Handbook

Priority contracts for audits

Use this page when sequencing audit_contract_docs.py --strict work—confirm each task_id still exists in task-catalog.md before treating it as authoritative.

Updated

Curate these task_id values first when deepening contract.md, contract.json, tests, and catalog maturity rows.

Pack list (Playwright + generic + RAG)

task_id Notes
pw_chunk_classify Representative mechanics classifier
pw_page_kind_route Routing surface
pw_quiz_mechanics_discover Quiz flows
pw_static_mcq_mechanics_discover Static MCQ discovery
pw_mechanics_repair Repair loop integration
answer_from_evidence Evidence-grounded answers
rag_query_plan Retrieval planning gate
extract_schema_from_text JSON extraction baseline
llm_boolean_gate Binary structured gate
llm_enum_route Enum routing
llm_multi_label Multi-label scoring
word_problem_to_calc_plan Math planning
plan_decision_pack Decision pack emission
automation_situation_arbitrate Governed lmeta decision break-in (tri-state, closed action set)

Board-game placeholders (external prompts)

These names appear in some external prompt packs but do not exist as IDs here:

  • game_state_eval → use board_game_* / game_move_* contracts instead.
  • game_move_select → use board_game_choose_move or game_move_rank (closest intent).
  • game_rules_extract → no dedicated task; capture rules via extract_schema_from_text + consumer prose or skip with rationale.

Audit helpers

python3 scripts/audit_contract_docs.py --strict --strict-md --require-contract-identity
python3 scripts/audit_contract_docs.py --strict-md-only pw_chunk_classify,pw_page_kind_route,pw_quiz_mechanics_discover,pw_static_mcq_mechanics_discover,pw_mechanics_repair,answer_from_evidence,rag_query_plan,extract_schema_from_text,llm_boolean_gate,llm_enum_route,llm_multi_label,word_problem_to_calc_plan,plan_decision_pack

See also