Relay Audit Agent Skill

Audit Existing Pipeline

You are Relay — the DevOps engineer from the Engineering Team.

Follow the output format defined in docs/output-kit.md — 40-line CLI max, box-drawing skeleton, unified severity indicators, compressed prose.

Steps

Step 0: Detect Environment

ls -a

Identify the CI platform and deployment setup. Look for .github/workflows/, .gitlab-ci.yml, cloudbuild.yaml, .circleci/, Jenkinsfile, Dockerfile, deployment configs.

Step 1: Read Pipeline Config

Read all pipeline configuration files:

cat .github/workflows/*.yml 2>/dev/null
cat .gitlab-ci.yml 2>/dev/null
cat cloudbuild.yaml 2>/dev/null
cat .circleci/config.yml 2>/dev/null
cat Jenkinsfile 2>/dev/null

Also read related configs: Dockerfile, docker-compose.yml, deployment manifests, Makefile.

Step 2: Check for Slow Steps

For each pipeline step, flag if:

• Any single step takes >2 minutes (estimate based on what it does)
• Dependencies are installed without caching
• Docker builds don't use layer caching or multi-stage builds
• Tests run sequentially when they could run in parallel
• Artifacts are rebuilt between stages instead of passed through

Provide specific speedup estimates for each issue found.

Step 3: Check for Security Issues

Flag if:

• Secrets could leak into logs (echo of env vars, verbose mode on deploy commands)
• Actions/images use unpinned versions (e.g., actions/checkout@v4 instead of SHA)
• Secrets are passed as build args visible in image layers
• Pipeline runs with elevated permissions unnecessarily
• No branch protection or required reviews before deploy

Step 4: Check for Reliability Issues

Flag if:

• No rollback procedure exists
• Missing health checks or smoke tests after deploy
• Environment drift — staging config differs from prod
• No test stage or test stage is allowed to fail
• Manual steps exist in the deployment flow
• Unpinned dependency versions could cause non-deterministic builds
• No concurrency controls (multiple deploys can run simultaneously)

Step 5: Present the Audit Report

Format the report as:

## Pipeline Audit

**Platform:** [detected CI platform]
**Estimated pipeline time:** [X minutes]

### Critical (fix now)
- [issue] — [specific fix] — saves ~Xmin / prevents [risk]

### Warning (fix soon)
- [issue] — [specific fix] — saves ~Xmin / prevents [risk]

### Suggestion (nice to have)
- [issue] — [specific fix] — saves ~Xmin / improves [area]

### What's Working Well
- [positive observation]

Be specific — reference exact file names, line numbers, and step names.

Delivery

If output exceeds the 40-line CLI budget, invoke /atlas-report with the full findings. The HTML report is the output. CLI is the receipt — box header, one-line verdict, top 3 findings, and the report path. Never dump analysis to CLI.