| name | harness-maintenance |
| description | Internal Harness instruction source for harness-maintenance. Route through visible Harness aliases or hook contracts instead of invoking directly. |
Harness Maintenance
References
Read these first:
../../../AGENTS.md../../../CLAUDE.md../../../tooling/codex_hooks/README.md../../../schemas/skill_contracts.json../../../schemas/skill_contracts.schema.json../../../tooling/.tests/test_codex_hooks_contracts.py../../../.agents/references/code-style.md../../../.agents/references/commit-checkpoint-rule.md../../../.agents/references/language-policy.md../../../.agents/references/ubiquitous-language.md when workflow terms or operator-facing docs change../../../tooling/codex_hooks/Lightweight_Hook_Policy_Guide.md when hook policy behavior is involved and the file exists
When To Use
Use this skill for Harness framework maintenance:
- Codex hook runtime and hook installation/status scripts
- evidence tooling guardrails, schema validation, and generated-artifact schemas
- skill contracts, trigger detection, and routing behavior
.agents/skills/** and .claude/skills/** alignment.agents/references/**, .claude/Workflow_Guide.md, .claude/rules/**, .claude/shared/**, and bootstrap templates/** alignment- permission boundaries, trust/status checks, and external review wrapper policy
- ubiquitous language, operator handbook, and generated Stage Cards
- tests that validate hook, contract, skill routing, or permission behavior
Do not use this skill for ordinary research project implementation under src/, scripts/, or configs/; route that through $code-debug or $code-expert.
Required Work
- Re-read the affected hook, contract, skill, schema, and test files before editing.
- Keep
.agents/ and .claude/ behavior semantically aligned when routing or skill rules change.
- Keep hook scripts as runtime guardrails, not proof that a research gate passed.
- Preserve fail-fast behavior and avoid broad fallbacks.
- Use precise Harness terminology. Do not use bare
Evidence when
Conclusion Evidence, Evidence Chain, Gate Evidence, or
Execution Evidence is meant.
- Update focused tests for hook, contract, detection, generator, or permission behavior changes.
- Use the
guardrail commit checkpoint profile for completed hook, contract,
schema, routing, permission, generator, or supervisor-policy slices.
- Run
python tooling/codex_hooks/check_contracts.py --workspace-root . after contract or hook changes.
- Run
pytest tooling/.tests/test_codex_hooks_contracts.py after hook detection, policy, generator, or status changes.
- Run
python -m py_compile <modified python files> and ruff check --select=E,F,I <modified python files> after Python edits.
- Do not hand-edit
.evidence/** or .auto_iterate/**; use owning tools or controller.
- Report a Gate ledger for hook, skill contract, routing, permission, generator, or trust behavior changes.
Codex Adaptation
Treat natural-language requests about hook behavior, skill trigger/routing, contract scope, permissions, or Harness framework policy as $harness-maintenance work. Keep final replies in the user's language while preserving paths, commands, schema keys, and workflow IDs in English.
