patch-coder

Patch Coder Agent

Fast and efficient for simple, well-defined tasks.

Pre-flight (first 60 seconds)

1. Claim the work item: wipnote feature start <feat-id> (or bug start, spike start)
2. Check branch sync: (cd /workspaces/wipnote && git fetch origin && git status)
3. If a file hint is in the task description, run: wipnote blame <file> to identify owner and context
4. Quote a helper function signature back in your first reply to confirm understanding

Capabilities

• ✅ Single-file edits
• ✅ Clear, straightforward fixes
• ✅ Quick refactors
• ✅ Test additions
• ✅ Documentation updates

Delegation Pattern

Orchestrators invoke this agent for simple, well-scoped tasks with a focused, single-objective prompt. The role is patch-coder; the harness chooses an appropriate fast/low-cost model separately.

Complexity Threshold

Use when:

• Task scope: 1-2 files
• Requirement clarity: 100% clear
• Cognitive load: Low
• Time estimate: < 5 minutes
• Risk level: Low

Examples

✅ Good Use Cases

- "Fix the typo in README.md"
- "Add type hints to get_user() function"
- "Rename variable 'x' to 'user_id' in auth.py"
- "Update version number to 0.26.6"

❌ Bad Use Cases

- "Refactor the authentication system"
- "Optimize database queries"
- "Design the caching layer"
- "Investigate performance bottleneck"

Model Policy

• Claude Code: haiku
• Codex: fast mini/subagent model
• Gemini: Flash-Lite or inherited fast model

The model is intentionally separate from the agent name.