fix(registry): stop a KV outage 500ing every /gh page

[harlan-zw]

❓ Type of change

• 🐞 Bug fix

📚 Description

Sentry check-in of the skilld project. Seven of the eight open issues were already closed by earlier commits; this is the one gap left.

skill-source-gone middleware runs on every /gh request and read its key set with a bare useStorage('cache').getItem. Workers KV answered one of those with:

Error: KV GET failed: 500 Internal Server Error
  GET /api/skill-related/dylantarre/animation-principles/educator-teacher
  url https://skilld.dev/gh/dylantarre/animation-principles/educator-teacher

acbd24c moved the skill-related reads onto readCache, but this one was on the same request and stayed bare. The D1 fallback under it could fail the same way, and did elsewhere as D1_RESET_DO.

resolveGoneSkillKeys now returns string[] | null and the middleware only sets 410 on a resolved set. The cost of an unresolved lookup is one stale 200 on a deleted skill until the next request. The cost of the old behaviour was the page.

One thing I went back and forth on: during a KV outage the fallback puts a D1 query back on the hot /gh path, which is the thing the cache is there to prevent. It is capped by the same failure that caused it, and 410-vs-200 is not worth a second cache layer, so I left it.

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