Skip to content

v2 review pass: guard hooks, honest demo, Windows fixes (v0.2.0) - #1

Merged
zacha0dev merged 1 commit into
mainfrom
claude/v2-review-pass
Sep 26, 2026
Merged

zacha0dev merged 1 commit into
mainfrom
claude/v2-review-pass

Conversation

@zacha0dev

Copy link
Copy Markdown
Owner

Fixes the failing CI on main and the gaps a full audit found. See CHANGELOG.md 0.2.0.

Highlights

  • adds the missing tools/hooks/guard.py + .github/hooks/guard.json (CI failed without them) with 43 tests
  • validator and mirror check work on Windows; validator now checks what the README claims
  • triage demo is honest (--runxfail shows the real failure); illustrative learning example labelled as such
  • agents that write files get bounded edit access; handoff order consistent
  • slash commands for Claude Code and Copilot; README why + quickstart; llms.txt, CITATION.cff, SECURITY.md, CONTRIBUTING.md

Verified locally (Windows, Python 3.14)

  • validate_fleet: ok - 18 agents, 10 skills, 4 flows
  • sync_claude --check: ok
  • pytest: 49 passed, 1 xfailed (--runxfail: the intended 1 failure)

🤖 Generated with Claude Code

https://claude.ai/code/session_01FYwugpFSkj3pMYZQ15vrMD

- add tools/hooks/guard.py (DENY[0] force-push, [1] reset --hard,
  [2] test deselection, [3] outbound sends) for Claude Code and Copilot,
  .github/hooks/guard.json, and 43 hook tests; CI was failing without them
- settings.json: python + $CLAUDE_PROJECT_DIR paths, Bash|PowerShell matcher
- Stop hook uses decision/reason and cannot loop
- validator: pathlib + UTF-8 (Windows crash), checks hook paths, agent
  names, skill references, flow handoffs and README counts
- sync_claude: UTF-8/LF so --check passes on Windows; .gitattributes
- lead, quality-auditor, reporter, triager get edit, bounded to their files
- triage demo runs --runxfail; report numbers match the real tree
- learning example labelled illustrative; "self-improving" claims softened
- handoff order triager -> test-writer -> implementer everywhere
- /sweep /triage /reinforce for Claude Code and Copilot
- deploy/: honest runner placeholder, run-once.sh path and placeholder checks
- mcp.json: read-only GitHub endpoint, pinned playwright with named tools
- README why + quickstart; llms.txt, CITATION.cff, SECURITY.md,
  CONTRIBUTING.md; CHANGELOG fleet history

Checks: validate_fleet ok; sync --check ok; pytest 49 passed, 1 xfailed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FYwugpFSkj3pMYZQ15vrMD
@zacha0dev
zacha0dev merged commit f0b51a9 into main Sep 26, 2026
2 checks passed
@zacha0dev
zacha0dev deleted the claude/v2-review-pass branch September 26, 2026 22:29
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