Optimise GitHub Actions CI costs

curl -fsSL --create-dirs 'https://theskillvault.ai/api/skills/optimise-github-actions-ci-costs/raw' -o ~/'.claude/skills/optimise-github-actions-ci-costs/SKILL.md' && curl -fsSL --create-dirs 'https://theskillvault.ai/api/skills/optimise-github-actions-ci-costs/files/LICENSE.txt' -o ~/'.claude/skills/optimise-github-actions-ci-costs/LICENSE.txt'

curl -fsSL --create-dirs 'https://theskillvault.ai/api/skills/optimise-github-actions-ci-costs/raw' -o '.cursor/skills/optimise-github-actions-ci-costs/SKILL.md' && curl -fsSL --create-dirs 'https://theskillvault.ai/api/skills/optimise-github-actions-ci-costs/files/LICENSE.txt' -o '.cursor/skills/optimise-github-actions-ci-costs/LICENSE.txt'

curl -fsSL --create-dirs 'https://theskillvault.ai/api/skills/optimise-github-actions-ci-costs/raw' -o '.agents/skills/optimise-github-actions-ci-costs/SKILL.md' && curl -fsSL --create-dirs 'https://theskillvault.ai/api/skills/optimise-github-actions-ci-costs/files/LICENSE.txt' -o '.agents/skills/optimise-github-actions-ci-costs/LICENSE.txt'

{
  "mcpServers": {
    "theskillvault": {
      "url": "https://theskillvault.ai/mcp"
    }
  }
}

CI that runs every job on every push burns minutes and makes people wait. Most of the waste comes from a few causes. This skill measures first, then applies only the fixes the evidence supports.

When to use

• Your GitHub Actions runs are slow, or you hit the free minutes limit.
• Every push to a pull request queues another full run while old ones keep going.
• A docs-only change still runs the whole test matrix.

When not to use

• You have no CI yet. Use configure-github-actions-ci first.
• A check is failing. Use fix-failing-ci-checks.

Steps

1. Measure. Run gh run list --limit 20 and open the slowest runs with gh run view <run-id>. Note which jobs take longest and how often each workflow runs. Without gh, paste the workflow files and say the result is static-only.
2. Scan the workflows: rg -n "on:|concurrency:|paths:|cache:|matrix:" .github/workflows.
3. List the waste you found, and pick at most three fixes, ranked by minutes saved per day (time saved per run x runs per day):

• Dependency caching keyed on the lockfile (actions/setup-node with cache: npm, or actions/cache).
• concurrency with cancel-in-progress: true on pull request workflows.
• paths or paths-ignore filters so docs-only changes skip heavy jobs.
• A smaller matrix on pull requests, and the full matrix on main or nightly.
• Removing a workflow that runs the same checks as another.
• Splitting one long serial job into parallel jobs on the critical path.

1. Check each fix against the guardrails below. Drop any that fails.
2. Apply the fixes in one small PR, with the before numbers in the description.
3. Prove it on a non-protected branch: push twice quickly to see the cancel, and push a docs-only change to see the skip.
4. After a week, compare minutes and wait time with the before numbers. Report total runner time and PR wait time separately.

Example

on:
  pull_request:
    paths-ignore: ["**.md", "docs/**"]
concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: 22, cache: npm }
      - run: npm ci && npm test

Guardrails

• Never skip release, migration, schema or shared-library checks to save minutes.
• Keep every matrix version or platform you promise to support.
• Do not cancel in-progress runs on main or release branches; a deploy could stop halfway.
• If a path-filtered workflow is a required check, a skipped run can block merging. Use a job-level filter or a small always-run job instead.
• Keep repo YAML changes separate from org or billing settings.

Gotchas

• Never cache folders that hold secrets, tokens or .env files. Cache only dependencies and build output.
• Caches are per branch with fallback to the default branch. A cache key that never changes serves stale packages.
• Bot or formatter jobs that commit back can trigger loops. Make them opt-in with a label or manual trigger.

Portability

These techniques work in any CI. If you use GitLab CI or another tool, use its equivalent: interruptible and rules: changes in GitLab, for example.

Done when

• At most three fixes shipped, each backed by a measured waste source.
• Required checks still run where they must, and the cancel and skip were seen live.
• You have before and after numbers for runner minutes and PR wait time.

Example prompt

"Our GitHub Actions bill is too high. Audit the workflows, show me where the minutes go, and propose the top three fixes without dropping any required checks."

---

Adapted from "github-actions-efficiency" in github/awesome-copilot (MIT, Copyright GitHub, Inc.). Original: https://github.com/github/awesome-copilot/blob/61fa1fa8fcc63d6dbd2f07a75aeb8e214df5acfb/skills/github-actions-efficiency/SKILL.md. See LICENSE.txt.

Bundled files

• LICENSE.txt

Licence

Adapted from mutl3y (github/awesome-copilot, github-actions-efficiency) (Copyright GitHub, Inc.), MIT License. Adapted by TheSkillVault on 2026-11-03: rewritten as a single self-contained audit with original YAML examples; the pointers to the upstream references/ files were dropped (those files are not included); added plain-language steps, gotchas and done-when. Not the original file.

Licence: MIT License. Attribution when you reuse it: Based on work by mutl3y (github/awesome-copilot, github-actions-efficiency) (Copyright GitHub, Inc., MIT License), https://github.com/github/awesome-copilot/blob/61fa1fa8fcc63d6dbd2f07a75aeb8e214df5acfb/skills/github-actions-efficiency/SKILL.md; adapted by TheSkillVault, https://theskillvault.ai