task-handler-internal locks (e.g. `file_gc_scheduler` locks candidate file rows). This makes the work-claim path safe across any number of worker-enabled backends.
+
+Two known race patterns survive multi-backend operation:
+
+- **Cron dedup is best-effort.** The lock on `scheduled_task` is released when the task body finishes. If two backends' cron timers fire for the same scheduled instant with a gap larger than the task body's runtime, both execute it. Penpot's cron entries are idempotent (`session-gc`, `objects-gc`, `storage-gc-*`, `tasks-gc`, `upload-session-gc`, `file-gc-scheduler`); the exceptions are `:telemetry` (would double-report) and `:audit-log-archive` (depends on archive target idempotency).
+- **`wrk/submit! ::dedupe true`** does a non-atomic `DELETE` then `INSERT`. Concurrent cross-backend submits can both bypass the `DELETE` (each sees the other's uncommitted insert as absent) and end up with duplicate `'new'` rows. Each row claims and runs once independently, so the underlying work is fine; the "at most one pending" guarantee weakens.
+
+Penpot in production lives with both: horizontal-scale deployments accept "exactly-once" as "essentially-once for idempotent operations." Devenv parallel instances handle it by running workers only on ws0 (see `mem:devenv/core`).
+
+## See also
+
+- Devenv composition and the ws0-only worker placement: `mem:devenv/core`.
+- Storage backend resolution, dedup, file-data lifecycle: `mem:backend/http-storage-filedata-subtleties`.
@@ -0,0 +1,33 @@
+# render-wasm Architecture and Workflow
+
+`render-wasm/`: Rust crate compiled to WebAssembly via Emscripten/Skia; frontend loads generated JS/WASM renderer. FFI/memory/tile behavior: `mem:render-wasm/ffi-rendering-subtleties`.
+
+## Stable Architecture
+
+- Exported functions live around `src/main.rs` / `src/wapi.rs` and are called from ClojureScript bridge namespaces under `frontend/src/app/render_wasm*`.
+- Updates are two-phase: ClojureScript calls exported setters to push shape data, then `render_frame()` performs Skia drawing.
+- Rendering is tile-based and shape data is stored separately from hierarchy.
+
+## Source Areas
+
+- `src/state*`: renderer state structures.
+- `src/render/` and `src/render.rs`: tile/surface render pipeline.
+- `src/shapes/` and `src/shapes.rs`: shape data and Skia drawing.
+- `src/wasm/`, `src/wasm.rs`, `src/mem.rs`: JS/WASM memory and interop helpers.
+- `src/math/` and `src/view.rs`: geometry and viewport helpers.
+
+## Build Environment
+
+`./build` sources `_build_env`, which sets the Emscripten paths and `EMCC_CFLAGS`. The WASM heap starts at 256 MB and uses geometric growth.
+
+## Commands
+
+From `render-wasm/`:
+- Build/copy frontend artifacts: `./build`.
+- Watch rebuild: `./watch`.
+- Rust tests: `./test` or `cargo test <name>`.
+- Cross-cutting testing principles and anti-patterns: `mem:testing`.
+- Lint: `./lint`.
+- Format check: `cargo fmt --check`.
+
+Do not change exported WASM function signatures without updating the corresponding frontend bridge and verifying the frontend renderer path.