Skip to content

fix: pin security-reviewer to sonnet, explicit model fields, drift test#39

Merged
Soul-Craft merged 2 commits into
mainfrom
claude/wizardly-kalam
Apr 16, 2026
Merged

fix: pin security-reviewer to sonnet, explicit model fields, drift test#39
Soul-Craft merged 2 commits into
mainfrom
claude/wizardly-kalam

Conversation

@Soul-Craft

Copy link
Copy Markdown
Owner

Summary

  • Pin security-reviewer agent from inheritmodel: sonnet so security review never runs on Haiku even when the main session is in fast mode
  • Add explicit model: inherit to cache-analyzer (was a silent default — now visible in frontmatter)
  • Add Group 11 to test-docs.sh: validates the agent model table in /session-execute against actual agent frontmatter at test time, preventing silent drift
  • Update CLAUDE.md: session-skill coordination note, explicit Phase 4 section, corrected test count (16 → 18)

Type of change

  • Bug fix
  • New feature (species, skill, patch type, etc.)
  • Refactoring (no behavior change)
  • Documentation
  • CI / tooling

Testing checklist

  • make test-docs — Group 11 passes (18/18 tests)
  • CLAUDE.md updated with session-skill semantics and Phase 4 note

Risk

No patching, backup/restore, or codesign changes. Docs and agent config only — the model: sonnet pin is the only behavioral change, and only affects the security-reviewer agent (previously ran at session model, now floors at Sonnet).

- security-reviewer: model inherit → sonnet (floor prevents Haiku runs)
- cache-analyzer: add explicit model: inherit (was silent default)
- session-execute: update agent table to match (security-reviewer sonnet)
- test-docs.sh: Group 11 validates table ↔ frontmatter at test time
- CLAUDE.md: session-skill coordination note, Phase 4 section, fix test count

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
@github-actions

Copy link
Copy Markdown

Local tests not yet uploaded

This repository runs macOS-dependent tests locally (not in CI) to keep
GitHub Actions costs down. Before this PR can merge, please:

  1. Run the full test suite on your macOS machine:
    scripts/test-all.sh
    
  2. Upload the results to GitHub:
    scripts/upload-test-results.sh
    

Results will appear either as a Check Run or a PR comment on this commit.
See CLAUDE.md for more detail on the testing architecture.

@Soul-Craft

Copy link
Copy Markdown
Owner Author

Local Tests (macOS)

183/183 passed — Local Tests (macOS)
Commit: afc3d7422a91 on claude/wizardly-kalam
Duration: 19s
Environment: Darwin 25.4.0 (arm64)
Swift: Apple Swift version 6.3 (swiftlang-6.3.0.123.5 clang-2100.0.123.102)

Tier Passed Failed Duration
Smoke
✅ smoke 13 0 4s
Core
✅ unit 98 0 11s
✅ security 25 0 1s
Real-world
✅ ui 23 0 2s
Full system
✅ snapshots 6 0 0s
Peripheral
✅ docs 18 0 1s
TOTAL 183 0 19s

Posted by scripts/upload-test-results.sh — Check Run API unavailable.

…ring

Root causes of ci-verify-local.yml always failing:
1. Check Runs API requires a GitHub App token (checks:write) — a regular PAT
   always gets 403. Replaced with Commit Status API (POST /statuses/{sha})
   which works with any repo-scoped PAT, no GitHub App needed.
2. Upload was called in /session-end before the commit existed, posting the
   status on the wrong (previous) commit SHA.
3. Even the PR comment fallback missed because CI fires on pr:opened before
   the upload had a chance to run.

Fix:
- upload-test-results.sh: POST to /statuses/{sha} instead of /check-runs
- ci-verify-local.yml: check commit status instead of Check Runs (Option 1)
- session-end SKILL.md: remove premature upload step (wrong SHA), add
  post-commit callout: run upload after push, before creating PR
- Docs (CLAUDE.md, CONTRIBUTING.md, README.md, PR template): correct
  ordering and wording throughout

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
@Soul-Craft Soul-Craft merged commit 98b7a7e into main Apr 16, 2026
6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant