/commit

Commit Changes

Process

1. Review Changes

git status
git diff HEAD
git diff --stat HEAD

Check for new untracked files:

git ls-files --others --exclude-standard

2. Stage Files

Add the untracked and changed files relevant to the current work.

Do NOT stage:

• .env or credential files
• Large binary files
• Files unrelated to the current task

3. Create Commit

Write an atomic commit message with a conventional commit tag:

• feat: — New capability or feature
• fix: — Bug fix
• refactor: — Code restructure without behavior change
• docs: — Documentation only
• test: — Test additions or fixes
• chore: — Build, CI, tooling changes
• perf: — Performance improvement

For monorepo changes spanning multiple packages, note the primary package in the scope:

feat(workflows): add DAG condition evaluator
fix(web): resolve SSE reconnection on navigation
refactor(isolation): simplify worktree resolution order

Commit message format:

tag(scope): concise description of what changed

[Optional body explaining WHY this change was made,
not just what changed. Include context that isn't
obvious from the diff.]

[Optional: Fixes #123, Closes #456]

4. Capture AI Context Changes

If any AI context assets were modified in this commit, add a Context: section to the commit body:

feat(orchestrator): add retry logic for session recovery

Added exponential backoff when SDK subprocess crashes mid-session.
Previously a single crash would fail the entire workflow.

Context:
- Updated .claude/rules/orchestrator.md with retry conventions
- Added .claude/commands/debug-session.md for session state inspection
- Surfaced issue: mock.module() in retry tests needs isolated batch

Fixes #482

What counts as AI context changes:

• .claude/rules/ — on-demand conventions added, updated, or removed
• .claude/commands/ — slash commands created or modified
• CLAUDE.md — global rules changes

Why this matters: Your git log is long-term memory. Future agents and sessions use git log to understand project history. If context changes aren't captured in commits, the AI layer's evolution becomes invisible — you lose the ability to trace WHY a rule exists or WHEN a command was added.