Agent-Skills.md

Agent Skills: Rules Compliance Review

Use after code changes, before final answers, or whenever rules/*.md compliance is uncertain. Runs deterministic rules-enforce first, then performs a semantic audit of codex/rules/*.md against the current diff. Do not use as a replacement for hooks; use it as the human/model-level second pass.

UncategorizedID: TakumiOkayasu/dotfile-work/rules-compliance-review

Repository

TakumiOkayasu/dotfile-work
TakumiOkayasu

Install this agent skill to your local

pnpm dlx add-skill https://github.com/TakumiOkayasu/dotfile-work/tree/HEAD/codex/skills/rules-compliance-review

Skill Files

Browse the full folder contents for rules-compliance-review.

Download Skill

Loading file tree…

codex/skills/rules-compliance-review/SKILL.md

Skill Metadata

Name
rules-compliance-review
Description
Use after code changes, before final answers, or whenever rules/*.md compliance is uncertain. Runs deterministic rules-enforce first, then performs a semantic audit of codex/rules/*.md against the current diff. Do not use as a replacement for hooks; use it as the human/model-level second pass.

Rules Compliance Review

This skill verifies that current changes comply with all active codex/rules/*.md markdown rules.

Mandatory sequence

  1. Run deterministic scanner first:
codex/hooks/rules-enforce.sh --report

If the repo is installed through plugin only, use the plugin copy when available:

~/.codex/plugins/dotfile-work-codex/hooks/rules-enforce.sh --report
  1. Read applicable rules:
codex-rules read
  1. Inspect the current diff:
git diff HEAD --
  1. Semantic audit checklist:
  • coding-conventions.md: strict equality, no boolean explicit comparison, guard clauses, small functions, no any, async/await, no empty/broad catch, logger usage, concrete test assertions.
  • implementation-policy.md: no unnecessary custom implementation, no unapproved dependencies, no latest/wildcard versions, ORM/migration/logger/validation/crypto/http-client boundaries.
  • hierarchical-architecture.md: dependency direction, no lateral layer access, no layer skipping, composition over inheritance, interface responsibility, Intent over Raw Input.
  • hallucination-prevention.md: no invented APIs/options/env/schema/test results, uncertainty marked with [要確認: reason].
  • RULES_CORE.md: no unread edits, no user-diff overwrite, no destructive/external write without approval, honest verification report.
  1. Output only one of these:
## Rules Compliance: PASS
- Checked: <rules>
- Deterministic scan: pass
- Semantic audit: pass
- Remaining risk: <none or concise>

## Rules Compliance: BLOCK
1. <file:line> - <rule> - <violation>
   Fix: <specific change>

Subagent policy

For large diffs, high-risk changes, or review tasks, ask one focused subagent to audit rules compliance only:

  • Scope: current diff + active rules/*.md
  • Prohibit: feature suggestions, style-only comments not backed by rules, implementation changes
  • Output: PASS or BLOCK with file:line, rule name, evidence, and fix

Do not rely on the subagent alone. Parent session must run rules-enforce.sh and make the final decision.