Scheduling

#Rules

• Keep --limit 1 unless you intentionally want multiple issues per invocation.
• Serialize runs so two Roark processes do not race.
• Run as a user with working gh auth status.
• Use a dedicated checkout or runner not shared with humans.
• Prefer explicit --repo and configured verify in non-interactive environments.

#cron

0 * * * * /usr/bin/flock -n /tmp/roark-auto.lock /usr/local/bin/roark auto --cwd /srv/roark/repos/app --repo owner/repo --limit 1 >> /var/log/roark.log 2>&1

#launchd

Run the launchd job in the user's login session so gh can access keychain credentials. Set the control checkout as the working directory and pass --cwd explicitly.

#GitHub Actions

name: roark-auto
on:
  schedule:
    - cron: "0 * * * *"
  workflow_dispatch:

concurrency:
  group: roark-auto
  cancel-in-progress: false

jobs:
  auto:
    runs-on: ubuntu-latest
    permissions:
      contents: write
      issues: write
      pull-requests: write
    steps:
      - uses: actions/checkout@v4
      - uses: oven-sh/setup-bun@v2
      - run: bun install
      - env:
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        run: bun run roark.ts auto --repo ${{ github.repository }} --limit 1

#Failure handling

If a scheduled run fails a gate, Roark posts recovery instructions on the issue. Inspect and resume it with roark continue from the same control checkout.