fix(ops): recover dead claims and retain rolling metrics

[harlan-zw]

🔗 Linked issue

Related to #618

❓ Type of change

• 📖 Documentation
• 🐞 Bug fix
• 👌 Enhancement
• ✨ New feature
• 🧹 Chore
• ⚠️ Breaking change

📚 Description

A failed Repository job keeps its progress claim until the 30-minute timeout. Reclaim it when the owning job ends, while preserving active jobs' claims.

Keep X Post and verified Skill counts on a fixed rolling day, regardless of the check-in baseline. Import the Skill detail handler during setup.

🤖 AI disclosure: Harlan Agent Kit modified this description. My AI open-source policy.

[harlan-zw]

🤖 Harlan Agent Kit Agent checked the live metric queries at 2026-10-11T01:50:37.829Z.

Read-only probes used that clock with one-hour and 48-hour baselines. Both returned 140 Posts and 47 verified Skills over the rolling day. No production data changed.

[harlan-zw]

🤖 Harlan Agent Kit Agent assessment of head SHA.

Reviewed head: 1d202340d0b660d1c0ac48879ed8a2fe03318a4b.

The Service Review remained queued 20 minutes after the recorded request. Its exact-head stop returned Stopped. One independent, read-only reviewer assessed the full diff, surrounding code, diagram, schema, and job lifecycle.

No evidence-backed findings. Rolling bounds exclude older and future records. Missing or terminal jobs permit atomic reclaim. Active and released jobs retain ownership. The job lookup uses its primary key.

Independent verification: 3 files and 39 tests passed. Review confidence: 95/100. Live D1 concurrency was not exercised.

Current-head CI passed: lint, typecheck, tests, build, harness proof, and declared Skills. The full local suite passed 3,872 tests.

This assessment is separate from Service READY. The exact-head Service Review remains stopped. A human merge decision is outstanding.