ci-cd-optimizer

Overview

Provides a systematic 5-step workflow to diagnose and fix slow, flaky, or costly CI/CD pipelines. Covers bottleneck identification, parallelization, smart test selection, caching hierarchy, artifact reuse, flaky test quarantine, and cost optimization techniques with concrete before/after metrics and GitHub Actions examples.

When to Use This Skill

• Pipelines take too long (developers are waiting).
• High flakiness rate.
• CI costs are rising.
• User says "make the build faster" or "fix flaky tests".

Prerequisites

• Existing CI configuration (GitHub Actions, GitLab CI, etc.).
• Access to CI logs, timing data, and cost reports.
• Test suite that can be run locally for experimentation.

Steps

1. Measure current state:
   • Record wall time for full pipeline on main.
   • Identify the slowest job(s) from timing graphs.
   • Count flaky test rate (retries needed).
2. Parallelize:
   • Split large jobs (e.g., test into shard by file or by package).
   • Use matrix strategy.
   • Run independent jobs in parallel (needs: only where required).
3. Caching hierarchy (most impactful):
   • Dependencies (node_modules, pip cache, Go modules).
   • Build cache (Next.js, Turborepo, Gradle, etc.).
   • Docker layer cache.
   • Test results / coverage cache.
4. Smart test selection:
   • Changed-files detection (only run tests for packages that changed).
   • Use tools like nx affected, Turborepo, or custom scripts.
   • Quarantine known flaky tests into a separate "flaky" job that runs less often.
5. Artifact & reuse:
   • Upload build artifacts from build job and download in deploy/test jobs instead of rebuilding.
   • Use actions/cache or native cache features.
6. Cost optimization:
   • Use larger runners only when needed.
   • Cancel in-progress workflows on new pushes to same branch.
   • Self-hosted runners for high-volume repos.
7. Output:
   • Diagnosis report template.
   • Optimized workflow diff.
   • Flaky test handling pattern.
   • Before/after timing expectations.

Examples

A before/after GitHub Actions workflow showing parallel test sharding, aggressive caching, affected tests only, and cancellation is included, along with a flaky test quarantine pattern.

Edge Cases & Error Handling

• Monorepos: Use workspace-aware tools (Nx, Turborepo, pnpm workspaces).
• Flaky tests that are actually bugs: Quarantine + create ticket, do not ignore.
• Cache invalidation bugs: Provide cache key best practices (hash of lockfile + OS).

Verification

1. Re-run the pipeline multiple times and record new average duration.
2. Flaky test rate drops (tracked in CI logs or a dashboard).
3. Cost report shows reduction (if applicable).
4. Developers report faster feedback on PRs.
5. Success: Pipeline is 30-70% faster, reliable, and cheaper while maintaining the same quality gates.

References

• GitHub Actions Caching
• Turborepo Remote Caching
• Nx Affected
• Flaky Test Management