Fix Tests Workflow

Run the test suite for the given scope, diagnose failures, fix them, and re-run. The fixed parts are the 3-attempt limit per failure, the final re-run of the original scope, and the summary; how you diagnose each failure is your call (the Step 3 checklist lists common causes in this codebase).

Step 1: Determine Scope & Mode

Parse the argument to determine what to run:

• backend or no argument → cd server && bundle exec rspec --format progress
• frontend → cd frontend && CI=true npm test
• e2e → cd frontend && npx playwright test
• A specific file path → run that file directly

Loop mode (--loop): Run the full fix cycle autonomously without asking for input. Make best-judgment fixes. Suitable for headless/unattended runs via claude -p "/fix-tests backend --loop".

Step 2: Run Tests

Execute the test command and capture output. Note all failures.

Step 3: Fix Loop (max 3 attempts per failure)

For each failing test:

1. Read the failing spec file and the source file it tests
2. Diagnose the root cause — check for:
   • Missing factory traits (check spec/factories/ and spec/factories/ai/)
   • Wrong test helpers (use user_with_permissions, auth_headers_for, json_response)
   • Missing shared examples (spec/support/shared_examples/)
   • Missing mocks or stubs for external services
   • Stale selectors in E2E tests (prefer data-testid)
   • Import path issues (use @/shared/, @/features/)
3. Fix the issue using Edit
4. Re-run just the failing file to verify the fix
5. If still failing after 3 attempts on the same test, stop and report — do not keep iterating

Step 4: TypeScript Check (frontend/e2e only)

If any TypeScript files were modified:

cd frontend && npx tsc --noEmit

Fix any type errors found (same 3-attempt limit).

Step 5: Final Run

Re-run the full suite for the original scope to confirm everything passes.

Step 6: Loop Mode Continuation (--loop only)

If --loop flag was provided and Step 5 still has failures:

1. Return to Step 3 with the NEW set of failures (not the originals)
2. Repeat the full cycle up to 3 total loop iterations
3. If failures persist after 3 full loops, stop and report

This enables autonomous fix pipelines: claude -p "/fix-tests backend --loop" --allowedTools "Bash,Read,Edit,Grep,Glob"

Step 7: Summary

Report:

• Total tests: pass / fail / pending
• Failures fixed (list each with one-line description of what was wrong)
• Failures remaining (if any, with diagnosis of why they couldn't be auto-fixed)
• Loop iterations completed (if --loop mode)