self-heal

Self-Healing Mode

Run a complete diagnostic and automatic fix cycle.

Process:

1. Diagnose — Run ALL checks and collect ALL failures:

   • Run full test suite → collect failures
   • Run type checker → collect errors
   • Run linter → collect warnings/errors
   • Run build → collect errors

2. Categorize — Group issues by type:

   • Type errors
   • Test failures
   • Lint errors
   • Build errors
   • Runtime errors

3. Fix — For each category, spawn a subagent to fix issues in parallel:

   • One agent fixes type errors
   • One agent fixes test failures
   • One agent fixes lint issues
   • Coordinate to avoid conflicts

4. Verify — After all fixes, run the full diagnostic again

5. Repeat — If issues remain, repeat (max 3 cycles)

6. Report — Final status report

Output Format:

## Self-Heal Report

### Cycle 1
- Found: X type errors, Y test failures, Z lint errors
- Fixed: A type errors, B test failures, C lint errors
- Remaining: ...

### Cycle 2 (if needed)
- ...

### Final Status
- ALL CLEAR / X issues remain (need manual intervention)
- Total fixes applied: N
- Files modified: [list]

Rules:

• Max 3 cycles — if still broken after 3 cycles, stop and report to user
• NEVER change test expectations to make tests pass — fix the code instead
• ALWAYS verify each fix doesn't introduce new issues
• ALWAYS show what was changed

$ARGUMENTS