Use when Codex is asked to perform or finish a full codebase audit, architecture audit, technical-debt audit, maintainability audit, security audit, over-engineering/de-bloat audit, or audit of an AI-assisted or AI-generated codebase. Focus on traditional engineering risks plus AI-agent failure modes such as local-optimum patches, duplicated abstractions, over-mocked tests, fragile fallbacks, unnecessary dependencies, and unreviewed generated logic. After an audit is complete, any commit that records the audit result must include the commit-message trailer `Maintenance-Audit: true`.
Review one pull request, commit range, or uncommitted diff, especially AI- or agent-authored code changes. Use for requested code reviews, merge-readiness checks, external PR review feedback, and small/medium/large change reviews where Codex should evaluate correctness, maintainability, tests, security, CI integrity, and AI-aware risks; not for full periodic codebase audits.
Use after a difficult or long debugging session when the user asks to turn .agent-work/debug records, resolved investigations, failed attempts, git history, or version-specific framework/host behavior into reusable docs/engineering-notes. Use when Codex must separate transferable engineering lessons from project-specific implementation notes, document official-doc gaps or contradictions with observed behavior, and make notes sufficient for someone to reproduce a similar capability or diagnose a similar issue without reading the original code. Do not use for actively debugging a live bug unless the task is specifically to archive or synthesize debug knowledge.
Send signed Feishu custom-bot text notifications for Codex task starts, periodic progress, and completion using a bundled reporter template. Use only when the user explicitly asks to use $feishu-message-sender, feishu-message-sender, or this Feishu notification skill to send them messages.
Use when a previous bug fix attempt did not resolve the issue, the user says the bug still happens, a test still fails after one repair round, Codex is starting a second or later debugging pass, or a debugging session is long/repetitive enough to need persistent state across context compaction. Also use when the user asks for systematic debugging, root-cause debugging, or durable debug notes in .agent-work/debug/. Do not use for the first ordinary bug-fix attempt unless explicitly requested or the session is already becoming long-running.
Use only when the user explicitly asks to create or use an isolated git worktree. Coordinate active-agent scope files under `.worktrees/current-agents/`, set up a `.worktrees/` workspace, verify existing isolation first, ensure `.worktrees/` is ignored before creation, and report whether the ignore rule was committed. Do not use automatically for ordinary implementation, debugging, or planning.
Use only when the user explicitly asks to brainstorm, clarify, design, or spec a change before implementation, or explicitly names this skill. Do not use automatically for ordinary build, fix, implementation, or code-edit requests.
Initialize or retrofit a repository with Codex-native workflow files. Use when the user asks to set up, migrate, bootstrap, or dry-run project-level Codex configuration, including AGENTS.md, .codex/config.toml, repo skills in .agents/skills, and optional Linear/Symphony WORKFLOW.md task orchestration.