Skip to main content
After a run, momentic ai triage investigates failed tests and repairs them in place. Based on your organization’s Healing settings, it then opens a pull request with the fix.
This is the post-run repair agent. It is separate from in-run auto-healing (locator resolution and smart waiting), which keeps a single run green without changing your test files.

​
Add the workflow

Add a workflow to run your tests on every pull request and triage whatever fails:
.github/workflows/momentic.yml
name: Momentic

on:
  pull_request:

permissions:
  contents: write
  pull-requests: write

env:
  MOMENTIC_API_KEY: ${{ secrets.MOMENTIC_API_KEY }}

jobs:
  test-and-triage:
    runs-on: ubuntu-latest
    timeout-minutes: 20
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - uses: actions/setup-node@v4
        with:
          node-version: 22
          cache: "npm"

      - run: npm ci
      - run: npx momentic install-browsers chromium

      - name: Run tests
        continue-on-error: true
        run: npx momentic run --output-dir test-results

      - name: Triage failing tests
        if: always()
        run: npx momentic ai triage test-results --yes --timeout-minutes 10
MOMENTIC_API_KEY authenticates the run so the agent picks up your organization’s healing behavior. Create one here and add it as a repository secret. continue-on-error on the test step lets the triage step run even when tests fail, and --timeout-minutes caps total triage time. --yes skips confirmation prompts (on by default when CI is set). Quarantined tests are healed by default. Pass --skip-quarantined to exclude them. The triage step automatically uploads the updated local results archive after processing, so the dashboard and connected reporting receive the final results and validation signal. Pass --no-upload to keep the post-triage archive local. The contents: write and pull-requests: write permissions let the Momentic GitHub App open the fix PR, which uses your repository’s pull request template.

​
Choose what a heal produces

Set On successful heal in Settings > Healing to control what a successful repair produces:
  • pull-request: open a PR with the fix (default). Best when you want the fix reviewed and merged like any other change.
  • draft-pull-request: open the PR as a draft so a human verifies it before it is marked ready. Requires momentic@2.123.0+.
  • direct-commit: commit the fix straight onto the branch under test, no PR. Writes through the Momentic GitHub App server-side, so it works under a detached HEAD CI checkout and needs no contents: write. On a protected branch (e.g. main) it opens a draft PR instead.
  • patch: print a git apply-ready diff instead of opening a PR. Needs no GitHub App or pull-requests: write, so it is the option for forked-PR runs or when you would rather apply the fix yourself.
On failed heal decides what happens to tests the agent cannot fix:
  • warn: log the failures and keep the job green. Use this while introducing auto-heal to an existing suite.
  • fail: exit 1 so the check blocks the PR once you trust it.
  • quarantine: exit 1 for this run and quarantine the tests the agent could not fix.

​
Skip triage when run is clean

Triage often needs your app’s dev servers running (and browsers installed) before it can re-run and repair tests — expensive setup to spin up on every PR. When every non-quarantined run already passed cleanly there is nothing to triage, so you can short-circuit all of that setup with momentic results check. results check reads a finished results directory and exits 0 only when every non-quarantined run passed cleanly — no failures, cancellations, in-run failure recovery, or failure classifications. Any unclean run exits 1. Gate the triage setup on that signal:
.github/workflows/momentic.yml
- name: Run tests
  continue-on-error: true
  run: npx momentic run --output-dir test-results

- name: Check whether triage is needed
  id: triage_check
  run: |
    if npx momentic results check test-results; then
      echo "needed=false" >> "$GITHUB_OUTPUT"
    else
      echo "needed=true" >> "$GITHUB_OUTPUT"
    fi

# Expensive setup only runs when something is actually broken.
- name: Start dev servers
  if: steps.triage_check.outputs.needed == 'true'
  run: npm run dev &

- name: Triage failing tests
  if: steps.triage_check.outputs.needed == 'true'
  run: npx momentic ai triage test-results --yes --timeout-minutes 10
On a clean run the results check step exits 0, needed stays false, and the dev-server and triage steps are skipped entirely. This is the same gate Momentic uses in its own CI. See the momentic results check reference for --json output and the exact “clean” definition.

​
Variations

For large suites, run healing on a schedule or on demand instead of on every PR by swapping the trigger. This mirrors the AI heal demo in momentic-ai/examples: 
on:
  workflow_dispatch:
  schedule:
    - cron: "0 9 * * 1" # every Monday at 09:00 UTC
Narrow the triage queue when you only want to touch some tests: 
# Only quarantined tests, two sessions in parallel
npx momentic ai triage test-results --only-quarantined --parallel 2

# Skip known-flaky tests by name
npx momentic ai triage test-results --exclude "checkout flow" "^Legacy "

# Preview the triage plan without healing or uploading the local archive
npx momentic ai triage test-results --dry-run --no-upload
--dry-run still uploads the local results archive by default. Combine it with --no-upload to skip the archive upload. See the momentic ai triage reference for every flag. The same run-then-triage pattern works on GitLab CI, CircleCI, Jenkins, and custom setups.