fix-loop

Autonomous Fix Loop

Iteratively fix failing tests until the full suite passes, with a hard cap to prevent infinite loops.

Process

1. Run Full Suite

pytest --tb=short -q

Capture the output. If all tests pass, stop - nothing to do.

2. Triage Failures

Before fixing anything, group and prioritize:

1. Group by error type/module - failures with the same stack trace or in the same module likely share a root cause
2. Identify cascading failures - if module A is broken, tests in modules B and C that import A will also fail. Fix root causes first.
3. Identify bad tests - a test is wrong if: it asserts behavior that contradicts documented requirements, was written against a previous API version that intentionally changed, has hardcoded values that no longer match current schema, or the "fix" to pass it would break other tests. Check git blame to see if the test or its source changed more recently.
4. Fix root causes first, re-run, and see how many failures disappear before addressing the next group.

3. Fix Each Failure Group

For each failure group from triage:

1. Read the test file and the source file it tests
2. Diagnose the root cause - bug in source, bad test, or cascading from another failure?
3. Fix the source code (not the test) unless the test itself is wrong
4. Re-run that specific test to confirm the fix: pytest path/to/test.py::test_name --tb=short
5. If it still fails, re-diagnose and try again (max 3 attempts per test)
6. If 3 attempts fail, flag as UNRESOLVED in FIX_LOG.md with diagnosis

4. Re-run Full Suite

After fixing all groups, run the full suite again. New failures may have appeared from interactions between fixes.

5. Loop

Repeat steps 1-4 until either:

• All tests pass
• 10 full iterations completed (hard cap)
• Two consecutive iterations produce no change in failing tests (stalled - exit early)

6. Write FIX_LOG.md

After each iteration, append to FIX_LOG.md in the project root (use the Write tool, not heredocs):

## Iteration N - [timestamp]
- Tests passing: X/Y
- Fixed: [list of tests fixed and what was wrong]
- Remaining failures: [list with brief diagnosis]
- Files changed: [list]

At hard cap or early exit, add a Final Summary:

## Final Summary
- Total tests fixed: X
- Total iterations: N
- Remaining failures: [list with full diagnosis, whether present since iteration 1, suspected root cause]
- Recommended next steps: [e.g., "These appear to be bad tests" or "Requires architectural changes"]

Rules

• Fix source code, not tests - unless the test is genuinely wrong (see triage criteria above)
• During the fix loop, operate autonomously without asking for input. If truly ambiguous, log it in FIX_LOG.md and move on.
• Only parallelize fixes that affect different source files. If multiple failures trace to the same module, fix them sequentially.
• Stop at 10 iterations or when progress stalls (no change in 2 consecutive iterations)
• Never introduce new failures - if a fix breaks other tests, revert it and try a different approach
• Commit after each iteration that reduces failing test count - preserves progress even if suite never goes fully green. Follow project commit message format.

Quick Reference

Step	Command	Stop if
Full suite	pytest --tb=short -q	All pass
Single test	pytest path::test -v	Passes
Check coverage	pytest --tb=short -q after fixes	New failures appear
Hard cap	Count iterations	10 reached
Stall detection	Compare failing tests	No change in 2 iterations