[RaviTejaKNTS]

Repairs directory automations that were broken or silently stale, and adds published CLI apps to the shared GitHub refresh.

Changes

• CLI apps: unions published CLI project links with self-host apps and fetches each GitHub project once, preserving self-host memberships. Updates CLI maintenance only when it changes, using the same 110/395-day push-age rule as the CLI pipeline. Existing triggers touch the public date and queue /cli-apps refreshes. The combined production queue has 4,677 unique projects: 500 per run twice daily gives a full attempt cycle of about five days.
• Technical refresh: fixes the qa-gate.mjs crash (path is not defined). The collector now tolerates up to 5% unreadable sources and lists them in the job summary. Install options stay dry-run.
• GitLab/Codeberg refresh: uses each repo's own GitLab host (gitlab.torproject.org, framagit.org, and others), keeps nested paths, and sends GITLAB_TOKEN only to gitlab.com. Adds bounded retries and a partial-failure summary.
• GitHub refresh: a new migration adds attempt and failure tracking on github_projects, so deleted or moved repos stop blocking the oldest-first queue. checked_at is never faked. The migration has a 10s lock timeout and a 120s statement timeout. The script falls back to the old behavior until the migration is applied.
• Docker Hub discovery: switches to a search route that works without a credential. GitHub credentials go only to api.github.com. Dedupe now runs even when one catalog fails.
• monitoring.json: corrects five job descriptions.

Checks

• npm run quality: 116 core tests pass.
• 41 focused tests pass with no skips, including isolated Postgres tests for the shared queue, old-schema fallback, CLI maintenance, public-date touches and cache refresh requests.
• The original automation fixes passed Sol review; its three follow-ups are addressed in 2122c65.

The production health-check script on OVH was fixed separately on the server.

🤖 Generated with Claude Code