CI CD Pipeline

CI/CD pipelines (GitHub Actions + GitLab CI)

Focused on the four rules that matter in real pipelines. Syntax details live in the upstream docs — this file only captures the conventions teams get wrong by default.

Rules

1. Build once, deploy many. Produce one artifact tagged with the commit SHA, then promote it across environments. Rebuilding per environment is how staging and prod drift.
2. Idempotence. Same commit → same artifact. Use npm ci / pip install --require-hashes / cargo build --locked, pin Docker base images to an explicit version, and pin third-party actions by full 40-char SHA (not @v4, not @main).
3. Fail fast under 10 min. Run lint, unit tests, and security scans in parallel jobs. A sequential pipeline is the biggest avoidable latency in CI.
4. Separate build from deploy. Two jobs, two responsibilities. The build job never touches deploy credentials.

GOOD vs BAD

# BAD — action pinned by tag, image rebuilt per env
- uses: actions/checkout@v4
- run: docker build -t my-app:staging .    # and again for prod

# GOOD — action pinned by SHA, image built once with SHA tag
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683  # v4
- run: docker build -t my-app:${{ github.sha }} .

Secrets and credentials

• Prefer OIDC over long-lived API keys (GitHub Actions → AWS/GCP/Azure role assumption, GitLab id_tokens:).
• Scope secrets per environment: environment: staging and environment: production get different secret stores.
• Never echo $SECRET, never pass a secret as a CLI argument on a shared runner.
• The minimum-permissions default for a GitHub workflow is permissions: contents: read. Add write scopes per-job only when they're needed.

Parallelism

• GitHub Actions: use needs: to build a DAG. Jobs with needs: [] start immediately; jobs with needs: [lint, test] wait for both.
• GitLab CI: needs: turns stages into a DAG (no longer strictly sequential). parallel: runs matrix variants side-by-side.

Pipeline audit checklist

Use this when reviewing an existing workflow:

• Every third-party action pinned by 40-char SHA
• Docker base images pinned to explicit version (not :latest, not :lts)
• Top-level permissions: declared and minimal
• Secrets scoped per environment; no production secrets in build jobs
• OIDC used instead of static cloud keys where the target cloud supports it
• npm ci / pip install --require-hashes / cargo build --locked — no unlocked installs
• Lint, test, and security scan run in parallel, not sequentially
• Rollback procedure documented or scripted

Helper script

scripts/action-pin-check.sh <workflow.yml>... — scans workflows for uses: references that aren't full SHAs. Wire it into CI so SHA pinning is enforced, not aspirational.

Anti-patterns

• ❌ uses: foo/bar@v4 or @main — pin by SHA
• ❌ docker build -t my-app:staging and a separate my-app:prod build — build once, retag on promotion
• ❌ npm install in CI — use npm ci
• ❌ :latest or :lts image tags — pin to an exact version
• ❌ One giant sequential pipeline — split into parallel jobs with needs:
• ❌ Production secrets exported in the build job — scope per environment
• ❌ echo $SECRET or secrets in set -x output — they end up in logs

References

• GitHub Actions: https://docs.github.com/en/actions
• GitLab CI: https://docs.gitlab.com/ee/ci/
• OIDC for cloud auth: https://docs.github.com/en/actions/deployment/security-hardening-your-deployments/about-security-hardening-with-openid-connect

Source: felixhennequin-gif/claude-code-config-template — distributed by TomeVault.