Skip to content

ci-systems.md

Latest commit

 

History

History
66 lines (56 loc) · 2.97 KB

ci-systems.md

File metadata and controls

66 lines (56 loc) · 2.97 KB

Supported CI systems

crag analyze parses gate commands from the following 12 CI formats. Each extractor handles inline scalars, block scalars (run: | / run: >-), list forms, and the system's equivalents. Output is normalized to drop shell plumbing, dedupe matrix expansions, and filter background processes.

System File(s)
GitHub Actions .github/workflows/*.yml (recursive)
Forgejo / Gitea Actions .forgejo/workflows/*.yml, .gitea/workflows/*.yml (recursive)
GitLab CI .gitlab-ci.yml
CircleCI .circleci/config.yml
Travis CI .travis.yml
Azure Pipelines azure-pipelines.yml, .azure-pipelines/**/*.yml
Buildkite .buildkite/pipeline.yml
Drone / Woodpecker .drone.yml, .woodpecker.yml, .woodpecker/*.yml
Bitbucket Pipelines bitbucket-pipelines.yml
Jenkins Jenkinsfile, jenkins/Jenkinsfile, ci/Jenkinsfile (declarative + scripted)
Cirrus CI .cirrus.yml (*_script: keys)
Ad-hoc shell CI ci/*.sh, .ci/*.sh, scripts/ci-*.sh (canonical names only)

Normalization pipeline

Every CI extractor feeds into the same normalizer in src/analyze/normalize.js. The normalizer:

  1. Canonicalizes matrix tokens (${{ matrix.X }} → <matrix>, ${{ env.X }} → <env>, ${{ ... }} → <expr>)
  2. Strips YAML-style wrapping quotes iteratively (handles nested wrappers like "'npm test'")
  3. Decomposes compound commands on && and ;
  4. Filters ~40 noise patterns including shell builtins (break, exit, continue, shift, trap), version probes (node --version, which node, shellcheck --version), variable assignments (shell-style and embedded-code), subshell fragments, cross-language prints (puts, print, console.log, die), and curl | bash installers
  5. Deduplicates across workflows (same command in 5 files → reported once)
  6. Caps the output at 8 canonical gates per CI system to prevent matrix explosion

The result is the same gate list regardless of which CI system the project uses. crag diff compares against this normalized list so drift detection is CI-agnostic.

Task runners

Task runners are mined independently from the CI layer:

Runner File(s) What we extract
Make Makefile, GNUmakefile, makefile .PHONY directives + column-0 targets matching GATE_TARGET_NAMES
Task Taskfile.yml, Taskfile.yaml tasks: sub-keys
just justfile Recipe names at column 0

Canonical gate target names recognised: test, tests, spec, specs, check, checks, ci, unit, unittest, unit-test, integration, integration-test, itest, e2e, e2e-test, kselftest, selftest, sanity, smoke, regress, regression, lint, lints, format, format-check, fmt, fmt-check, format-fix, formatting, linting, build, compile, compile-check, typecheck, type-check, type_check, types, tc, verify, validate, audit, security, sec, sast.