How to use the GitHub Actions Efficiency Auditor agent skill

What is an agent skill, and what does this one do?

An agent skill is a folder of instructions and reference files that an AI agent loads when a task matches it. This skill is named github-actions-efficiency and is licensed under MIT.

When installed, the agent inspects a repository's GitHub Actions workflow files in .github/workflows/ and, where it has access, recent run history. It looks for missing dependency caches, missing concurrency cancellation, over-broad triggers, duplicate workflow coverage, and expensive jobs that run on every change regardless of scope.

• Proposes up to three fixes, ranked by estimated daily CI minutes saved.
• Checks each fix against guardrails so required tests, schema, migration or release checks are not removed.
• Lists what was verified live and what was checked only from files.

Install the skill from the ZIP download

Download the skill ZIP from this page and unzip it. Place the resulting folder in one of two locations, depending on how you want to use it.

For a personal install available across your projects, put it in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder so it applies only there and can be shared with the team through the repository.

If you use the Claude apps instead, you can upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/github-actions-efficiency/
• Project: .claude/skills/github-actions-efficiency/ inside your repository

Trigger the skill with natural requests

You do not need a special command. Describe the goal in plain language and mention GitHub Actions or your workflows. The skill is meant to activate when you want to reduce runtime, CI cost or wasted runs, or when you ask about caching, concurrency, path filters, matrix reduction or job optimization.

The skill also covers creating a first workflow or CI baseline for a repository that has none.

• "Audit our GitHub Actions workflows and tell me where we waste CI minutes."
• "Our PR checks take 25 minutes. Find the fixes with the biggest daily savings."
• "Add concurrency cancellation to our CI so stale runs stop."
• "Set up a minimal GitHub Actions baseline for this new repo."

What the agent does during an audit

The skill follows a fixed sequence. First it measures, searching the workflow files for triggers, concurrency, path filters, strategy matrices and caches, and, if gh is available, listing recent runs and reading failed logs. Then it applies five guardrails to every candidate fix, for example dropping any fix that removes release, schema, migration or shared-library checks, and flagging formatter or bot jobs that run automatically so they can use an opt-in trigger instead.

Next it keeps only fixes supported by evidence and passing the guardrails, up to three, ranked by per-run savings multiplied by runs per day. Finally it verifies, using a live test push on a non-protected branch when gh is available, or stating clearly that validation was local only.

Know what the skill needs

The skill needs read access to the repository's .github/workflows files. The GitHub CLI (gh) is optional but recommended, because it provides live run history and validation. Live validation also requires a non-protected branch for test pushes.

Without shell or gh access, you can paste the contents of .github/workflows/ and the output of gh run list --limit 10 into the chat. If only partial files are provided, the agent is designed to say so and begin its answer with a static-only analysis notice, since its findings may be incomplete.

• Repository with GitHub Actions workflows (it has little to audit otherwise)
• Read access to .github/workflows/
• GitHub CLI (gh), optional but recommended
• A non-protected branch for live test pushes

Safety notes: scripts and network access

This skill runs scripts and uses the network. Its measurement and validation steps run shell commands such as rg and gh, and live validation pushes a test change to a non-protected branch. Review the commands before letting the agent run them, and do not point it at a protected branch.

Do not treat an audit as proof of correctness. The skill itself warns that fixes to CI can silently remove checks if applied without testing, so review every proposed change and confirm required checks still run.

• Keep test pushes on a non-protected branch.
• Read proposed YAML diffs before committing them.
• Confirm that required tests and release checks still run after any change.

Tips for better results

Give the agent the context it cannot see in the files: which checks are required for merging, which release steps matter, and which runtime versions you support. The skill's own guidance is to define a baseline before optimizing, and it treats matrix legs without an explicit version or platform commitment as candidates for removal.

If you want before-and-after numbers, collect run data before the change. The skill separates expected savings from measured savings and warns that caches warming up, changed matrix breadth or unusually small pull requests can distort comparisons. Ask for the report explicitly if you need that kind of summary.

• Name your required checks and supported versions up front.
• Ask for a before/after report and gather a baseline sample first.
• Keep PR wall-clock time and total runner time as separate numbers.