crclabs-hq/GateTest Quality Gate

120-module AI-powered quality gate with self-healing CI

Trust Signals

Scorecard Score: not yet scored
Maintenance Recency: Activelast commit Jul 10, 2026
License: MIT

Pinned Snippet

workflow.ymlSHA-pinned

uses: crclabs-hq/gatetest@ddc2c3f7f28f9e9a01e880e66b828111d106f297 # v1.1.3

tags can be moved; commit SHAs can't. why a SHA?

Inputs

name	description	required	default
suite	Which GateTest suite to run: quick (41 modules) / standard (45) / full (88) / nuclear (95). Unknown values fall back to standard.	no	quick
diff-only	On pull_request events, scan only the files the PR changes (vs the base branch). Keeps PRs from being flooded with pre-existing repo-wide findings. Set to false to scan the whole repo on every event. Non-PR events always scan the full repo.	no	true
auto-fix	When true and the gate blocks, run the AI CI-fixer to propose a patch (requires ANTHROPIC_API_KEY).	no	false
node-version	Node.js version to use (GateTest requires Node 20+).	no	22
working-directory	Directory to scan. Defaults to the repository root.	no	.
report-format	Report format produced by the gate: console / json / sarif / junit.	no	console
fail-on-warning	When true, warning-severity findings also block the gate (default is errors-only).	no	false
track-non-fixable	Open a GitHub Issue per error-severity finding the auto-fixer could not resolve. Idempotent — re-scans never duplicate. Default off.	no	false
threshold	Minimum acceptable letter grade (A/B/C/D) computed from the scan (same A-F scale as gatetest.ai/playground). When set, the gate ALSO fails if the computed grade is below this — on top of, not instead of, the `block`/`fail-on-warning` error-count gate. Default empty = grade never gates on its own.	no	""
summary-comment	Post (and keep updated across re-runs) one PR comment with the grade, error/warning counts, and top findings. Default true on pull_request events.	no	true
block	When true, the gate FAILS the workflow on any confident error finding. Default FALSE — fresh installs report findings as a PR comment but never block CI, so pre-existing rot in a mature codebase does not stop deploys on day 1. Flip to `true` once the team has triaged the baseline and wants the gate to actually enforce.	no	false
share-learnings	OPT-IN: sync successful fix recipes (anonymised — no paths, no repo names, no source code outside the 2KB before/after snippet) to gatetest.ai so every other customer benefits from this fix. Default off. See https://gatetest.ai/trust for what is and is not shared.	no	false
recipe-store-url	HTTP endpoint to sync fix recipes to. Defaults to https://gatetest.ai/api/recipes when share-learnings is true. Set to your own URL for a private recipe store (the Action will sync there instead).	no	""
mutation	Run mutation testing after the gate (Nuclear tier or opt-in). Requires the customer test suite to run.	no	false
chaos	Run live-browser chaos / runtime checks against a URL (Nuclear tier or opt-in). Requires playwright.	no	false
chaos-url	Live URL to chaos-test (deployed staging or prod, NOT localhost). Required when chaos: true.	no	""
mutation-blocks-merge	When true, a failing mutation score blocks the merge. Default false — mutation is coaching, not a gate.	no	false
chaos-blocks-merge	When true, chaos / runtime failures block the merge. Default false — chaos surfaces resilience gaps as warnings.	no	false

Outputs

name	description
gate-status	The gate verdict: "passed" or "blocked".
error-count	Number of error-severity findings from the scan.
warning-count	Number of warning-severity findings from the scan.
report-path	Filesystem path to the generated report (empty when report-format is "console").
grade	Letter grade (A/B/C/D/F) computed from the scan — same A-F scale as gatetest.ai/playground. Empty when no JSON report was produced.
score	Numeric 0-100 score the grade above is derived from.