一键导入
ci-cd-audit
Audit CI/CD workflows for correctness, caching, action pinning, secret hygiene, and job-graph health. Returns impact-ordered findings tables.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Audit CI/CD workflows for correctness, caching, action pinning, secret hygiene, and job-graph health. Returns impact-ordered findings tables.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Audit agent configuration surfaces (AGENTS.md, CLAUDE.md, MCP servers, tool registrations, settings, hooks) for consistency, contradictions, and silent overrides. Returns impact-ordered findings tables.
Audit an agent's main loop for halting conditions, mutation budgets, loop breakers, sensitive-path guards, error recovery, and replayability. Returns impact-ordered findings tables.
Review a database schema migration for production safety: lock contention, backfill strategy, online vs offline, rollback path, foreign-key timing, replication lag. Returns impact-ordered findings tables.
Production readiness audit for a service: health checks, rollback strategy, runbook presence, observability wiring, capacity, dependencies. Returns impact-ordered findings tables.
Audit an AI eval suite for coverage, leakage between datasets, deterministic vs stochastic measurement, ladder correctness, and signal quality. Returns impact-ordered findings tables.
Generate or audit an incident runbook for an alert/service from code, dashboards, and alert rules. Output is a runbook plus a findings table of gaps. Findings follow the shared output contract.
| name | ci-cd-audit |
| description | Audit CI/CD workflows for correctness, caching, action pinning, secret hygiene, and job-graph health. Returns impact-ordered findings tables. |
Use this skill when the task is to review a repository's CI/CD configuration as a whole — workflow files, job graphs, caching, secrets usage, and reusable-action pinning. For pure security review of GitHub Actions, prefer gh-actions-security. For the release-publishing path specifically, prefer release-pipeline-audit.
.github/workflows/ (and circleci, gitlab-ci.yml, bitbucket-pipelines.yml, azure-pipelines.yml if present). For each, record: triggers, jobs, runners, timeout, concurrency rules.needs:), where the parallelism is, and where the critical path is.@v3, @main) instead of by full commit SHA. First-party actions (actions/checkout, actions/setup-*) may use major-version tags if that's the repo convention — note the convention.if: conditions, env: at workflow level that leak to forked PRs, or ${{ secrets.* }} in run: steps where the secret ends up in process args. Defer hard security findings to gh-actions-security.pull_request vs pull_request_target — flag pull_request_target runs that check out untrusted code. push triggers on tags vs branches. workflow_dispatch inputs and their defaults.ubuntu-latest pinning where reproducibility matters.concurrency: groups with cancel-in-progress set correctly so deploys don't race and PR runs don't pile up.timeout-minutes set. Flag jobs without one.actions/*) from third-party actions before flagging pin-by-tagubuntu-latest unless the workflow has reproducibility requirements (releases, security scans)if: github.repository == ... when auditing fork behavior — they don't run on forksuses: ./.github/workflows/... or uses: org/repo/.github/workflows/...), recurse into itFindings follow docs/skills/OUTPUT_CONTRACT.md.
Return:
ready / ready with caveats / not ready + a small table:
Workflows scanned, Jobs, Third-party actions, Self-hosted runners?@v3.2.1, @a1b2c3d, or @main) for action pinning findings