CI-esta: make GitHub Actions faster, with evidence

Read the documentation · Related tools

See what holds up your PR, avoid repeating validated work, and run the remaining tests efficiently. CI-esta adds reports, environment caching, reusable execution evidence and test scheduling to your existing GitHub Actions workflows. No account, hosted service or application-language change is required.

Choose your first improvement

Use the adoption tutorial and checklist to choose an order, verify each improvement and find every action/CLI capability. Run ci-esta adoption --root . for a read-only inventory with source-linked suggestions. It helps find gaps; successful hosted runs and measurements verify whether an integration is useful.

Start with one action

In an existing uv test job, keep checkout and your test command. Replace Python setup, uv setup and dependency installation with:

- uses: actions/checkout@v7
- uses: gigaverse-app/ci-esta/actions/setup-cached-uv@v1
  with:
    python-version: '3.12'
    sync-args: '--group test'
- run: uv run --no-sync pytest

Use your project's interpreter and dependency groups. Every action lives under actions/NAME. Use gigaverse-app/ci-esta/actions/ci-observability@v1 for reports.

Use actions/NAME@v1 for automatic Actions API upgrades. That alias is independent of Python package versions. Older .github/actions/NAME and top-level named callers must migrate to actions/NAME; each action has one implementation. Action API and versioning.

Copy the complete reporting or cache workflow. The actions set up their own tools; using them does not require adding CI-esta to your app. For Python execution commands and plugins, install ci-esta[pytest] in the test environment. Reporting and evidence actions also work with non-Python projects; Bun users have a native selection adapter.

Give your agent a concrete task

Read https://raw.githubusercontent.com/gigaverse-app/ci-esta/main/skills/ci-esta/SKILL.md
Measure a completed CI run and improve its largest avoidable cost. Use the public
action entry points and tested quickstart examples. Compare one large runner with
multiple runners if tests dominate. Keep required checks and full-suite coverage,
and show the workflow changes, test inventory, elapsed time and runner usage.

Install or update the portable skill. The skill, README and quickstart use the same recipes. Updating the Python package does not update an installed agent skill.

Results you can inspect

These are scenario-specific observations. Selection savings belong to pytest-selective and the consumer's declared inputs; they are not receipt-reuse savings. Runner-seconds are summed execution, not a billing invoice. Measure your own cold/warm setup, full/selected runs and runner layouts before claiming a gain.

How the safeguards fit

A duration estimate changes where a test runs, never whether it belongs in the plan. An environment cache accelerates setup; locked synchronization still runs. A completion receipt can justify reusing the same declared work, but an old green workflow alone cannot. Missing or contradictory proof executes the work again. Explicit reruns and scheduled full checks retain their meaning.

Input declarations, service isolation and conformance requirements belong to your project. Evidence reuse and strict gates explain these boundaries. The optimization guide connects selection, timing feedback, execution and measurement in order.

Further documentation

• Public actions and full toolkit catalog
• Critical path and slack, cost reporting, and environment caching
• Learned timings, runner pools, and Python CI tuning
• Integrations, alternatives, and revision-bound adoption audit
• Contributing, releases, and source provenance

Install the standalone CLI with pip install ci-esta. Base reporting needs Python 3.11+ and neither pytest nor pytest-selective. For fixtures, schemas, source scanners, and subprocess helpers, follow the pytest-selective sidecar tutorial. Declare inputs at their readers so unrelated documentation does not select tests.

Use ci-esta[selection] for Python selection reports or ci-esta[pytest] for pytest execution tools. Before 1.0, review each minor release's migration notes.

Test execution profiles

Separate affected work from special-occasion execution with test execution policy. The JSON planner works with pytest, Bun, Playwright and custom runners; resource resolution is optional.

New policy adopters can declare work directly in plan-test-policy without a policy file. CI-esta provides ordinary, full, weekly, smoke, performance and probe defaults. Optional configuration contains only exceptional metadata and profile overrides; existing selectors remain authoritative for work IDs.