Agents often know what fields a target request's body should contain
but not which captured request carries it. --filter lets them declare
the field set and get back only matching entries.
Matching is "any-segment": a field matches when it equals any segment
name of any inferShape() path (ignoring root $, array indices, and
bracket-quoted key syntax). Multiple fields AND together. Case-sensitive.
- invalid_filter for empty / commas-only values
- invalid_args when combined with --detail (mutually exclusive)
- 0 matches is a valid empty result, not an error
- persisted cache stays unfiltered so later --detail lookups still resolve
Envelope gains `filter` (echo) and `filter_dropped` (count of entries
passing the static-resource filter but not --filter). Existing --raw
and --all compose normally.
* feat(browser): remove silent html truncation, add --as json tree output
`browser get html` had two agent-hostile defaults:
1. A silent 50000-char cap on the returned HTML — agents that got a
truncated page had no signal they were looking at half the DOM.
2. Only raw HTML string output, forcing agents to re-parse for
structured extraction.
Changes:
- Default output is now the full outerHTML, no truncation
- `--max <n>` opts in to a character cap; when the cap actually
trips, the HTML is prepended with
`<!-- opencli: truncated N of M chars; re-run without --max ... -->`
so agents always see the signal
- `--as json` returns `{selector, matched, tree}` where `tree` is
`{tag, attrs, text, children}` recursively. `matched` is the full
count of selector matches so agents know when more elements exist
beyond the first. `text` is the node's own direct text children,
whitespace-collapsed; child elements live in `children`.
- `--selector` not matching any element now emits structured
`{error:{code:"selector_not_found", ...}}` with a non-zero exit
code, in both raw and json modes (was `(empty)` stdout previously,
indistinguishable from empty element)
- Invalid `--as` / negative `--max` emit structured
`invalid_format` / `invalid_max` error codes
Extracted the tree serializer as `src/browser/html-tree.ts` so the
JS expression can be unit-tested against a DOM stub.
* fix(browser get html): structured errors for invalid selector & strict --max
Both edges previously bypassed the structured-error contract introduced in
#1102, which agents rely on for branching:
- Invalid CSS selector: querySelector(All) would throw SyntaxError through
page.evaluate into the generic exception path. Wrap the lookup in try/catch
inside page context for both raw and --as json paths; surface as
{error:{code:"invalid_selector", message}} + non-zero exit.
- --max validation: parseInt silently accepted "1.5" -> 1 and "10abc" -> 10.
Switch to a strict /^\\d+$/ check so fractional, negative, and non-numeric
values all return {error:{code:"invalid_max"}}; validation runs up front so
bad values never reach the page.
Covered by new unit tests in cli.test.ts (fractional, non-numeric, invalid
selector on raw + json) and html-tree.test.ts (SyntaxError -> invalidSelector
envelope).
Co-authored-by: freemandealer <freeman.zhang1992@gmail.com>
---------
Co-authored-by: freemandealer <freeman.zhang1992@gmail.com>
* feat(browser): rewrite network command for agent-native discovery
Replace the index-based list + pretty-printed --detail flow with a
structured JSON interface built around stable keys, body-shape previews,
and a persistent capture cache. Agents can now reference captured
requests by operationName (GraphQL) or `METHOD host+pathname` (REST)
instead of array indexes that shift on every rerun.
- `browser network` now emits JSON: `{workspace, captured_at, count,
filtered_out, entries: [{key, method, status, url, ct, size, shape}],
detail_hint}` — no body payloads by default
- Shape inference (src/browser/shape.ts) walks response JSON into a
flat path -> descriptor map with depth cap 6 and a 2KB budget per
entry, so agents see structure without paying body tokens
- Stable key generator (src/browser/network-key.ts) derives
`operationName` from graphql URLs and `METHOD host+pathname`
elsewhere, disambiguating collisions with `#N` suffixes
- Persistent cache (src/browser/network-cache.ts) snapshots every
capture to `~/.opencli/cache/browser-network/<workspace>.json` with
a 24h TTL, so `--detail <key>` survives later commands
- `--detail <key>` returns `{key, url, method, status, ct, size, shape,
body}` with structured error codes (cache_missing / cache_expired /
cache_corrupt / key_not_found, the latter including available_keys)
- Add `--raw` for agents that want every full body inline, `--ttl` for
cache lookups
- Update opencli-adapter-author + opencli-autofix skill docs to
reference `--detail <key>` and the shape-first discovery flow
Supersedes the cache prototype in #1051.
Co-authored-by: freemandealer <freeman.zhang1992@gmail.com>
* fix(browser): structured errors for capture/save, shape budget guard
Self-review findings on the network refactor:
- captureNetworkItems throwing (browser crashed / CDP dropped) now emits
`error.code: capture_failed` on stdout rather than leaking a bare
stderr line from browserAction's generic handler — agents get a
parseable JSON blob on every failure path, matching the design goal.
- saveNetworkCache throwing (disk full, read-only path) is a soft
failure: the captured data is already in hand, so surface a
`cache_warning` field in the envelope and keep going instead of
aborting. `--detail` lookups on that run will miss the cache but the
listing still reaches the agent.
- shape.ts: guard the sub-walk on `add()`'s return value so the
"budget hits on the array/object descriptor itself" path can never
emit a stray child without its parent marker.
- network-key.ts: document that `#N` suffixes start at `#2` — the first
occurrence stays bare, there is no `#1`. Matches test + code.
Added regression tests: `capture_failed` on readNetworkCapture throw,
`cache_warning` on persistence failure, shape budget hit on array descriptor.
Co-authored-by: freemandealer <freeman.zhang1992@gmail.com>
---------
Co-authored-by: freemandealer <freeman.zhang1992@gmail.com>
* feat(twitter): add tweets command for fetching a user's recent posts
Adds `opencli twitter tweets <username> [--limit N]` to pull a user's
most recent chronological tweets via the UserTweets GraphQL endpoint.
Long posts resolve via note_tweet, pinned entries are skipped, and
retweets are flagged. QueryIds resolve dynamically through
`resolveTwitterQueryId` with hardcoded fallbacks.
* fix(twitter): expose retweet flag in tweets output
* refactor: consolidate 6 skills into 3, remove mechanical commands
Replaces opencli-oneshot / opencli-explorer / opencli-browser /
opencli-usage with a single opencli-adapter-author skill that takes
the AI agent end-to-end: site recon, API discovery, field decoding,
adapter coding, and `opencli browser verify`.
Removes the mechanical commands (`explore`, `synthesize`, `generate`,
`cascade`, `record`) and their src/tests — they were codegen scaffolding
meant for agents, which the new skill handles more flexibly via
`opencli browser` primitives.
Skill highlights:
- Top-level decision tree + 12-step runbook
- 5 site patterns (SPA / SSR / JSONP / Token / Streaming)
- 5-layer API discovery (network → initial state → bundle → token → interceptor)
- Field decode playbook (self-explanatory → codes → sort-key comparison)
- Output design guide (columns, types, order, ≤15 per adapter)
- Two-layer site memory: in-repo seeds for eastmoney/xueqiu/bilibili/tonghuashun
plus local `~/.opencli/sites/<site>/` runtime workspace
Kept skills: opencli-autofix (now points to adapter-author for rewrites),
smart-search. Kept primitives: `browser *`, `doctor`, `list`, `validate`,
`verify`, `<site> <cmd>`, `plugin *`, `completion`.
No backward compatibility shims. Full test suite (1605 tests) passes.
* review fixes: honest coverage, hard memory-hit path, typo, stale docs
- site-memory hit path no longer jumps to writing adapter; forces Step 5
endpoint re-verification + Step 7 field check, and 30-day expiry
- site-memory.md now specifies exact schemas for endpoints.json /
field-map.json / notes.md / fixtures + write-back timing rules
- coverage-matrix.md marks unverified patterns as 🟡 with an evidence
section citing coingecko dry run + PR #1091 eastmoney + bilibili
- eastmoney seed typo: resolveSecids -> resolveSecid (and splitSymbols)
- docs/developer/ai-workflow.md rewritten to teach the adapter-author
skill + opencli browser * primitives (dropped generate/synthesize/
cascade/explore references)
- ts-adapter.md, getting-started.md, CHANGELOG.md:87 updated to point
at opencli-adapter-author
* fix(ci): resync package-lock + drop stale built-in list reference
- Regenerate package-lock.json to restore @emnapi/core + @emnapi/runtime
entries that got dropped during the rebase — `npm ci` was failing on all
CI jobs (build / audit / docs-build / bun-test / unit-test)
- docs/guide/getting-started.md: built-in list dropped `explore`, now
reads (list, validate, verify, browser, doctor, plugin...)
* fix(ci): restore package-lock.json from main (unrelated lockfile churn)
Mirror the remaining 13 read-oriented adapters and the shared _secid.js
helper from the author's local workspace into the repo, so that
clis/eastmoney/ becomes the full Phase A codegen regression oracle
described in OpenCLI Improvement Spec v1.1 §B.10.
Total repo oracle after this PR: 14 adapters under clis/eastmoney/
(hot-rank.js already exists; this PR adds the other 13) plus the
_secid.js normalize helper.
Covers the two schema-expressiveness gaps discovered during prep:
- CSV row_format: kline.js decodes "YYYYMMDD,open,close,..." strings
- :row_index source: convertible.js derives rank = i + 1
_secid.js is the canonical example of the v1.1 §B.7 helper contract
(pure normalize/derive function, serializable I/O, no env/fs/net/session
access, does not drive pagination/retry/fallback).
This PR is oracle-only, carries no framework changes. Phase A framework
PR depends on this merging first so the codegen diff target is stable.
Refs: task #177 / spec v1.1 §B.10
* feat(download): show saved file path in web read and weixin download output
Closes#1038
* test(download): cover saved article path
---------
Co-authored-by: Benjamin Liu <beneecs@Benjamins-Mac-mini.local>
Co-authored-by: jackwener <jakevingoo@gmail.com>
* feat(browser): add cross-origin iframe support via CDP execution contexts
Enable interaction with cross-origin iframes through CDP's execution
context mechanism, without requiring content scripts or all_frames.
- Track frame execution contexts via Runtime.executionContextCreated events
- Add 'frames' action to list all child frames (including cross-origin)
- Support frameIndex in 'exec' action to evaluate JS in specific frames
- Add Page.frames() and Page.evaluateInFrame() APIs for CLI consumers
- Tag cross-origin iframes with [F0]/[F1] indices in DOM snapshots
- Add Page.getFrameTree to CDP allowlist
Closes#1077
Change-Id: Id03361ddb616912dff3bfa8e59e8b68716de590b
* fix(browser): align cross-origin iframe routing contract
* fix(browser): unify iframe frame-index routing
---------
Co-authored-by: xuezhangying <xuezhangying@bytedance.com>
Co-authored-by: jackwener <jakevingoo@gmail.com>
From first principles, `opencli browser` commands exist for AI Agents to
operate websites through the browser skill. Reframe both READMEs to reflect
this: show users how to install the skill into their AI agent and describe
tasks in natural language, rather than listing raw CLI commands.
* feat(twitter): rewrite lists via GraphQL + add list-tweets
The DOM-scraping / detail-click approach in PR #1053 remained fragile
against X's frequent overview-page rendering changes and slow (N+1 page
loads per list). Rewrite `twitter lists` to call
`ListsManagementPageTimeline` GraphQL directly — one request returns all
owned + subscribed lists with id/name/member_count/subscriber_count/mode.
Also add `twitter list-tweets <listId>` for pulling the tweet stream from
a list, completing the read-side chain (lists → pick an id → list-tweets).
- lists: drop positional `user` arg (GraphQL returns only logged-in
user's lists), add `id` column, change followers to exact integer from
subscriber_count.
- list-tweets: same GraphQL pattern as bookmarks/likes (BEARER + ct0 +
dynamic queryId with static fallback + cursor pagination).
- Delete obsolete lists-parser.js and lists.d.ts.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(twitter): add list-add / list-remove with Save-button commit
Two new commands to toggle list membership. X's list dialog uses a
"click-to-stage, Save-to-commit" pattern — clicking a row only updates
optimistic UI; the actual POST fires when the user clicks the top-right
"Save" button. Pressing ESC or the close-X silently cancels the change.
Implementation:
- Resolve listId → name via ListsManagementPageTimeline GraphQL, so we
match the dialog row by name (dialog rows have no data-testid listId).
- Open profile page → DOM click "…" menu → "Add/remove from Lists".
- Scroll dialog to locate target row (virtualized list).
- page.nativeClick on row — trusted CDP Input.dispatchMouseEvent fires
React's onclick, flips aria-checked (.click() alone does not suffice;
X ignores non-trusted events for list mutations).
- page.nativeClick on the Save button — commits to server.
- Verify by re-fetching ListsManagementPageTimeline and diffing
member_count: success only if N→N±1. No silent successes.
This fixes the pattern where batch `list-add` calls returned success for
every user but committed zero to the server (optimistic UI lied).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix: stabilize twitter list manifest and query ids
* docs: add twitter list command discoverability
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: jackwener <jakevingoo@gmail.com>
* fix(zsxq): separate content field from title, remove title truncation
- Split getTopicText to return only title, add getTopicContent for body text
- Remove .slice(0, 120) that was truncating titles
- content field now contains full body text instead of duplicating title
* fix(zsxq): preserve title fallback for body-only topics
---------
Co-authored-by: huzekang <huzekang@opencode.ai>
Co-authored-by: jackwener <jakevingoo@gmail.com>
PR #1059 migrated xiaoyuzhou from SSR scraping to authenticated API.
The E2E tests run without credentials, producing exit code 78
(CONFIG_ERROR). The existing `isExpectedChineseSiteRestriction` guard
only caught FETCH_ERROR, PARSE_ERROR, and NOT_FOUND — not config
errors from missing auth credentials.
Three improvements from the design debt audit:
1. Remove deprecated `tabId` field and `getActiveTabId()` method
- Delete `tabId` from DaemonCommand (daemon-client.ts) and Command (protocol.ts)
- Delete `getActiveTabId()` from IPage interface (types.ts) and Page class (page.ts)
- Update extension resolveCommandTabId() to remove legacy fallback
- Update handleTabs select case to remove tabId check
- The tab→page migration is now complete
2. Unify argument validation into single code path
- Remove `normalizeArgValue()` from commanderAdapter.ts
- Commander adapter now passes raw values to prepareCommandArgs()
- All coercion (bool, int, number) and validation (required, choices)
happens once in coerceAndValidateArgs() in execution.ts
- Eliminates duplicated boolean normalization
3. Remove dead plugin filesystem wrappers
- Delete `promoteDir()` — never called in production code
- Delete `replaceDir()` — thin wrapper over beginReplaceDir, never called
- Remove corresponding test-only exports and tests
- Rename PromoteDirFsOps → ReplaceDirFsOps to match remaining usage
- Transaction infrastructure (runTransaction, beginReplaceDir,
beginReplaceSymlink) retained — used by publishStandalonePlugin
and publishMonorepoPlugins for atomic multi-step operations
* fix(extension): per-workspace idle timeout for browser sessions (#1058)
The global 30s WINDOW_IDLE_TIMEOUT was too aggressive for interactive
`opencli browser` commands where users type manually between invocations.
- browser:*/operate:* workspaces now default to 10 min idle timeout
- Adapter workspaces keep the existing 30s timeout
- Support custom timeout via OPENCLI_BROWSER_TIMEOUT env var (seconds)
or command-level idleTimeout parameter
- Surface sessionExpired warning when a new window is created after
the previous session timed out
- Fix stale comment (said 120s, actual was 30s)
Closes#1058
* fix: resolve sessionExpired double-delete race and timeout override lifecycle
Addresses @codex-coder review blockers:
1. sessionExpired flag was never set because getAutomationWindow()
consumed expiredWorkspaces before handleCommand() could check it.
Fix: use .has() in getAutomationWindow, only .delete() in handleCommand.
2. workspaceTimeoutOverrides was never cleaned up — once set, it
persisted until extension restart. Fix: clear override on idle
timeout expiry, explicit close-window, and borrowed-session detach.
Adds 5 tests covering:
- browser:* uses 10min timeout (not 30s)
- sessionExpired flag is set and consumed correctly
- workspaceTimeoutOverrides cleared on idle expiry
- workspaceTimeoutOverrides cleared on explicit close
- idleTimeout from command applies to workspace override
* refactor: remove sessionExpired warning per product decision
@WAWQAQ decided session-expired warning is not needed.
Remove expiredWorkspaces tracking, sessionExpired flag from protocol,
and related CLI-side warning code. Keep per-workspace timeout and
override lifecycle cleanup.
* fix: clean up workspaceTimeoutOverrides on user-initiated window close
The windows.onRemoved listener was missing workspaceTimeoutOverrides
cleanup, causing stale overrides to persist across sessions when users
manually close the automation window.
* fix(xiaoyuzhou): migrate from broken SSR scraping to authenticated API (fixes#1023)
Xiaoyuzhou removed SSR rendering — /podcast/<id> and /episode/<id> pages
now return 404, breaking fetchPageProps() which scraped __NEXT_DATA__.
Migrate podcast, podcast-episodes, episode, and download commands to use
the existing authenticated API client (requestXiaoyuzhouJson) that
transcript.js already uses successfully.
Changes:
- podcast.js: use /v1/podcast/get API endpoint
- podcast-episodes.js: use /v1/podcast/listEpisode API endpoint
- episode.js: use /v1/episode/get API endpoint
- download.js: use /v1/episode/get API endpoint
- utils.js: remove unused fetchPageProps, keep format helpers
- Update all affected tests (download.test.js, utils.test.js)
- Change strategy from PUBLIC to LOCAL (requires credentials)
* fix(xiaoyuzhou): align local strategy contract
* fix(xiaoyuzhou): align local api metadata
---------
Co-authored-by: jackwener <jakevingoo@gmail.com>
#1025 merged hot-rank adapters for eastmoney/tdx/ths but the
accompanying docs were missing. This breaks the Doc Check CI
workflow on every PR ('--strict' mode, exits non-zero when
`scripts/check-doc-coverage.sh` finds adapters without docs),
blocking merges across the board.
Adds a doc page per adapter, registers them in the adapters
index table, and adds sidebar entries in the VitePress config.
* feat: add hot stock ranking adapters for eastmoney, tdx, ths
Add three new site adapters for Chinese stock hot rankings:
- eastmoney/hot-rank: 东方财富热股榜
- tdx/hot-rank: 通达信热搜榜
- ths/hot-rank: 同花顺热股榜
All use Strategy.COOKIE browser mode with page.evaluate() DOM scraping.
Each includes co-located tests (13 tests total, all passing).
* fix(tdx,ths): add symbol validation and deduplication in evaluate()
Add seen Set for deduplication and skip entries with empty symbol/name,
matching the pattern already used in eastmoney/hot-rank.js.
* fix: refine hot-rank selectors based on browser inspection
- eastmoney: use table.rank_table tbody tr with td index-based extraction,
fix name from a[title] to avoid post content contamination
- tdx: use div.top-cell[data-code] data attributes for reliable extraction,
add tags column from div.tips-item.gnbk
- ths: use card-based layout selectors, remove price column (not in UI),
extract tags from div.tag.PFSC-R
* fix(hot-rank): align tdx and ths columns with actual output
* fix: register hot stock ranking adapters
---------
Co-authored-by: dengjingren <dengjingren@cn.wilmar-intl.com>
Co-authored-by: jackwener <jakevingoo@gmail.com>
* fix: preserve network capture and surface extension mismatch diagnostics
Older Browser Bridge installs can still connect to the daemon while
missing two capabilities we now rely on: the network-capture actions
and the extension version handshake. That created three user-facing
failure modes with real impact:
1. `opencli explore ...` crashed with `Unknown action: network-capture-start`
against an old extension, so exploration stopped before any site
analysis finished.
2. `opencli doctor` and `opencli daemon status` could show a healthy
connection even when the extension never reported a version, which
hid the compatibility problem and sent users toward the wrong fix.
3. After reloading a new extension, `explore` could still report
`Endpoints: 0 total, 0 API` because `handleNavigate()` detached the
debugger before top-level navigation and cleared the active network
capture state right before the page load we needed to observe.
Fix this in two layers:
- Teach `Page` to treat unsupported `network-capture-*` actions as an
old-extension compatibility case. It now warns once, memoizes the
unsupported state, and returns empty capture data instead of throwing.
- Teach `doctor` and `daemon status` to treat "connected but version
unknown" as a warning instead of a healthy state, so version-handshake
failures are visible immediately.
- Preserve the debugger attachment while network capture is armed, so
the initial navigation keeps the capture state alive and the extension
can record requests from the first page load.
Before:
- `opencli explore ...` -> `Error: Unknown action: network-capture-start`
- `opencli doctor` -> `[OK] Extension: connected` / `Everything looks good!`
- `opencli daemon status` -> `Extension: connected` even when the
extension version was missing
- `opencli explore ...` after reloading the extension -> `Endpoints: 0 total, 0 API`
After:
- `opencli explore ...` on an old extension -> warns once and continues
- `opencli doctor` -> `[WARN] Extension: connected (version unknown)`
- `opencli daemon status` -> `Extension: connected (version unknown)`
- `opencli explore ...` on the reloaded extension keeps network capture
armed across navigation instead of clearing it before the page load
* fix: reset network capture flags on closeWindow()
Prevents stale _networkCaptureUnsupported flag from persisting across
sessions when the user reinstalls or reloads the extension mid-session.
* fix: startNetworkCapture returns boolean to prevent false-positive on old extensions
When the extension doesn't support network-capture-*, startNetworkCapture()
now returns false instead of silently resolving. This ensures browser open/
network correctly falls back to the JS interceptor on old extensions.
---------
Co-authored-by: jackwener <jakevingoo@gmail.com>
* fix: auto-restart stale daemon and improve connection error messages
When daemon is running but extension never connected (stale daemon started
before extension was installed), the CLI now auto-restarts the daemon to
give the extension a fresh WebSocket endpoint, instead of just waiting
and then telling the user to install the extension.
Also improves error messages across cli.ts, bridge.ts, and doctor.ts to
suggest "opencli daemon stop && opencli doctor" as the quick fix, since
that's what actually resolves the issue.
* fix: version-aware stale daemon detection and improved error messages
- Daemon /status now includes `daemonVersion` field
- bridge.ts: when daemon is running but extension not connected, checks
daemonVersion vs CLI version. Only auto-restarts if version mismatch
(stale daemon from older CLI). Same-version daemon shows improved error
message with "opencli daemon stop && opencli doctor" hint.
- doctor.ts: explicitly identifies stale daemon (version mismatch) in
diagnostics report, shows daemon version in status line
- cli.ts: error message changed to suggest "opencli daemon stop && opencli doctor"
* fix: treat missing daemonVersion as stale, verify shutdown before respawn
- Missing daemonVersion (pre-version daemon) is now treated as stale,
covering the most common user scenario (old daemon without version field)
- After requestDaemonShutdown(), poll until daemon actually stops (port
released) before spawning new one, with 3s timeout
- If shutdown request fails, log warning instead of silently proceeding
- doctor.ts also treats missing daemonVersion as stale with clear message
* fix: fail explicitly when stale daemon replacement fails
- If shutdown request fails or port isn't released within 3s, throw
'Stale daemon could not be replaced' instead of blindly spawning on
an occupied port
- Add tests for all three stale-daemon branches: same-version (no
restart), missing daemonVersion (stale), mismatched version (stale)
* fix: use type-based error dispatch in browserAction instead of string matching
browserAction() now checks `instanceof BrowserConnectError` first and
renders both message and hint, instead of string-matching on message
content. This ensures stale daemon errors ("Stale daemon could not be
replaced") surface the actionable hint to the user.
* feat(grok): add image command for grok.com image generation
Add `opencli grok image <prompt>` which submits a prompt via the existing
grok.com browser session and returns the generated image URLs from the
latest assistant bubble.
Because assets.grok.com URLs are gated by Cloudflare and cannot be
downloaded with a plain HTTP client, the --out flag triggers an in-page
fetch(credentials: 'include') so the browser session's cookies and
referer are attached, then writes the decoded blob to disk.
Flags:
- --new start a fresh chat before sending
- --timeout max seconds to wait for the image (default 240)
- --count minimum number of images to wait for before returning
- --out directory to save downloaded images
Ships with unit tests for the helpers (isOnGrok, normalizeBooleanFlag,
dedupeBySrc, imagesSignature, extFromContentType, buildFilename).
* fix(grok): harden image composer and bubble detection
* fix(grok): harden image flow and docs
---------
Co-authored-by: jackwener <jakevingoo@gmail.com>
When ChatGPT macOS app is set to Chinese language, the "Options"
button label becomes "选项". This change checks for both English
and Chinese labels to find the button.
Co-authored-by: mad <mademing@maddeMac-mini.local>
Co-authored-by: Claude <noreply@anthropic.com>
* feat: implement Ref-Backed Locator for browser actions
Introduces a unified target resolution system with fingerprint
verification and structured error diagnostics.
Snapshot phase:
- Each interactive element now gets a fingerprint (tag, role, text,
ariaLabel, id, testId) stored in window.__opencli_ref_identity
- Zero overhead: metadata is already available during DOM walk
Resolution phase (new target-resolver.ts):
- Numeric input → ref path with fingerprint verification
- CSS-like input → querySelectorAll with uniqueness check
- No more silent first-match: ambiguous selectors are rejected
Error model (new target-errors.ts):
- stale_ref: element identity changed since snapshot
- ambiguous: CSS selector matched multiple elements (with candidates)
- not_found: element not in DOM or invalid input
- All errors include actionable hints for AI agents
base-page.ts:
- click() and typeText() now use two-phase resolve-then-act
- Existing CDP fallback for click preserved
* feat: migrate scrollTo to unified resolver pipeline
scrollTo now uses the same two-phase resolve-then-act pattern as
click and typeText, getting fingerprint verification and structured
error diagnostics (stale_ref/ambiguous/not_found) for free.
* fix: address review — stronger fingerprint verification & surface TargetError in CLI
1. Fingerprint verification now uses the full identity vector (tag, id,
testId, ariaLabel, role, text) instead of just tag/role/text. Strong
identifiers (id, testId) are decisive; remaining signals use majority
voting. Fixes false negatives where same-tag elements swapped.
2. browserAction() now renders TargetError with code, hint, and
candidates list instead of just the message string.
* fix: migrate get/select/type-autocomplete to unified resolver
- browser get text/value/attributes now resolve via resolveTargetJs
instead of raw querySelector, getting fingerprint verification and
structured errors for free
- browser select uses selectResolvedJs on __resolved element
- type command's autocomplete detection uses isAutocompleteResolvedJs
on the already-resolved element
- Fix empty-string text prefix match: fp.text="Login" + text="" no
longer falsely passes fingerprint check
jackwener