Make the chapter 06 Feature Workbench exercise reliable

[DanWahlin]

Why

I ran the chapter 06 hands-on exercises step by step in an isolated GitHub Copilot app persona. Exercises 1 and 2 worked. Exercise 3 (Feature Workbench) stopped twice and needed repair prompts that were not in the README:

1. Run baseline: Tests and the build passed, but Implement stayed disabled. The generated canvas required "before" screenshots saved as PNG files. The app's Browser tab returns screenshots to the model only, not as files.
2. Browser validation: The action timed out at 15 minutes because the plan required real Tab/Enter keyboard checks that the Browser tab cannot run. Run final checks stayed disabled.

The canvas also showed a false "No correlated agent response was saved" error, and the session stayed in Autopilot mode after each action.

What changed

06-canvases/README.md

• Exercise 3 prompt:
  • Screenshots are optional and never block. Only saved PNG files count.
  • Browser validation checks only the proposed behavior in the app's Browser tab. Checks that the tab cannot run are notes, not failures.
  • Run final checks stays available after Implement, even if Browser validation fails.
  • An action is done when its evidence is recorded. The canvas does not wait for a matching chat reply.
  • The session goes back to Interactive mode after each action.
• Added a row to the plain-language table, and two repair prompts.
• Exercise 1: explains that moving a card to Plan starts a read-only planning agent.
• Exercise 2: tells learners how to add the Changes tab.
• Exercise 3: expected run times, the correct Working status, and an Autopilot note for Run baseline.
• Troubleshooting: an action stays disabled, a false error, the session stays in Autopilot mode, and extra canvas tabs open.

appendices/troubleshooting-reference.md

• Two new chapter 06 rows.

Validation

With the updated prompt, I ran Exercise 3 twice in fresh worktree sessions. I removed the generated canvas before each run.

Step	Run 1	Run 2
Create canvas	about 10 min	about 10.5 min
Generate plan, then Approved	Pass	Pass
Run baseline	4 of 4 tests, build passes	4 of 4 tests, build passes
Implement	Pass, 4 files	Pass, 5 files
Browser validation	Pass, about 3.5 min	Pass, about 4 min
Run final checks	12 of 12 tests	14 of 14 tests

Neither run needed a repair prompt. I also ran the tests and build in each worktree, and the results matched the canvas.

Screenshots

• I retook app-create-canvas-screenshot and the six app-feature-workbench-* images from a third clean run with the updated prompt. That run also passed every step with no repair prompt (14 of 14 tests). File names are unchanged, and the boxes and arrows are at the new positions of the same controls.
• I updated the alt text and step text to match. Run baseline results can come from npm output or from a check tool. The dev server can run under Background or in a Terminal tab.
• app-create-canvas-command does not change with the canvas, so I kept it.

Learner review fixes

After a full read-through from a beginner's point of view:

• Each repair prompt has a label, and the troubleshooting entries refer to the prompts by label.
• Step 4 explains why to select Exit plan mode and I will prompt myself instead of Approve and implement this plan.
• Browser validation says where to check the app (http://127.0.0.1:5173, in the Book App Web Browser tab).
• Implement lists the files that learners can expect to change.
• Exercise 3 has an Expected Output line, as Exercise 2 does.
• Exercise 1 says to move only one card to Plan, because each card starts an agent run.
• Exercise 2 gives the exact summary prompt.

Follow-ups (not in this PR)

• The training fork includes course-author skills in .github/skills. Learner sessions loaded github-copilot-app-automation.

[Copilot]

Copilot review overview

🟡 Changes recommended

The expected Working status conflicts with the generation prompt and screenshot, which still specify Working....

Review effort: Balanced
Findings: 1

Open (1)

• Align generation prompt and screenshot labels with expected status · New

What changed in this PR

Improves Chapter 06 canvas exercises, particularly Feature Workbench reliability.

Changes:

• Makes screenshots and unsupported browser checks non-blocking.
• Adds repair prompts, timing guidance, and mode explanations.
• Expands canvas troubleshooting guidance.

File	Description
06-canvases/​README.md	Revises exercises and troubleshooting.
appendices/​troubleshooting-reference.md	Adds Workbench failure remedies.

---

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.