perf(registry): skip hourly AI catalog scans when no work is missing

[harlan-github-agent]

🔗 Linked issue

Closes #664

❓ Type of change

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

📚 Description

The d1-query-review routine measured 373 seconds and 19 million row reads per week on the hourly AI work scans, and the cost lands even when there is nothing to do: both work scans sort the entire eligible catalog by stars before LIMIT, so an idle hour pays for a full sort and returns zero rows.

The submit task now counts the missing work per kind first and only runs the ordered scan when the count is nonzero. The count keeps the CASE guard that skips trimming completed sources and has no sort step. Local fixture, 20k skills, fully idle:

== ordered select (the hourly scan being skipped) ==
  SCAN s
  CORRELATED SCALAR SUBQUERY 1
  SEARCH generated USING INDEX sqlite_autoindex_skill_generated_1 (owner=? AND repo=? AND name=? AND kind=?)
  SEARCH r USING INDEX sqlite_autoindex_repos_1 (owner=? AND repo=?)
  USE TEMP B-TREE FOR ORDER BY
== CASE-guarded count (the idle guard) ==
  SCAN s
  CORRELATED SCALAR SUBQUERY 1
  SEARCH generated USING INDEX sqlite_autoindex_skill_generated_1 (owner=? AND repo=? AND name=? AND kind=?)
  SEARCH r USING INDEX sqlite_autoindex_repos_1 (owner=? AND repo=?)

On the same fixture the count agreed with the scan in both directions: zero and empty on an idle catalog, one row each after adding a single unprocessed skill, so a zero count is a safe skip signal. Work that appears between the count and the scan waits for the next hourly run.

One thing I could not verify locally: a scoped typecheck of the scheduled task file. An ad-hoc project pulls both workers' Env declarations and reports unrelated binding errors, including on untouched lines, so that file is covered by CI's nuxt typecheck instead.

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

[harlan-github-agent]

🤖 READY · 93/100

Harlan Agent Kit posted this automated review. It is not Harlan's personal review or approval. AI open source policy. A person still decides the merge.

• Merge gate: Passed.
• Review gate: Passed. No material issues.
• CI gate: Passed.

[harlan-github-agent]

Harlan Agent Kit posted this automated review. AI open source policy.

Automated Review: READY.

Read the Review results