BETA · GateTest is in active polish ahead of public launch. Some flows are rough. Found a bug? hello@gatetest.ai — we're reading every message.
5 minutes from install to first auto-fix PR

Get your CI self-healing
in four steps.

You install one workflow, add one secret, and the next time your CI fails you get a pull request with the fix already written.

1

Install the workflow

~30 seconds

From the root of any GitHub repo, run the one-liner. It drops three files: the CI workflow, a pre-push hook, and a protection marker. Nothing else changes.

curl -sSL https://raw.githubusercontent.com/crclabs-hq/gatetest/main/integrations/scripts/install.sh | bash

Public or private repos both work. The workflow file only readsyour code — nothing is sent anywhere until CI runs.

2

Add the ANTHROPIC_API_KEY secret

~2 minutes

This is what unlocks auto-fix PRs. Without this secret, CI still runs the gate — but no PR opens when something fails.

⚠ This is the step most people skip.

If you skip it, your CI will still detect bugs but won't open fix PRs. You'll see a yellow “auto-repair not configured” warning on every failing run.

  1. Open the secrets page for your repo or org:
    https://github.com/<your-org>/<your-repo>/settings/secrets/actions

    Or for the whole org at once: github.com/organizations/<your-org>/settings/secrets/actions

  2. Click New repository secret (or organization secret).
  3. Name: ANTHROPIC_API_KEY
  4. Value: your Anthropic API key. Get one at console.anthropic.com — pay-as-you-go, no minimum, a typical fix PR costs ~$0.02 in API spend.
  5. Save.

We never see or store your key — it goes from GitHub Secrets directly to Anthropic at fix-time. You're billed by Anthropic for the API usage, not by us.

3

Push a commit (deliberate or real)

~1 minute

On your next push or pull request, the workflow runs. If everything passes, the gate shows green and you move on. If it finds a fixable bug, Step 4 kicks in.

Want to see it work end-to-end? Add an obvious bug and push:

# Pick any JS/TS file in the repo, add a stray console.log + commit
echo 'console.log("debug");' >> src/some-file.js
git add -A && git commit -m "test: trigger gate" && git push

The gate flags console.log in library code as error-severity. CI goes red. Auto-fix runs.

4

Watch the fix PR open

~1 minute

Within ~60 seconds of CI failing, a new pull request appears in your repo titled AI CI-fixer: repair workflow run #<id>. It contains:

  • The actual code fix — console.log replaced with process.stderr.write, or whatever was appropriate for the specific finding.
  • A before/after scan summary in the PR body.
  • A regression test for the bug (so it can't silently come back).
  • On Scan + Fix and Forensic Scan tiers: a pair-review comment from a second Claude scoring the fix on 4 axes.

Review it like any other PR. Merge it if you're happy. Your gate stays red until either this PR merges or you fix it yourself.

Nothing happened on a failing run?

The workflow ran but no fix PR appeared.

Check the workflow output for a yellow “auto-repair not configured”warning. If you see it, Step 2 didn't land — the ANTHROPIC_API_KEYsecret isn't set on the repo (or org).

The workflow says “auto-repair could not generate any fixes.”

The fix engine ran but couldn't produce a verified patch. Common causes: file too large (> 50KB), config-level finding with no file:line to anchor a fix, or the finding wasn't a straightforward code change (architecture, dependency choice). Check the per-finding [skipped: …] lines in the workflow log for the reason.

CI passes but I want to see a fix PR anyway.

The gate only opens PRs when something fails. Try Step 3's “add a deliberate bug” trick — cheapest way to see the loop work end-to-end on a real repo.

I'm on GitLab/Jenkins/CircleCI, not GitHub.

The CLI works in any CI — npx gatetest --suite full from your pipeline runs the same 110 modules. Auto-fix PRs currently require the GitHub workflow path; CLI is scan-only.

You're live. What's next?

The free path covers most of what you need. Upgrade tiers if you want deeper analysis, pair-review, and cross-finding attack-chain correlation.