Failed keywords log the error and continue to the next one.
Not marked complete, so they'll be retried on next run.
Also await async onPage callback.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Browser crash recovery: per-keyword error handling, auto-recreate
browser and re-login if page dies mid-search
- Platform-level retry: if browser creation or login fails entirely,
retry up to 3 times with escalating waits (5/10/15 min)
- Progress saved after each search track (not just at end)
- Unhandled rejection handler to prevent silent process death
- Fixed async callback in classifyExternalJobs
- Added ensureLoggedIn to session.mjs for searcher flow
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Searcher creates browser first then verifies login, unlike applier
which checks auth before browser creation. Both paths now work.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- ensureLocalFile() downloads binary files (resume PDF) from S3 to temp
- Applier downloads resume from S3 before applying
- Cached in /tmp to avoid re-downloading each run
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
storage.mjs is now a single interface: loadJSON() and saveJSON()
route to either local disk or S3 based on settings.storage.type.
The app never touches disk/S3 directly.
- All queue/log functions are now async (saveQueue, appendLog, etc.)
- All callers updated with await
- Data validation prevents saving corrupt types (strings, nulls)
- S3 versioned bucket preserves every write
- Config: storage.type = "local" (disk) or "s3" (S3 primary)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- New lib/storage.mjs: async S3 backup on every queue/log save
- Versioned S3 bucket (claw-apply-data) keeps every revision
- Auto-restore from S3 if local file is missing or corrupt
- saveQueue/saveLog now validate data type before writing
(prevents the exact bug that corrupted the queue)
- IAM role attached to EC2 instance for credential-free S3 access
- Config: storage.type = "local" (default) or "s3"
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Invisible reCAPTCHA + server round-trip can take 10-30s. Increased
from 6x2.5s (~15s) to 10x3s (~30s) polling window.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Playwright on AWS doesn't have pressSequentially, use type() instead.
Extend submit polling from 4 to 6 iterations (~15s total) to handle
slower reCAPTCHA verification.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Use .ashby-application-form-field-entry class to find question
containers, then Playwright locators for reliable button clicking.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Ashby: detect Yes/No button questions for work auth, sponsorship,
and consent. Click appropriate button in beforeSubmit hook.
Greenhouse: use pressSequentially instead of fill() for React
Select comboboxes (Country, Location). Click the dropdown option
after typing.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Country inputs on Greenhouse use React Select comboboxes that
need option click after typing. Add 'country' to the autocomplete
trigger list alongside city/location.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Prefer resume-specific file inputs (#resume, #_systemfield_resume)
over autofill inputs. Skip autofill class inputs in fallback.
Also fix Ashby beforeSubmit to target #_systemfield_resume directly.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
After clicking submit, poll up to 4 times (2.5s + 3x2s = ~8.5s total)
for success text or form disappearance. Handles invisible reCAPTCHA
verification delay. Stops early on validation errors.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
External ATS jobs now show their actual apply URL instead of the
LinkedIn listing URL. Also added Ashby-specific "job not found"
text to closed detection.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Greenhouse wraps phone inputs in a fieldset with legend "Phone"
that gets picked up as an unanswered radio group. Skip these
since the actual phone input is handled separately.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Skip inputs with placeholder="Search" or pre-filled values when
they have a "Phone" label — these are country code pickers, not
the actual phone input.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Form filler now defaults to page root instead of scoping to
[role="dialog"]. LinkedIn Easy Apply passes its modal selector
explicitly. Fixes external ATS forms being scoped to wrong
container. Also improved Greenhouse handler with targeted
resume upload and form detection.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Ashby wraps file input inside #_systemfield_resume container — search
for the actual input[type="file"] element. Also capture and log
validation errors from the page when submit returns incomplete.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Generic handler now accepts options (transformUrl, formDetector,
submitSelector, resumeSelector, beforeSubmit, verifySelector, etc.).
Each ATS handler passes its overrides instead of reimplementing.
Registry resolves handlers by: apply_type -> URL domain -> generic fallback.
New ATS handlers only need to export SUPPORTED_TYPES and an apply() that
calls genericApply with platform-specific options.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Page scraping was grabbing wrong elements (e.g. "Location" instead of
company name on Ashby). Queue already has correct metadata.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Only classify jobs without an apply_url — those with one have already
been visited and don't need re-classification.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Ashby and unknown_external work. Greenhouse, Lever, Jobvite have
visible CAPTCHAs. Workday requires login. Keep those disabled until
proper handlers are built.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Invisible reCAPTCHA (size=invisible) fires on submit and usually passes
automatically. Only block on visible CAPTCHA challenges. Also fix form
detection to work without <form> wrapper elements.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Best-effort form filler for any career page with standard HTML forms.
Handles single-page and multi-step flows, resume upload, login wall
and CAPTCHA detection. All ATS stub handlers now delegate to generic.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Iterate over the full queue array instead of getJobsByStatus() results,
and pass it to saveQueue(). The previous code passed no argument, which
would corrupt or silently fail the save.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Secondary processes (standalone classifier, ad-hoc scripts) now write
to queue_updates.jsonl via writePendingUpdate() instead of modifying
jobs_queue.json directly. Primary processes pick up updates on next
loadQueue() call using atomic rename-then-read.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
External Apply buttons are <a> tags with LinkedIn redirect URLs, not
<button> elements. Extract the real URL from the redirect's query
parameter instead of clicking and waiting for a new tab.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replaces flat "Breakdown: ..." line with indented sub-items showing
Easy Apply, Wellfound, Greenhouse etc. counts under Ready to apply.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When LinkedIn is rate-limited, its jobs were filling the maxApps quota
but then getting skipped, leaving 0 applied. Now excludes easy_apply
jobs from selection during cooldown so Wellfound jobs get picked up.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
After LinkedIn search completes, visits each unknown_external job page,
clicks the Apply button, captures the redirect URL, and matches against
known ATS patterns to identify the actual application platform.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When rate limited, writes timestamp to data/linkedin_rate_limited_at.json.
Subsequent runs skip LinkedIn until 6 hours have passed. Other platforms
(Wellfound) continue unaffected. Cooldown file auto-deleted on expiry.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
LinkedIn shows different text on direct /apply/ URL vs button click:
- Button: "Easy Apply limit"
- Direct URL: "limit daily submissions" / "apply tomorrow"
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Don't break the entire run when LinkedIn rate limits — skip LinkedIn
jobs and continue to Wellfound and other platforms.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When LinkedIn shows "Easy Apply limit" message, detect it in the
modal-open check, return rate_limited status, put job back in queue,
send Telegram alert, and stop the entire run. Prevents burning retries
and wasting browser sessions when rate limited.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Some LinkedIn listings have image-only upload fields (JPG/PNG).
Don't attempt to upload a PDF resume to these inputs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Prevents OpenClaw from relaying full report as plain-text duplicate.
sendTelegram() handles formatted delivery, stdout just says "sent".
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- answerFor returning '' (intentionally blank) was treated as falsy,
falling through to AI which fabricated "123 Main Street". Now
empty string skips the field without triggering AI or reporting unknown.
- status.mjs was printing to stdout AND sending via Telegram, causing
OpenClaw to relay a duplicate plain-text copy. Now only prints to
stdout when Telegram isn't configured.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
LinkedIn validates number fields even when not marked required in the
DOM. Previously these were skipped (no AI call, no answer). Now number
fields always trigger AI fallback and are reported as unknown if empty.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When a batch completes but scores aren't written back (collection
error), jobs get stuck with filter_batch_id set and never re-submitted.
Now checks: if no filter_state.json exists (no batch in flight) but
jobs have batch markers without scores, clear them so they get
re-submitted on the next run.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
