CI failure fixes

When a CI run fails on your repo, the failure itself becomes the task: your agent reads the failing run, diagnoses it against the actual code, and posts what broke and how to fix it to your chat channel — often before you've opened the logs.

What you need

• The GitHub channel set up (a webhook with a secret)
• A chat channel for the report to land in
• The repo registered as a project

How it behaves

Deliveries are verified with your webhook secret, so only real GitHub events trigger work. A failed run turns into a diagnosis in your chat; green runs stay silent. The agent has the repo checked out, so "the test broke because commit abc123 changed the fixture shape" is the kind of answer you get — not a restated log line.

Make it yours

Route the reports to a stronger model if your failures tend to be subtle — either nudge the channel up a tier, or bind it to an agent with the exact model (and thinking depth) you want:

channels:
  github:
    tier: strong          # simplest: escalate this channel

agents:
  debugger:
    model: claude-opus-5
    reasoning: high
channels:
  github:
    agent: debugger       # precise: this agent, this model, everywhere it answers