SPA frameworks (React Router, Vue Router, Next.js) require a prior
mouseMoved event to establish pointer hover state before the click
event synthesizes correctly. Without it, the router's click listener
never fires and the URL remains unchanged (url_changed: false).
Add a single mouseMoved dispatch at (x, y) with button=none before
the existing mousePressed/mouseReleased loop — matches Playwright's
standard click implementation.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Injects a SPA-style fixture (history.pushState on click with
preventDefault) into about:blank and verifies that browser click
triggers the router handler and returns url_changed=true.
Tests the mouseMoved fix — will fail until dispatch_click sends
mouseMoved before mousePressed in click.rs.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Address Codex bot P1 review finding: skipping favicon/data: URLs on
requestWillBeSent without also skipping their loadingFinished would
undercount active requests, causing premature network-idle.
Changes:
- Maintain `skipped_request_ids` HashSet to track suppressed requests
- loadingFinished/loadingFailed checks the set and suppresses delta
- Add EventSource exclusion (persistent like WebSocket)
- Handle redirect double-counting (redirectResponse → don't re-increment)
- Add comprehensive unit test covering all skip/undercount scenarios
WebSocket upgrade requests fire `Network.requestWillBeSent` but never
`Network.loadingFinished`, leaving the pending counter stuck at ≥1
and causing `wait network-idle` to always time out on sites with
persistent WebSocket connections (analytics, real-time features).
Align with Playwright's behaviour by excluding:
- WebSocket upgrades (type == "WebSocket")
- Favicon requests (url ends with /favicon.ico)
- data: URLs (not intercepted by Network Service)
Before: healthline.com 30s timeout, flightaware.com ~25s
After: healthline.com 523ms, flightaware.com 514ms
Run sync-versions.js to align 6 platform packages with CLI version.
Fixes broken release path where optionalDependencies would point to
unpublished npm versions.
Port search and get action commands to cli-v2 with identical logic:
- API client module for search_actions and get_action_by_area_id
- --api-key global flag (env: ACTIONBOOK_API_KEY)
- Updated help text with new commands
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* [packages/cli-v2]feat: auto-restart daemon on version mismatch
CLI now compares full BUILD_VERSION (not just major.minor) with the
running daemon. On mismatch, sends SIGTERM to old daemon, waits for
flock release, and starts a fresh daemon automatically.
Changes:
- client.rs: Replace check_version(major.minor) with versions_match
(full string). Add restart_daemon() + wait_for_daemon(). Socket
connect path now auto-restarts instead of returning VersionMismatch.
- server.rs: Write daemon.version file on startup, clean up on shutdown.
- Tests updated to verify exact version matching semantics.
* [packages/cli-v2]fix: treat missing daemon.version as version mismatch
Old daemons that never wrote daemon.version were silently accepted
instead of being restarted. Now missing/empty version file returns
false from versions_match(), triggering auto-restart on upgrade.
* [packages/cli-v2]test: E2E tests for daemon version auto-restart
Two scenarios:
- daemon_version_mismatch_auto_restart: tamper daemon.version to
simulate rebuild, verify CLI auto-restarts daemon with new PID
- daemon_missing_version_file_triggers_restart: delete daemon.version
to simulate old daemon, verify restart is triggered
* [packages/cli-v2]fix: harden daemon restart — wait for version file, abort on timeout
- P1: restart_daemon() now returns error if old daemon is still running
after 5s SIGTERM timeout, instead of force-cleaning files
- P2: connect() waits up to 1s for daemon.version file to appear before
treating missing version as mismatch (avoids killing starting daemon)
* [packages/cli-v2]fix: track target PID during restart instead of global flock
Use kill(pid, 0) to check if the specific old daemon process exited,
instead of is_daemon_running() which checks global flock state. This
prevents false "did not exit" errors when a concurrent CLI has already
started a new daemon that holds the flock.
* [packages/cli-v2]fix: guard cleanup against concurrent restarts + wait for version file
- P1: Before cleanup_stale_files(), check if a concurrent CLI already
started a matching-version daemon. If so, skip cleanup to avoid
deleting the new daemon's socket/ready/version files.
- P2: When daemon is running but not connectable, wait up to 1s for
version file to appear before treating absence as mismatch. Prevents
killing a same-version daemon still booting up.
* [packages/cli-v2]fix: treat EPERM as alive in PID liveness check
kill(pid, 0) returns EPERM when the process exists but signaling is
not permitted. The previous code treated this as "dead", which could
cause restart logic to delete a live daemon's files. Now only ESRCH
(no such process) is interpreted as definitely dead.
* [packages/cli-v2]fix: reject non-positive PID before sending SIGTERM
kill(2) treats pid<=0 specially (0=caller's process group, negative=
process group). Validate pid>0 to prevent accidental mass kill from
corrupted daemon.pid file.
* [packages/cli-v2]fix: check SIGTERM result + use canonical paths in cleanup
- send_sigterm() now returns bool (false = ESRCH, process already dead)
restart_daemon() skips wait loop when process is already gone
- cleanup_stale_files() uses server::socket_path().with_extension()
instead of hardcoded filenames for forward compatibility
* [packages/cli-v2]fix: skip cleanup when PID file is missing but daemon is running
When daemon.pid is missing/corrupt, don't blindly delete socket files.
Check flock first — if daemon is still running, return error. If daemon
is truly gone, clean up and start fresh.
* [packages/cli-v2]test: add E2E tests for orphan Chrome recovery and SIGHUP shutdown (#219)
Two new tests in browser_lifecycle.rs:
- lifecycle_daemon_sigkill_orphan_recovered: SIGKILL daemon, verify orphan
Chrome stays alive, then browser start must succeed without timing out
- lifecycle_daemon_sighup_closes_chrome: SIGHUP triggers graceful shutdown
and kills Chrome (same as SIGTERM)
Tests will fail until implementation in start.rs (chrome.pid orphan detection),
close.rs (chrome.pid cleanup), and server.rs (SIGHUP handler) is added.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* [packages/cli-v2]fix: recover from orphan Chrome after daemon SIGKILL (#219)
Three changes:
1. start.rs: Before launch_chrome, read chrome.pid from user-data-dir.
If the recorded PID is still alive (kill -0 probe), SIGKILL it and
wait 500 ms, then remove chrome.pid. After successful launch, write
the new Chrome PID to chrome.pid so a future restart can detect it.
2. close.rs: On normal browser close, remove chrome.pid before deleting
the profile directory so a stale file cannot trigger false orphan
detection on the next browser start.
3. server.rs: Register SIGHUP handler in run_daemon(). Terminal close
(SIGHUP) now triggers the same graceful shutdown path as SIGTERM,
killing all Chrome processes before the daemon exits.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* [packages/cli-v2]fix: re-read source coordinates after destination scroll in drag
resolve_center() now scrolls to viewport center unconditionally. For
selector-to-selector drags, the destination scroll can invalidate the
source's viewport coordinates captured earlier. Fix by re-reading
source position via get_center() (no scroll) after destination resolve.
Also returns nodeId from resolve_center() so callers can refresh
coordinates without re-resolving the selector.
* [packages/cli-v2]fix: preserve source frame_id when refreshing drag coordinates
get_center() now accepts an explicit frame_id parameter so drag can
pass the saved source frame context. This prevents cross-frame drags
from querying getBoxModel in the wrong CDP session after destination
resolution overwrites resolved_frame_id.
[packages/cli-v2]test: remove unreliable activeElement.id assertion from focus_json (#218)
In headless Chrome, document.activeElement reverts to document.body
after the CDP focus sequence completes (no OS-level window focus event).
The pre/post reference comparison used by focus_changed runs inside the
CDP sequence where focus is still held, so focus_changed=true is the
authoritative signal. Remove the post-command activeElement.id and
__ab_last_focused eval checks that were flaky in CI.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* [packages/cli-v2]fix: resolve 4 CI E2E failures (#217)
1. start.rs: update native_tabs[0] URL after Page.navigate so registry
reflects the navigated URL (fixes lifecycle_open_with_url_json)
2. browser_lifecycle.rs: verify --open-url via `browser url` instead of
`browser status` text (§7.3: status text excludes tab URLs)
3. focus.rs: add Runtime.callFunctionOn after DOM.focus to durably update
document.activeElement in headless Chrome
4. wait.rs: mark wait_network_idle_json_happy_path #[ignore] with CI note
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* [packages/cli-v2]fix: collapse nested if-let in focus.rs (clippy collapsible_if)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
All element interaction commands now scroll the target element to the
viewport center before operating, following Playwright's approach. This
prevents edge cases where elements at the viewport edge are intercepted
by sticky headers/footers.
Changes:
- element.rs: Replace DOM.scrollIntoViewIfNeeded with
scrollIntoView({ block: 'center' }) via Runtime.callFunctionOn
- type_text.rs, fill.rs, select.rs, focus.rs: Add scroll_into_view()
call before operating (previously had no scroll at all)
- click and hover inherit the center-scroll automatically via
resolve_center() and scroll_into_view()
- 7 new E2E tests verifying offscreen element operations
* [packages/cli-v2]feat: enhance stealth anti-detection for bot.sannysoft.com
- Add Playwright/Puppeteer trace removal (__playwright, __pw_manual, __PW_inspect)
- Enhance chrome.loadTimes() with realistic staggered timestamps
- Add navigator.maxTouchPoints = 0 for desktop consistency
- Add screen.colorDepth/pixelDepth = 24 for headless consistency
- Add canvas fingerprint noise via temporary offscreen canvas (non-destructive)
- Add CDP Emulation.setDeviceMetricsOverride (1920x1080) in attach()
- Replace window.__stealth_applied__ with closure-scoped guard to avoid detection
- Update contract tests and CDP mock test for new capabilities
* [packages/cli-v2]fix: use non-enumerable property for stealth idempotency guard
Replace var-scoped __stealth_guard__ (broken due to IIFE hoisting) with a
non-enumerable, non-configurable property on Navigator.prototype. This
prevents double injection while avoiding detection by anti-bot scripts.
Addresses Codex review feedback on #436.
[packages/cli-v2]docs: update api-reference §10.1 snapshot — path replaces content, cursor default true
- §10.1 JSON data: replace `content` (inline text) with `path` (file path to saved snapshot)
— content is now saved to a file only; `nodes` array remains for programmatic use
- §10.1 text output: update example to show "output saved to <path>" format
- §10.1 cursor param: default changed to `true` (always enabled)
- §1.3/§1.4/§5.2: session_id auto-generated as sN (global counter, mode-agnostic)
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
[packages/cli-v2]feat: snapshot --cursor enabled by default (#212)
Change cursor field default from false to true in SnapshotCmd.
Add cursor_default() serde helper so IPC deserialization also defaults
to true when the field is absent.
Scope: no --no-cursor flag per mcfeng confirmation.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
[packages/cli-v2]test: E2E tests for snapshot --cursor default on (#211)
Add /cursor-fixture deterministic page to harness.rs local server
(cursor:pointer, onclick, tabindex divs) and url_cursor_fixture() helper.
Add snap_cursor_on_by_default test to snapshot.rs: verifies that the
default browser snapshot (no flags) includes cursor-interactive elements.
Expected to fail until #212 implementation (cursor default = true) lands.
Scope: cursor on by default, no --no-cursor per mcfeng confirmation.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* [packages/cli-v2]test: update E2E default session ID from SLOCAL-1 to s1 (#208)
Change DEFAULT_LOCAL_SESSION_ID const from "SLOCAL-1" to "s1".
Update all literal [SLOCAL-1 / [SLOCAL-1] text assertions to [s1 / [s1].
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* [packages/cli-v2]test: fix cargo fmt in browser_lifecycle.rs (#208)
Collapse short assert! to single line per rustfmt.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
[packages/cli-v2]feat: rename auto-generated session IDs from SLOCAL-N/SCLOUD-N/SEXT-N to sN
- types.rs: is_valid() drops the uppercase prefix special cases; all IDs
now use the single lowercase format. auto_generate(n) no longer takes
a mode param and emits sN (s1, s2, ...). from_profile() drops mode param.
- registry.rs: generate_session_id() uses a single "s" prefix for
max_active_prefix_n; removes per-mode prefix match. Mode is no longer
encoded in auto-generated IDs.
- Unit tests updated: SLOCAL-N/SCLOUD-N/SEXT-N assertions replaced with
sN equivalents; added test that old uppercase prefixes are now rejected.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Allow callers to specify a custom tab ID instead of relying on auto-assignment.
Includes duplicate ID validation with TAB_ID_CONFLICT error and rollback on conflict.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* [packages/cli-v2]fix: list-tabs uses CDP real-time state, syncs registry
list-tabs now builds the tab list from CDP Target.getTargets instead of
cross-matching with the registry. This ensures it always reflects actual
browser state even after restart or reconnect.
Registry sync logic:
- Matching native_id → keep short tab ID, update url/title
- Stale registry tabs (not in CDP) → removed
- New CDP tabs (not in registry) → assigned new short ID via push_tab
* [packages/cli-v2]fix: attach newly discovered tabs in list-tabs sync
When list-tabs discovers a CDP page not in the registry, it now calls
cdp.attach() after assigning a short tab ID. Without this, subsequent
commands on the new tab would fail with "no CDP session for target".
* [packages/cli-v2]fix: collapse nested if per clippy
Snapshot output is now written to a timestamped file under
~/.actionbook/sessions/{session_id}/ instead of being returned inline.
The session data directory is created on browser start and cleaned up
on browser close with absolute path safety checks. The snapshot response
now includes a ref usage hint and the file path.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
[packages/cli-v2]fix: describe --nearby 4 bug fixes in JS (children/siblings/state/parent)
Fix 1 (children): relax isig() — add explicit role attr check and
e.innerText check so nav/semantic elements with descendant text pass.
Fix 2 (siblings): same updated isig() catches role="tab" elements that
only have text inside a child span (no direct text node).
Fix 3 (state qualifiers): sm() now appends [disabled/checked/selected]
when detected via element properties or aria-selected attribute.
Fix 4 (parent): add isig_par() that returns true when any direct child
has an explicit role attribute, preventing always-null parent for
structural wrappers that contain semantic children.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
- Update all e2e test assertions from dot notation ("ok browser.start")
to space-separated ("ok browser start") matching the refactored
COMMAND_NAME constants
- Fix type command clap panic: change from separate selector/text fields
to `#[arg(num_args = 1..=2)] args: Vec<String>` pattern (same as fill)
to avoid "non-required positional before required" clap assertion
- Fix storage test command_name helper to use space separator
Reduces e2e failures from 57 (on release/1.0.0 CI) to ~5 pre-existing
timing/flaky issues.
Address PR review: when Network.enable fails after Target.attachToTarget
succeeds, send Target.detachFromTarget before rolling back the local
mapping to prevent an unreachable browser-side session.
[packages/cli-v2]fix: return contentDocument.body.innerText for iframe/frame elements in browser text
iframe.innerText is always empty; the text content lives in the embedded document.
Detect IFRAME/FRAME via tagName and read contentDocument.body.innerText instead.
Cross-origin iframes catch SecurityError and return empty string.
Closes#203
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add 60s timeout to CdpSession::execute() to prevent indefinite hangs
on half-open sockets or unresponsive CDP commands (was infinite wait)
- Propagate Network.enable errors in attach() with session mapping rollback
to avoid half-initialized ghost tabs
- Return error from cdp_navigate() on WebSocket EOF instead of silent Ok(())
- Clean up event subscriptions on detach() via unsubscribe_all()
- Remove requestServedFromCache from network pending counter (Chrome CDP
spec: it's an informational event, not a lifecycle terminator)
- Add 8 unit tests covering all fixes
[packages/cli-v2]feat: add coordinate and selector-less support to fill and type commands
- Extract ClickTarget enum and parse_target() from click.rs to element.rs (shared)
- fill: accept optional selector (CSS/XPath/@ref/x,y) or omit to fill activeElement
- type: accept optional selector (CSS/XPath/@ref/x,y) or omit to type into activeElement
- Both commands dispatch mouse click to focus when coordinates are given
- Both commands return NO_FOCUSED_ELEMENT error when no selector and no focused element
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
load waits for ALL resources (including third-party scripts like Adobe
Analytics, Google Maps SDK) which causes timeouts on heavy pages like
Costco Travel. domcontentloaded returns when the DOM is ready —
sufficient for snapshot + click workflows.
Verified: Costco Travel Regent page loads in 1.6s with domcontentloaded
vs 60s+ timeout with load.
mcfn