Skip to content

Releases: andrewaws26/claude-works

Release list

v0.3.0

Choose a tag to compare

@andrewaws26 andrewaws26 released this 04 Jul 18:07

The self-sufficient release: clone it, type /setup, and the system runs for you.

  • Built-in live discovery: the new default boards source queries the public Ashby/Greenhouse/Lever posting APIs over your seed_boards list (policy.json). No private scripts, no keys; 15 default seed boards returned 3,300+ live roles at release time.
  • fetch_job_description: headless plain-text JD for any Ashby/Greenhouse/Lever URL, so scoring runs on the JD, not the title.
  • fetch_verification_code: completes emailed-code submit gates from the applicant's own inbox (scoped, read-only, revocable app password). Captchas and attestations still always park.
  • PLAYBOOK.md: the full, sanitized operator manual: every per-ATS form-driving lesson learned in production (Ashby spam wall, Greenhouse geocode/iframe/email-code, Lever captcha protocol, Workday wizards, Gem, Hirebridge postbacks, the submit-debug checklist).
  • OPERATING.md: the loop's operating model: queue-first, 4-gate resumes, de-dup semantics, hard walls, self-improvement mandate.
  • /setup: bundled Claude Code command that onboards a new candidate from example jobs + their resume: derives policy, angles, seed boards, and confirmed resume fragments, registers the claude-works + Playwright + JobDataLake MCP stack, and smoke-tests the pipeline.
  • 88 tests, ruff + mypy clean, CI on Python 3.10-3.13.

v0.2.0

Choose a tag to compare

@andrewaws26 andrewaws26 released this 04 Jul 17:42

First PyPI release.

  • Per-candidate policy: drop a policy.json in JOBSEARCH_DATA_DIR to replace the rails (excluded companies/domains, over-level terms, skill gaps), comp floor, pursue threshold, scoring vocabularies, and curation lanes without editing code. See examples/policy.sample.json.
  • Demo mode: sanitized fixtures in examples/ so every tool works from a fresh clone with zero network.
  • Hardening: hard-capped roles rank below clean fits in discovery; whole-word excluded-company matching; flock-locked atomic ledger writes with role_key de-dup; all identity PII env-only; honest render-failure findings.
  • New tool: curate_queue triages the discovery queue with auditable park reasons.
  • CI (ruff + mypy + pytest) on Python 3.10-3.13; py.typed shipped.

Install: pip install claude-works or run directly with uvx claude-works.