Lab 9 - CI

Put an agent where it helps most and risks least: in CI, explaining failures. Push a change that breaks a test, open a pull request on your fork, and watch an agent explain your mistake in public. You keep the merge button.

Module	9 - CI & Toolchain Integration
Time	15 minutes
Shop preset	clean: CI starts its own shop from the pinned image, whatever your local shop does
You need	Lab 0 done: your fork, and gh auth status signed in
You start from	Your fork, with your work of the day committed

Steps

1. Enable Actions on your fork: open its Actions tab on GitHub and confirm that you want to run its workflows. Three are provided in .github/workflows/:

   Workflow	Runs	Does
   run-tests.yml	on every push and pull request	runs the suite against the pinned shop and uploads the results
   agent-triage.yml	when a pull request's tests fail	posts one comment with the failed tests and, with a credential, an agent's root cause
   heal-suggestions.yml	when you start it	turns heals into a suggestion pull request (the stretch goal)

   A fourth workflow builds this workshop's website. It runs only in the workshop's own repository, never on your fork.

2. Optionally give the triage agent a credential. Without one, the lab still works: the comment then lists the failed tests, without an analysis. With one, an agent adds the root cause. Set a spending cap on any key first.

   You have	Set these secrets on your fork	The analysis comes from
   a Claude subscription	CLAUDE_CODE_OAUTH_TOKEN, from claude setup-token	the Claude Code Action
   an Anthropic API key	ANTHROPIC_API_KEY	the Claude Code Action
   a key for any OpenAI-compatible endpoint, for example your healing key	TRIAGE_MODEL, TRIAGE_BASE_URL, TRIAGE_API_KEY	one call to that model

   gh secret set CLAUDE_CODE_OAUTH_TOKEN --repo <your-handle>/ai-engineering-robotframework

   gh secret set asks for the value, so it never lands in your shell history.

3. Make a branch and break something. For example, change an expected value in a test you wrote today, or in tests/api/smoke.robot. Commit it yourself, in the terminal:

   git switch -c lab-09-break
   git commit -am "Break a test on purpose"
   git push -u origin lab-09-break

4. Open a pull request on your fork, not on the workshop's repository:

   gh pr create --repo <your-handle>/ai-engineering-robotframework --base main --head lab-09-break --fill

5. Watch the checks until they finish:

   gh pr checks --repo <your-handle>/ai-engineering-robotframework --watch

6. Read the comment:

   gh pr view --repo <your-handle>/ai-engineering-robotframework --comments

   Does the root cause it names match what you broke? What evidence does it give? Would you trust it on a change you had not made yourself?

7. Close the pull request without merging:

   gh pr close --repo <your-handle>/ai-engineering-robotframework lab-09-break

Stretch

Start the heal-suggestion workflow from the Actions tab (Run workflow). It runs the suite against a drifted shop with healing on, and opens a pull request that proposes the heals as changes. Review it the way you triaged heals in Lab 8: merge nothing you would reject. It needs your healing model as the secrets HEAL_MODEL, HEAL_BASE_URL and HEAL_API_KEY; without them, it ends with a notice and changes nothing. Forks do not let GitHub Actions open pull requests by default: then the run's summary links the branch with the heals, and you open the pull request yourself.

Compare with the reference

When you are done, compare your result with the reference: what the rehearsal produced, why it is a good result, and the answers the debrief covers. Open it after the lab: it gives the answers away.

If your agent fails

This lab needs no agent of your own: the agent runs in CI. If Actions will not run on your fork, follow the recorded walkthrough of this lab.