Fix skill install for qualified namespace/name references and missing parent directory

[samuv]

Summary

• Fix Install to fall back to registry catalog lookup when a namespace/name reference (e.g. io.github.stacklok/skill-creator) fails OCI pull because the namespace is parsed as a registry host. Names with an explicit tag or digest (e.g. ghcr.io/org/skill:v1) or more than one / (e.g. ghcr.io/org/skill) are unambiguously OCI and do not fall back
• Extract installFromResolvedRegistry helper to share dispatch logic between installFromRegistryLookup and the new fallback path
• Fix WriteFiles failing to acquire the lock when the parent skills directory does not yet exist (e.g. first install to ~/.cline/skills/)
• Fix pre-existing bug in installFromResolvedRegistry OCI branch: pass result.Skill.Metadata.Name instead of opts.Name (the OCI ref string) to installAndRegister, so group registration and rollback use the correct skill name

Behavioral change note

Previously, a registry entry that existed but had no installable OCI or git package fell through and returned 404 Not Found. It now returns 422 Unprocessable Entity ("resolved but no installable package"), which is a more accurate response.

Type of change

• Bug fix

Test plan

• Unit tests (task test)
• Linting (task lint-fix)

Changes

File	Change
pkg/skills/skillsvc/skillsvc.go	Add OCI→registry fallback in Install and GetContent; tighten guard to also block multi-segment refs; add debug log before fallback; fix opts.Name bug in OCI branch of installFromResolvedRegistry
pkg/skills/skillsvc/skillsvc_test.go	Extend TestInstallQualifiedNameOCIFallback with digest, multi-segment, and ambiguity cases; pin Pull mock arguments
pkg/skills/gitresolver/writer.go	MkdirAll parent before acquiring file lock
pkg/skills/gitresolver/writer_test.go	Add test for missing parent directory

Does this introduce a user-facing change?

Yes — thv skill install io.github.stacklok/<skill-name> now works correctly instead of failing with "no such host".

[codecov]

Codecov Report

❌ Patch coverage is 80.48780% with 8 lines in your changes missing coverage. Please review.
✅ Project coverage is 69.22%. Comparing base (5342824) to head (f528c88).
⚠️ Report is 10 commits behind head on main.

Files with missing lines	Patch %	Lines
pkg/skills/skillsvc/skillsvc.go	84.61%	5 Missing and 1 partial ⚠️
pkg/skills/gitresolver/writer.go	0.00%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4841      +/-   ##
==========================================
- Coverage   69.25%   69.22%   -0.04%     
==========================================
  Files         537      537              
  Lines       55380    55401      +21     
==========================================
- Hits        38354    38351       -3     
- Misses      14080    14105      +25     
+ Partials     2946     2945       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[JAORMX]

Hey! Nice fix. The core approach is solid... the :@ heuristic to distinguish "ambiguous namespace/name" from "unambiguous OCI ref with tag/digest" is clean and easy to reason about. The installFromResolvedRegistry extraction is a good refactor, and the writer.go fix addresses a real first-install bug.

I verified the locking story and it's fine. The fallback path in Install() never holds a lock when calling downstream installers, so no deadlock risk. InstallOptions is passed by value and slice fields are replaced wholesale, not mutated in place. All good there.

A few things I'd like addressed before merging:

The multi-segment OCI edge case. A ref like myregistry.com/org/skill (3 segments, no tag) passes the guard and triggers the fallback. The namespace filtering makes a false match unlikely, but it's a supply-chain confusion vector in principle. Adding || strings.Count(opts.Name, "/") > 1 to the guard closes this cleanly. See inline comment.

The writer.go fix needs a test. All existing WriteFiles tests pre-create the parent directory, so no test actually proves this fix works. A test with a non-existent parent would take ~10 lines and directly exercise the bugfix.

CodeQL alert. The #nosec / lgtm annotations won't suppress CodeQL (different tool). This needs to be dismissed in GitHub's Security tab.

The rest are nits and nice-to-haves (logging the original OCI error before fallback, noting the 404->422 behavioral change in the PR description, tightening gomock.Any() in tests). See inline comments for details.