refactor(browser): remove model-backed page extraction (#120101)

* refactor(browser): remove model-backed extraction

* docs(browser): remove stale extract reference

docs/tools/browser-control.md

@@ -23,7 +23,7 @@ agent tools, but nothing listens on the loopback control port.
 - Status/start/stop: `GET /`, `GET /doctor`, `POST /start`, `POST /stop`, `POST /reset-profile`
 - Profiles: `GET /profiles`, `POST /profiles/create`, `DELETE /profiles/:name`
 - Tabs: `GET /tabs`, `POST /tabs/open`, `POST /tabs/focus`, `DELETE /tabs/:targetId`, `POST /tabs/action`
-- Snapshot/screenshot/extract: `GET /snapshot`, `POST /screenshot`, `POST /extract`
+- Snapshot/screenshot: `GET /snapshot`, `POST /screenshot`
 - Actions: `POST /navigate`, `POST /act`
 - Hooks: `POST /hooks/file-chooser`, `POST /hooks/dialog`
 - Downloads: `POST /download`, `POST /wait/download`
@@ -49,27 +49,6 @@ For tab endpoints, `targetId` is the compatibility field name. Prefer passing
 handles such as `t1` are also accepted. Raw CDP target ids and unique raw
 target-id prefixes still work, but they are volatile diagnostic handles.
 
-### Page extraction
-
-The agent tool accepts `action="extract"` with required `query` and optional
-`targetId`, `timeoutMs`, `selector`, `ignoreSelectors`, and `schema`. `selector`
-is a CSS selector that limits capture to matching subtrees; a no-match response
-is an error and never falls back to the whole page. `ignoreSelectors` is an
-array of CSS selectors removed from the captured subtree before readable text
-conversion, so navigation, footers, ads, and banners do not consume the model
-context window. The reported `chars` count reflects the scoped, converted text.
-
-`schema` is a JSON Schema object for structured extraction. A successful result
-stores the validated value in `details.json` and shows compact JSON in the
-wrapped text block. Invalid JSON or a schema mismatch gets one correction retry;
-if that also fails, retry without `schema` or adjust the schema. Without
-`schema`, extraction keeps its free-text answer and `NOT_FOUND` behavior.
-
-The CLI mirrors these fields with `--selector <css>`, repeatable
-`--ignore-selector <css>`, and `--schema <json>`. The private `POST /extract`
-capture route accepts `targetId`, `timeoutMs`, `selector`, and
-`ignoreSelectors`; schema validation happens in the calling agent tool or CLI.
-
 If shared-secret gateway auth is configured, browser HTTP routes require auth too:
 
 - `Authorization: Bearer <gateway token>`
@@ -105,7 +84,7 @@ Other runtime failures may still return `{ "error": "<message>" }` without a
 
 ### Playwright requirement
 
-Some features (navigate/act/AI snapshot/role snapshot, extract, element
+Some features (navigate/act/AI snapshot/role snapshot, element
 screenshots, PDF) require Playwright. If Playwright isn't installed, those endpoints return
 a clear 501 error.
 
@@ -128,7 +107,6 @@ What still needs Playwright:
 - AI snapshots that depend on Playwright's native AI snapshot format
 - CSS-selector element screenshots (`--element`)
 - full browser PDF export
-- page-question extraction
 
 Element screenshots also reject `--full-page`; the route returns `fullPage is
 not supported for element screenshots`.
@@ -214,8 +192,6 @@ openclaw browser snapshot --urls
 openclaw browser snapshot --selector "#main" --interactive
 openclaw browser snapshot --frame "iframe#main" --interactive
 openclaw browser snapshot --out snapshot.txt
-openclaw browser extract "What is the page's main conclusion?"
-openclaw browser extract "List the releases" --selector "main" --ignore-selector "nav" --schema '{"type":"array","items":{"type":"object"}}'
 openclaw browser console --level error
 openclaw browser errors --clear
 openclaw browser requests --filter api --clear
@@ -283,12 +259,6 @@ openclaw browser set device "iPhone 14"
 
 Notes:
 
-- Use `browser extract "<question>"` or agent-tool `action="extract"` when you
-  need an answer from the current page but do not need interaction refs. It
-  sanitizes readable page content, caps it at 80,000 characters, runs one
-  model call, and returns only the wrapped answer. The overall timeout defaults
-  to 60 seconds and is clamped to 5–120 seconds. If extraction fails, fall back
-  to `snapshot`; existing-session profiles do not support extraction.
 - The agent-facing `browser` tool exposes `action=download` (required `ref` and
   `path`) and `action=waitfordownload` (optional `path`). Both return the saved
   download URL, suggested filename, and guarded local path. Explicit download

docs/tools/browser.md

@@ -102,16 +102,10 @@ Plugin-bundled skills are listed in the agent's available skills when the
 plugin is enabled. The full skill instructions load on demand, so routine
 turns do not pay the full token cost.
 
-For “read this page and answer X,” use browser `action="extract"` with a
-`query`. It sends sanitized, bounded readable text through one model call and
-returns only the answer; keep `snapshot` for choosing actions and obtaining
-refs. Extraction requires a Playwright-backed profile and falls back to a
-snapshot workflow when it cannot complete.
-
-On large pages, pass `selector` to capture only the relevant CSS subtree and
-`ignoreSelectors` to remove repeated chrome before conversion. Pass a JSON
-`schema` when the caller needs validated machine-usable fields in
-`details.json`; without it, extraction remains a free-text answer.
+For page text, use a selector-scoped snapshot or `act:evaluate` that returns
+only the relevant text or structured data, then let the active agent model
+reason over that bounded result. Use efficient snapshots for controls and
+action discovery; they intentionally omit most non-interactive prose.
 
 ## Missing browser command or tool