stacklok / toolhive / 23433487141

push

github

Resolve plain skill names from registry during install (#4200)

* Resolve plain skill names from registry during install

When `thv skill install <name>` does not find the skill in the local OCI
store, the code previously created a dead-end "pending" DB record. Now it
queries the configured skill registry/index for an exact name match and,
if found, extracts the OCI reference and installs from it. If no match is
found anywhere, a 404 with an actionable error message is returned.

This is Phase 2 of #4015. The resolution cascade is now:
  1. Local OCI store (from `thv skill build`)
  2. Registry/index lookup (new)
  3. Actionable 404 error (replaces silent pending record)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Use lazy skill lookup to pick up registry config changes

The SkillLookup was created once at server startup from the config at
that point. If the registry was changed at runtime (via the API or
thv config set-registry), the skill lookup still used the old provider,
so plain-name installs could not find skills in the new registry.

Replace the eager buildSkillLookupOption with a lazySkillLookup that
calls GetDefaultProviderWithConfig on each SearchSkills invocation.
After ResetDefaultProvider (called on every registry config change),
the next lookup creates a fresh provider from the current config.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Update E2E tests for registry-based skill resolution (#4202)

* Update E2E tests for registry-based skill resolution

Plain name installs without a prior build now return 404 instead of
creating a dead-end "pending" record. Update all E2E tests that relied
on the pending path to use the build-then-install flow, and add a test
verifying the 404 behavior for unresolvable names.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Add E2E test for registry-based skill name resolution

Exercises the full registry lookup chain end-t... (continued)

83 of 108 new or added lines in 2 files covered. (76.85%)

50503 of 77838 relevant lines covered (64.88%)

73.12 hits per line