[mountainowl]

Requested by Ajay Singh · project thread

What changed

• Before: with analytics on, every review rebuilt the project profile from scratch. That meant a git ls-tree of the checkout, reading up to 100 manifests, and one SCM metadata API call, even when the same project had been reviewed minutes earlier.
• After: the stack scan and the SCM metadata are cached per project per UTC day. Later reviews that day reuse them. The events sent to PostHog are unchanged.
• How:
  • Each review runs in its own worker process, so the cache lives in a profiles table in the existing analytics_projects.sqlite, keyed by the random project ID.
  • A failed metadata read (metadata_status: unavailable) isn't cached, so the next review retries it.
  • The declared domain and project_type still come from config on every review, so a config change applies right away.

Verification

• tests/test_profile_cache.py runs the real collect_profile against a real git checkout and counts stack scans and metadata reads.
• On main, 3 reviews on one day did 3 scans and 3 reads. With this change they do 1 and 1, and a new day refreshes the cache.
• The tests also cover separate projects, a failed metadata read being retried, and a declared-profile change applying the same day.
• The full suite on Python 3.14 gives 1932 passed. The only failures are the 2 container-only test_ui_export failures, which also fail on main.
• ruff check, ruff format --check and cz check pass.

---

Generated by Claude Code