Skip to content

Conversation

@victormartingil
Copy link
Member

Summary

  • Thinking feedback: Spinner now evolves through contextual messages while the LLM processes (Thinking → Analyzing → Planning → Preparing → elapsed timer), so users know it's not stuck
  • Git tools cwd fix: Fixed simpleGit to use { baseDir } object form and added explicit cwd: process.cwd() to all execSync calls across git-simple.ts and git-enhanced.ts — fixes "Not in a git repository" error on global npm installs
  • Git tools for subagents: Added read-only git tools (status, diff, log, branch) to all agent types so agents can review PRs and analyze repos
  • authorize_path tool: New tool lets the LLM proactively request directory access via interactive user prompt, instead of telling users to manually run /allow-path
  • UX improvements: Project folder name highlighted in welcome path display, .tape/.gif added to .gitignore
  • Version bump: 1.2.2 → 1.2.3

Test plan

  • pnpm typecheck — no errors
  • pnpm lint — 0 errors (1 pre-existing warning)
  • pnpm test — 172 test files, 4360 tests passed
  • Manual: run pnpm dev, ask a complex question, verify spinner evolves through messages
  • Manual: run pnpm dev, ask "review the PR of this branch vs main", verify git tools work
  • Manual: install globally via npm, run in a git repo, verify git status works
  • Manual: ask LLM to access a path outside project, verify authorize_path tool triggers prompt

- Add evolving spinner messages during LLM thinking (Thinking → Analyzing → Planning → Preparing → elapsed time)
- Fix git tools cwd bug for global npm installs (simpleGit object form, explicit cwd in execSync)
- Add read-only git tools (status, diff, log, branch) to all subagent types
- Create authorize_path tool so LLM can request path access conversationally
- Highlight project folder name in welcome screen path display
- Update system prompt to use authorize_path instead of manual /allow-path
- Add .tape/.gif to .gitignore
- Bump version to 1.2.3
…coverage checks

- Add excludePaths to Pattern interface; skip console.log detection for CLI/REPL files
- Make checkTestCoverage() verify test files on disk before flagging missing tests
- Downgrade "no test" to info severity when test file exists but wasn't in the diff
- Add no-AI-attribution guideline to system prompt
- Replace console.log+chalk rendering with markdown builder + renderMarkdown()
- Review output now uses formatted tables (marked-terminal box-drawing)
- Findings shown as scannable table rows: File | Severity | Category | Issue
- Output also returned as markdown string in skill result for LLM consumption
- Add JSDoc to exported analyzePatterns function
- Skip test-coverage finding entirely when test file exists on disk and
  source changes are < 15 additions (refactors/fixes covered by existing tests)
- Only emit info-level nudge for large changes (>= 15 additions) with existing tests
- Fix oxlint no-control-regex warning in renderer.ts (intentional ANSI escape match)
- Linter now fully clean: 0 warnings, 0 errors
@victormartingil victormartingil merged commit fbf02e6 into main Feb 10, 2026
6 checks passed
@victormartingil victormartingil deleted the feat/v1.2.3-thinking-feedback-git-fixes branch February 10, 2026 13:12
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