Engineering

What changed upstream, before stand-up

Every weekday at 8:30 it checks a public code project your team builds on, for new releases and for open issues that look like bugs or security problems.

The 8:30 run, marked "Errors found" with the check's three notes, then its four calls to GitHub and the cost of the writer and the checker.
Recorded in the staging demo workspace, using public APIs.

One recorded run

Ten issues worth a look, three lines questioned

Who it is for

Engineering teams who build on open source projects and want to hear about new releases and serious bugs before they run into them.

On 16 November at 8:30 the agent looked at the Node.js project on GitHub for the past day. There was no new release (the latest was from 12 November), and it listed ten open issues worth a look, among them a crash when a worker thread stops while compression is starting up, and a crash when copying files to the top of a Windows drive.

Its answer check, a model from a different provider, questioned three lines: two issues listed as bugs were feature requests, and one label was not on the issue. The report still arrived on time, with those notes at the top.

Recorded once in the staging demo workspace against public APIs. The numbers are copied from that run; another run, other data or another model can give a different result.

What the agent does

  • Runs on its own every weekday at 8:30
  • Lists new releases in the time window, and the latest release
  • Picks the open issues whose labels or titles point to bugs, regressions or security problems
  • Has a model from a different provider check the report, and notes what it questions

What you need

  • The project's name on GitHub, for example nodejs/node
  • Two AI models from different providers: one writes, the other checks
  • Nothing else for a public project: GitHub answers without a key, with lower limits

Set up your own

Three steps to a version of this.

The docs walk through an agent of this kind step by step. It is a guide to the pattern, not a copy of this exact agent.

Read the guide
  1. Connect GitHub

    Paste the link to GitHub's API description and keep the actions for repositories, releases and issue search.

  2. Build the agent

    Describe the report, and add an answer check with a model from another provider.

  3. Set the schedule

    Every weekday at 8:30, with the project and the time window in the message.

Keep exploring

Other recorded examples.

Start with a job like this one.

Connect the service, describe the result you want and check the first run yourself.