ci-cd-automation
Pipeline setup/maintenance standards, CI best practices, Shift Left principle. Broader than toolchain-fallback (toolchain detection); this covers the full CI/CD automation pipeline. Write pipelines be
SKILL.mdskills/ci-cd-automation/SKILL.md
GitHubCI/CD Automation
Problem it solves
Without structured CI/CD practices, pipelines become unreliable and slow:
- Tests pass locally but fail in CI (environment drift)
- Pipeline configs drift from project requirements
- Long feedback cycles — developers wait too long for CI results
- Flaky tests ignored or disabled instead of fixed
- Security checks added as afterthoughts
- No automated deployment process — manual steps that cause errors
Protocol
1. Shift Left — test early, test often
Run quality checks as early as possible in the pipeline:
- Pre-commit: Lint, format, type-check (fast, <1 min)
- Pre-push: Unit tests, security scan (medium, <5 min)
- CI: Integration tests, e2e tests, performance benchmarks (longer)
- CD: Deploy to staging, smoke tests, deploy to production
A bug found at pre-commit costs seconds. Same bug found in production costs hours.
2. Pipeline structure
Every project should have a CI pipeline that runs on every push:
# Minimum CI pipeline
jobs:
lint: # linting and formatting
typecheck: # type checking
test: # unit + integration tests
build: # verify it builds
Optional but recommended:
security: # dependency audit, SAST
benchmark: # performance regression check
e2e: # end-to-end tests
3. Pipeline reliability
- Deterministic: Same code → same result every time
- Idempotent: Running twice produces the same result
- Fast: Keep under 10 minutes. If longer, split into parallel jobs.
- Isolated: Each job runs in a clean environment
- Reproducible: Pin dependency versions, use lockfiles, use CI cache
4. Fail fast
- If lint fails, don't run tests
- If unit tests fail, don't run e2e
- If staging smoke tests fail, don't deploy to production
- Each step should gate the next
5. CI/CD is code
- Pipeline configs belong in version control
- Review pipeline changes like code changes
- Test pipeline changes in a branch before merging to main
- Document pipeline structure and what each job does
6. Secret management
- Never hardcode secrets in pipeline configs
- Use CI/CD platform secret store (GitHub Secrets, GitLab CI variables, etc.)
- Rotate secrets regularly
- Audit who has access to secrets
Detection triggers
Activate when:
- Setting up a new project
- Adding or modifying CI/CD configuration
- Pipeline is failing or flaky
- User says "setup CI", "add pipeline", "configure deployment"
- User reports tests passing locally but failing in CI
When NOT to use
- Prototyping or throwaway projects
- User explicitly says "no CI needed"
Cross-references
toolchain-fallback — Use together to detect available toolchains before writing pipeline configs. toolchain-fallback detects tools; this skill wires them into pipelines.
release-changelog — CI/CD pipelines should trigger release workflows (version bump, tag, publish). release-changelog defines the release process; CI/CD automates it.
git-workflow-conventional-commits — CI should enforce conventional commit format and branch naming conventions.
test-driven-development — CI runs the tests. TDD ensures tests exist for CI to run.
code-review — CI checks should pass before review. Review should check CI config changes.
project-scripts — Use to discover existing build/test/deploy scripts before writing the pipeline.
security-best-practices — Integrate security scanning into the pipeline (dependency audit, SAST, secrets scan).
anti-premature-termination — CI passing is a completion criterion. Do not declare done until pipeline is green.