AI coding agents CI on GitHub Actions

Agent-driven pull requests multiply CI volume and cost overnight.

Why it happens

Agents push far more frequently than humans, and every push runs the full pipeline.

What to change

1. Use concurrency groups so superseded agent pushes cancel their own in-progress runs
2. Gate expensive jobs behind path filters and a ready-for-review condition
3. Set per-repository budget caps so an agent loop cannot run up an unbounded bill
4. Give agent branches a smaller runner tier than release branches

AI coding agents on runnerhut

concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: true
 
jobs:
  test:
    if: github.event.pull_request.draft == false
    runs-on: runnerhut-4vcpu-ubuntu-2404

The only runnerhut-specific line is `runs-on: runnerhut-4vcpu-ubuntu-2404`. Everything else is standard GitHub Actions — the same actions, the same secrets and the same permissions model you use today.