/check-approach

Check if your planned approach matches any past failures:

Arguments: $ARGUMENTS

Core Principle

Learn from mistakes. Do not repeat approaches that failed before.

Phase 1: Parse Approach

1. Extract approach description from arguments
2. Identify key elements:
   • Files involved
   • Technique/pattern being used
   • Problem being solved

Phase 2: Query Failure Memory

3. Read .claude-harness/memory/procedural/failures.json
4. For each failure entry:
   • Calculate similarity score based on:
     • File overlap (same files affected)
     • Technique similarity (same approach)
     • Problem similarity (same type of issue)
   • If similarity > 0.7, flag as potential match

Phase 3: Report Matches

5. If matches found:

   SIMILAR APPROACH FAILED BEFORE
   
   Failure: {failure description}
   When: {timestamp}
   Files: {affected files}
   Error: {error messages}
   Root Cause: {why it failed}
   
   Prevention Tip: {how to avoid}
   

Phase 4: Suggest Alternatives

6. Read .claude-harness/memory/procedural/successes.json
7. Find successful approaches for similar problems
8. Report alternatives:

   SUCCESSFUL ALTERNATIVE
   
   Approach: {description}
   When: {timestamp}
   Files: {files}
   Why it worked: {rationale}
   

Phase 5: Recommendation

9. If high-similarity failure found:

   • Recommend NOT proceeding with current approach
   • Suggest specific alternative

10. If no matches:

    • "No similar failures found. Proceed with caution."
    • Still recommend running tests frequently