Scheduled detection

Instead of checking for patterns manually, you can schedule a GitHub Action to run detection automatically and open issues when new patterns are found.

How scheduled detection works

1. A cron job runs weekly (or on your schedule)
2. The action calls flaky-tests check against your shared store
3. If new patterns are found, a GitHub issue is opened automatically with an AI investigation prompt embedded in the body
4. If the same pattern is already open as an issue, a new one is not created (deduplication by test name)

Setting up the workflow

Add this workflow to your repository:

.github/workflows/flaky-check.yml

name: Flaky test detection
on:
  schedule:
    - cron: '0 9 * * 1'   # Monday at 9am UTC
  workflow_dispatch:        # Allow manual runs

jobs:
  detect:
    runs-on: ubuntu-latest
    permissions:
      issues: write
    steps:
      - uses: brewpirate/flaky-tests@v1
        with:
          store: turso
          connection-string: ${{ secrets.TURSO_URL }}
          auth-token: ${{ secrets.TURSO_AUTH_TOKEN }}
          github-token: ${{ secrets.GITHUB_TOKEN }}
          window-days: '7'
          threshold: '2'
          create-issues: 'true'

Customizing the schedule

The cron field uses standard cron syntax. Some examples:

Schedule	Cron expression
Every Monday at 9am UTC	0 9 * * 1
Every day at midnight	0 0 * * *
Every 6 hours	0 */6 * * *
First of the month	0 9 1 * *

Tip

Weekly detection with a 7-day window is a good default — it gives enough time to accumulate meaningful data while still catching problems quickly.

What the issue looks like

When a new pattern is detected, the action opens an issue like this:

Title: Flaky test: auth > login > should redirect on success

Body:
**File:** tests/auth.test.ts
**Failures:** 8 in the last 7 days (0 in the prior 7 days)
**Error type:** timeout

<details>
<summary>Investigation prompt</summary>

The following test has started failing intermittently...
[full structured prompt for Claude/Cursor/Copilot]
</details>

Deduplication

Before opening a new issue, the action searches for existing open issues with the same title. If one is found, no duplicate is created. Close the issue when the problem is resolved — the next detection run will open a fresh one if it reappears.

Running detection manually

You can trigger the workflow manually from GitHub’s Actions tab without waiting for the schedule. This is useful when you want to check for patterns immediately after deploying a change.

Or run locally:

FLAKY_TESTS_STORE=turso \
  FLAKY_TESTS_CONNECTION_STRING=libsql://... \
  FLAKY_TESTS_AUTH_TOKEN=... \
  bunx @flaky-tests/core --create-issue \
  --repo your-org/your-repo \
  --window 7 \
  --threshold 2