Skip to Content
Use casesNightly CI triage

Wake up to last night’s CI failures, already sorted

Stop opening red builds one by one. Every day a deputy sweeps the failed workflow runs across your repositories and investigates each failure. It reads the logs, groups them by likely cause, and tells you which look like flaky tests and which look real. Your team starts the day knowing where to look.

How it works

  • Trigger. A daily schedule, for example 02:00 in your time zone.
  • What the run does. It lists the failed workflow runs from the last 24 hours, reads their failed logs, and groups them by cause.
  • Who is contacted. The deputy’s owner gets a heads-up with the summary and links to the failing runs, on Home and in Slack when your admin has connected Slack.
  • What you see. One message each morning. It is a short list with links.

Set it up

  1. Connect GitHub with the Personal access token method. See Connect GitHub.
  2. Create a manifest with the packages claude-code and github.
  3. Open Deputies, select New deputy, and fill in Name, Description, Manifest and Owner.
  4. In the deputy, select New mission. Paste the prompt below into Prompt, select Dry run, then Activate.
  5. Expand the mission. In the Triggers list, select Schedule.
  6. Set Repeats to daily, At to 02:00, and pick your Timezone. Or select Advanced and enter 0 2 * * * in Cron expression.
  7. Select Create.

What you should see: the schedule in the trigger list with its next run time. After the first night, a heads-up on Home.

Example mission prompt

Review CI health for the repository acme/web.

1. List the workflow runs that failed in the last 24 hours.
2. For each failed run, read the failed log.
3. Group the failures by likely cause. Mark each group as "looks flaky"
   or "looks real", and give one line of evidence.
4. Send me a heads-up with the groups, the counts, and links to the
   failing runs. Keep it short.
If there were no failures, send a heads-up that says so.

Make it yours

  • Keep a list of known flaky tests in the deputy’s Context so they are sorted first.
  • Connect your tracker, for example linear-mcp, and ask the deputy to file one ticket per real failure. See Integrations.
  • Ask for approval first: “Ask me with the choices File tickets and Skip before creating anything.”
Delegate from your AI clientMonday engineering digest