Exercise prompts

Prepare a fresh workspace before sending a task. Run commands from the course checkout after setup.

Exercise	Complete task and run instructions
1. Harness choice	Pagination repair
2. Tool design	PR explanation and MCP modes
3. Memory	Prior customer agreement
4. Routing	Counter and controlled security repair
5. Goals	Staged checkpoint and native continuation
6. Delegation	Reservation and optional ledger prompts

Open your exercise page above, follow its configuration steps, and copy the task prompt into a new Canvas conversation. Use a separate fresh workspace for each condition; the updated pages provide preparation commands.

Where a prompt contains /absolute/path/to/.workshop-venv/bin/python, replace it with the Python interpreter path printed by prepare-all. Keep the rest of the task unchanged for a harness comparison.

Optional: get a prompt from the terminal

Original W1/W2 prompt copies are saved in your session's prompts/ folder with the interpreter path filled in. For updated W3-W6 tasks, use classroom.py; original session prompts for those workshops are reserves. You can also print one without creating a workspace:

.workshop-venv/bin/python workshops/odsc-2026/labs/classroom.py prompt w3-agreement-accepted

Keep the course checkout outside the agent workspace. Run its external checker from your own terminal after the agent stops. The worksheet records predictions, evidence, and decisions. Saved results provide a fallback when live access is unavailable.