A /screenshot without a path had nowhere to go in the REPL: the image
went to the model when one was attached and was refused otherwise.
Probe the terminal once (graphics query followed by Device Attributes,
which every terminal answers and so bounds the reply) and, when it
speaks the kitty protocol, transmit the PNG it already produces, scaled
by the terminal into the window's width and a screenful of rows. The
tool is only asked for an image when the terminal or a model can take
it, so an unsupported terminal keeps the tool's own refusal; when the
model alone took it, the outcome line says so. The PNG is rendered once
and shared between the terminal and the conversation.
The raw-mode and window-size helpers the picker and spinner carried
move to tty.zig so the probe doesn't grow a third copy.
Zig 0.16 forwards the result type through the unary float builtins, so
@trunc produces the integer directly and @intFromFloat is deprecated as
redundant with it. Every site was a truncating cast, or already held an
integral value, so the behaviour is unchanged.
The transport already drops its staging buffer past 256 KB, but the
connection buffer that receives the copy lives for the whole keep-alive
connection and was only ever cleared, so one screenshot pinned its
capacity per open connection. Apply the same threshold there.
Claude-Session: https://claude.ai/code/session_01S9t1TX3vTKunaXdnatbcjB
The renderer knows that layout reflows to the width and that height 0
means the whole content, so the bounding rule moves next to measure()
and only measures when the strip isn't already fixed: the default
viewport shot no longer pays a second layout pass. The limits are
constants spliced into the tool description, not a caller-tunable that
nothing tuned. expireImages runs inside prune so there is one end-of-
turn hook, and re-homes a stripped result instead of casting away const.
An inline screenshot is re-sent on every request for as long as it sits
in history, and a full-page render could reach 1920x16384. Inline images
are now rendered at most 1280 wide and 4096 tall (measured after the
reflow, so short pages aren't padded), files keep full size; the
conversation keeps only the newest two images, older tool results keep
their text with a note; the MCP transport releases its buffer after a
large response.
The slash path's result has two consumers: the terminal, which can't
show an image, and the conversation, which can. Opt in when a model is
attached and forward the image through the same adapter the model-driven
path uses, and stop mapping a failed adapter to a text-only success.
MCP's ImageContent and CallToolResult take their payload type like
TextContent does; resolveScope reuses resolveTarget; needsLocator folds
into replayRequires.
zenai tool results now carry image parts (lightpanda-io/zenai#12, #13),
so the model-driven tool path opts into inline images: a screenshot
without `path` reaches the model as text plus the PNG, on every backend.
The slash-command path still needs `path`, its result goes to the
terminal.
Callers pass CallOpts.inline_image; execScreenshot rejects a path-less
call before navigating or rendering, which removes the per-consumer
guards and covers the model-driven tool path that had none. MCP image
content is a protocol type, the screenshot recording rule joins the
recorder's replayRequires predicate, and the viewport-to-Opts mapping,
node-scope ladder and save-path helpers are shared instead of copied.
The PNG renderer was reachable from CDP and fetch --dump png only. The
tool renders the page or one node; with `path` it writes the file and
returns its location (agent, PandaScript, MCP), without it MCP returns
the image inline as base64 content. ToolResult carries the prepared
image so the transport streams it; the agent and script runtime reject
the inline form since their tool results are text. An inline screenshot
is not recorded, as it has no replayable form.
Terminal.zig references md_term and prompt_assist from a test block, but
nothing referenced Terminal, so those 30-odd tests never ran. Adding it
to Agent.zig's discovery block surfaced one rotted test sink in
js_highlight.zig still using ArrayList.writer(), ported to
Io.Writer.Allocating.
Similar to https://github.com/lightpanda-io/browser/pull/3289.
This pattern:
```zig
errdefer freesomething()
try transfer.submit();
```
is dangerous.`sumbit()` guarantees that the errorCallback is called on any
error. So if the transfer's errorCallback also does `freesomething()` then we
end up with a double-free.
The code was generally cleaned up to be:
```zig
{
errdefer freesomething();
try transfer.addHeader(....);
}
transfer.submit() catch {};
```
Mutating a searchParam that was created from a url.searchParams now keeps the
URL in sync. This helps ensure the that url.search keeps its original form.
Improves a WPT url-searchparams.any.html and urlsearchparams-stringifier along
with a few other cases.
By default, iframes and workers no longer loaded. Use `--load-resources iframe`
and `--load-resources worker` to restore the previous behavior. The disabling
makes resource loading more consistent.
To further make things more consistent, Config seems the following changes:
1. remove `--timeout` from `serve` which does nothing but has printed a
deprecation warning for a long time
2. added .deprecated field to CLI config flags which now logs the specified
deprecated warning when used
3. `--log-filter-scopes` is deprecated in favor of `--log-scopes`
4. `--disable_subframes` is deprecated. Iframe loading is disabled by default,
use `--load-resources iframe` to enable iframe loading
5. `--disable_workers` is deprecated. Worker loading is disabled by default,
use `--load-resources worker` to enable worker loading
6. `--enable_external_stylesheets` is deprecated. Stylesheets remain disabled
by default. Use `--load-resources stylesheet` to enable loading external
stylesheets
CLI log parameters now alter the logger behavior on parse. This helps minimze
the window where default log settings are in-play. It also means things like
this work:
```
./lightpanda --log-level fatal --disable_subframes --log-level warn
```
More seriously, there's now an optional `beforeParse` fired once the mode is
known. This is used by mcp to set the default log level to logfmt. Previously
this was done much later and could easily result in a mix of pretty and logfmt
logs.
A trusted backspace/delete or printable character now:
1 - fires beforeinput and textInput (which can cancel the edit)
2 - change the input's value
Input and TextArea share enough in common and this change has enough selection
logic that a new text_entry.zig helper is added to implement the shared logic.
Review feedback: markdown caps its own output through Opts.max_bytes while
the html dump left it to each caller, so the html tool and fetch each
carried a copy of the LimitedWriter wrapping. root and deep now do it,
with a no-cap fast path so innerHTML/outerHTML are unaffected.
A wait failure on one URL aborted the whole multi-URL fetch with no
output, the wait_ms budget was computed once before the loop so N pages
could take N times the budget, navigation failures exited 0 with an empty
dump, and HTTP 4xx/5xx were indistinguishable from success by exit code.
Each page now has its own error slot: wait, navigation and dump failures
are recorded there instead of returned, every page is still written (the
JSON envelope carries the name under "error"), then the first failure is
returned so the process exits 1. The remaining budget is recomputed per
page. --fail-on-http-error turns a status >= 400 into exit 22, curl's
code for the same condition; the dump is written first either way.
--strip-mode was silently ignored for --dump markdown, and fetch had no
way to scope a dump to one element or cap its size although the markdown
and html tools have both.
--selector dumps the first matching element in any dump mode, --max-bytes
caps html and markdown with the tools' truncation marker, and markdown
now honors strip ui (images; scripts, styles and hidden elements are
never rendered, so the other groups don't apply).
The links tool returned every a[href] verbatim: hidden nav entries,
one row per duplicate href, and null text for image and icon links even
though markdown already knows to use alt. Nav-heavy pages produced
thousands of rows for an agent looking for one link.
One entry per resolved href (the first, upgraded with text from a later
duplicate), hidden anchors skipped with the same check tree uses, text
falling back to aria-label, title, then a descendant img alt. The tool
gains an optional limit.
Adrià Arrufat