mirror/browser
1
0
Fork 0
mirror of https://github.com/lightpanda-io/browser.git synced 2026-06-11 01:25:53 -04:00
Code Issues Packages Projects Releases Wiki Activity

docs(browser.tools): clarify extract tool schema and examples

Browse Source
This commit is contained in:
Adrià Arrufat
2026-05-14 10:48:13 +02:00
parent 688924e7a1
commit faba86cf82
2 changed files with 15 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
src/browser/tools.zig
View File

@@ -136,11 +136,18 @@ pub const tool_defs = [_]ToolDef{
\\Extract structured data from the current page using a small JSON schema. Prefer this over `markdown` or `eval` whenever the user asked for a specific value or list (a score, price, count, profile field, headlines, …) — the result is returned as JSON AND the call is recorded as an `EXTRACT` PandaScript line, so a later replay (no LLM) prints the answer to stdout. Use `markdown` / `tree` / `interactiveElements` only to discover the right selector, then commit to one `extract` call.
\\
\\Schema is a JSON object literal (pass it as a string in `schema`). Each value picks what to lift out:
\\ "<sel>" → first match's textContent.trim() (string|null)
\\ ["<sel>"] → every match's text (string[])
\\ {"selector":"<sel>","attr":"<name>"} → first match's attribute (string|null)
\\ [{"selector":"<sel>","fields":{…}}] → array of objects, fields resolved relative to each match
\\Example: schema `{"karma": "#karma"}` → `{"karma":"42"}`.
\\ "<sel>" → first match's textContent.trim() (string|null)
\\ "" → element's own textContent.trim() (only meaningful inside `fields`)
\\ ["<sel>"] → every match's text (string[])
\\ {"selector":"<sel>","attr":"<name>"} → first match's attribute (string|null)
\\ [{"selector":"<sel>","attr":"<name>"}] → every match's attribute (string[])
\\ [{"selector":"<sel>","fields":{…}}] → array of objects, fields resolved relative to each match
\\
\\Examples (schema → result):
\\ {"karma": "#karma"} → {"karma":"42"}
\\ {"items": [".story .title"]} → {"items":["Title 1","Title 2"]}
\\ {"links": [{"selector":"a.title","attr":"href"}]} → {"links":["/a","/b"]}
\\ {"stories": [{"selector":".athing","fields":{"title":".titleline","rank":".rank"}}]} → {"stories":[{"title":"Foo","rank":"1"}]}
,
.input_schema = minify(
\\{
@@ -504,13 +511,7 @@ pub fn evalScript(
/// Schema-driven extraction. The schema is parsed in Zig so a syntax error
/// surfaces here instead of as a confusing V8 SyntaxError on the spliced
/// walker. Each value in the schema object is one of:
/// "sel" → first match's textContent.trim() (string|null)
/// "" → matched element's own textContent.trim()
/// ["sel"] → all matches' textContent (string[])
/// {selector, attr} → first match's attribute (string|null)
/// [{selector, attr}] → all matches' attributes (string[])
/// [{selector, fields}] → all matches, with `fields` relative to each (object[])
/// walker.
pub fn extract(
arena: std.mem.Allocator,
session: *lp.Session,

9
src/script.zig
View File

@@ -109,13 +109,8 @@ pub const mcp_driver_guidance =
\\ prints the value to stdout. Reading the page via `markdown` and
\\ answering only in chat does NOT survive replay.
\\- Use `markdown` / `tree` / `interactiveElements` to *discover* the right
\\ selector, then commit to one `extract` call. Schema examples:
\\ {"karma": "#karma"} → single value
\\ {"items": [".story .title"]} → list of texts
\\ {"links": [{"selector":"a.title","attr":"href"}]} → list of attrs
\\- Pass the schema as a JSON string in the `schema` argument. It must be
\\ a JSON object literal — see the `extract` tool description for
\\ the full schema grammar.
\\ selector, then commit to one `extract` call. See the `extract` tool
\\ description for the schema grammar and examples.
\\
;