Skip to content

fix: reject task-specific Hermes review candidates#37

Merged
cafitac merged 1 commit intomainfrom
fix/reject-task-specific-review-candidates
Apr 28, 2026
Merged

fix: reject task-specific Hermes review candidates#37
cafitac merged 1 commit intomainfrom
fix/reject-task-specific-review-candidates

Conversation

@cafitac
Copy link
Copy Markdown
Owner

@cafitac cafitac commented Apr 28, 2026

Summary

  • reject task-specific Hermes review constraints from becoming durable learning candidates
  • add regression coverage for Do not modify files and Do not assume prior reviews are correct
  • cover the adjacent Do not inspect or report unrelated files wording in the new rejection heuristic

Evidence

  • isolated replay of session_end-20260428_213448_79a706.json currently produced candidate-modify-files.md
  • isolated replay of session_end-20260428_173757_a990f7.json currently promoted candidate-assume-prior-reviews-are-correct.md
  • both came from review-task instructions, not reusable cross-session learning

Test Plan

  • PYTHONPATH=src .venv/bin/python -m pytest tests/test_pipeline.py::test_process_events_rejects_task_specific_review_constraint_do_not_modify_files tests/test_pipeline.py::test_process_events_rejects_task_specific_review_constraint_prior_reviews -q
  • PYTHONPATH=src .venv/bin/python -m pytest tests/test_pipeline.py::test_process_events_rejects_real_runtime_malformed_code_fragment_candidate tests/test_pipeline.py::test_process_events_rejects_task_specific_review_constraint_do_not_modify_files tests/test_pipeline.py::test_process_events_rejects_task_specific_review_constraint_prior_reviews tests/test_pipeline.py::test_extract_candidate_ignores_hermes_skill_wrapper_in_user_message -q
  • npm test
  • PYTHONPATH=src .venv/bin/python -m pytest -q

@cafitac cafitac merged commit 26dcec0 into main Apr 28, 2026
6 checks passed
@cafitac cafitac deleted the fix/reject-task-specific-review-candidates branch April 28, 2026 14:15
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant