vignesh2027 Overview
A slow CI pipeline is a productivity killer. A CI pipeline that lets bad code through is worse than no pipeline. This skill designs pipelines that are fast enough to not be skipped, and thorough enough to catch real problems.
When to Use
- When setting up CI/CD for a new project
- When the CI pipeline is slow (>10 minutes)
- When CI is consistently failing for reasons unrelated to code quality
- Before adding a new automated gate to CI
Process
Step 1: Define the pipeline stages
Standard stages:
- Fast checks (<2 min): lint, type check, compile
- Unit tests (<5 min): isolated, no network, no database
- Integration tests (<10 min): with real dependencies
- Security scan: dependency audit, SAST
- Build: Docker image, artifact
- Deploy to staging
- E2E tests against staging
- Deploy to production (gated by manual approval or automated checks)
Step 2: Optimize for speed
- Run stages in parallel where possible
- Cache aggressively: dependencies, build artifacts, Docker layers
- Fail fast: put the fastest checks first
- Target: full pipeline under 15 minutes
Step 3: Required gates before merge
Minimum required to merge a PR:
- All tests pass
- No new high/critical security vulnerabilities (dependency audit)
- Code coverage hasn't regressed (enforced threshold, not aspirational)
- Linting passes (no new lint errors)
Step 4: Required gates before deploy to production
- Staging deploy succeeded
- E2E tests on staging passed
- Performance tests within SLO (for perf-sensitive changes)
- Rollback procedure documented
Step 5: Branch protection
- Main/master branch requires PR (no direct pushes)
- PR requires CI pass + at least 1 review
- Stale approvals invalidated on new commits
Step 6: Environment variable management
- Secrets stored in CI environment variables, not code
- Different secrets per environment (dev/staging/prod never share production secrets)
- Rotate secrets on schedule, not just on breach
Step 7: Monitor the pipeline
Track: build success rate, build duration trends, flaky test rate. A flaky test that is retried silently is a reliability problem.
Verification Requirements
- Pipeline stages defined with clear purpose for each
- Full pipeline runs under 15 minutes
- Required gates enforce: tests, security, linting
- Branch protection enabled
- No secrets in pipeline code
- Flaky test rate < 1%