Pr Workflow

Pull Request Workflow

Complete workflow for creating pull requests following project standards.

Process

1. Verify staged changes exist with git diff --cached --name-only

2. Branch setup

   • If on main/master, create feature branch first: feature/brief-description or fix/brief-description
   • Use github-dev:commit-creator subagent to handle staged changes if needed

3. Documentation check

   • Update README.md or docs based on changes compared to target branch
   • For config/API changes, use mcp__tavily__tavily_search to verify info and include sources

4. Analyze all commits

   • Use git diff <base-branch>...HEAD to review complete changeset
   • PR message must describe all commits, not just latest
   • Focus on what changed from reviewer perspective

5. Create PR

   • Use /pr-creator agent or gh pr create with parameters:
     • -t (title): Start with capital letter, use verb, NO "fix:" or "feat:" prefix
     • -b (body): Brief summary + bullet points with inline markdown links
     • -a @me (self-assign)
     • -r <reviewer>: Find via gh pr list --repo <owner>/<repo> --author @me --limit 5

6. PR Body Guidelines

   • Summary: Few words or 1 sentence describing changes
   • Changes: Bullet points with inline links [src/auth.py:42](src/auth.py#L42)
   • Examples: For significant changes, include before/after code examples
   • No test plans: Never mention test procedures in PR

Examples

With inline source links:

Update Claude Haiku to version 4.5

- Model ID: claude-3-haiku-20240307 → claude-haiku-4-5-20251001 ([source](https://docs.anthropic.com/en/docs/about-claude/models/overview))
- Pricing: $0.80/$4.00 → $1.00/$5.00 per MTok ([source](https://docs.anthropic.com/en/docs/about-claude/pricing))
- Max output: 4,096 → 64,000 tokens ([source](https://docs.anthropic.com/en/docs/about-claude/models/overview))

With code changes:

Refactor authentication to use async context manager

- Replace synchronous auth flow with async/await pattern in [src/auth.py:15-42](src/auth.py#L15-L42)
- Add context manager support for automatic cleanup

Before:
\`\`\`python
def authenticate(token):
    session = create_session(token)
    return session
\`\`\`

After:
\`\`\`python
async def authenticate(token):
    async with create_session(token) as session:
        return session
\`\`\`