7. Exercises
In short
- Six exercises, from writing a spec to running your own small team of agents.
- Work on a fork of OpenLucid; every exercise has a clear ‘done when’.
On this page
Fork github.com/emolionl/openlucid, check out checkpoint-02, and get the backend tests running locally first (see backend/README.md).
1. Read before you build
Read backend/SPEC.md and answer: which principle decides what happens when user B asks for user A’s journal entry? Where in the code is it enforced? Which test proves it?
Done when: you can point to the principle, the line of code and the test.
2. Write a spec section
Feature: users can add tags to a night (for example stressed, new-moon, cue-night). Write the spec section: principles (if any), data model changes, endpoints, status codes, isolation and test rules. Don’t write code yet.
Done when: a classmate can implement it without asking you a question.
3. One agent, one feature
Give an agent your spec from exercise 2 and let it implement the tags feature, with tests. Then check its work as in lesson 5: rerun the tests yourself, read the queries that filter by user, check git status.
Done when: the full suite passes on SQLite and on PostgreSQL (scripts/pg_test.py), and you found at least one thing to improve in the agent’s work, or can explain why it’s correct.
4. Be the reviewer
Pick one lens from lesson 4 (security, isolation, contract, operations) and review the tags feature through that lens only. Write each finding with a concrete scenario as evidence.
Done when: you have at least one finding, or a written argument why the lens finds nothing.
5. Prove it with a failing test
For each finding from exercise 4: write a test that fails because of it. Then fix the code until the test passes. If you can’t make a test fail, the finding isn’t real; write down why.
Done when: every finding is either fixed with a test or rejected with a reason.
6. A small team
Write a workflow with three agents, modelled on examples/backend-workflow.js: one builds a feature, one reviews it through two lenses, one fixes the findings test-first. Use file ownership and a structured report for each agent.
Done when: the workflow runs end to end, and you can explain in two sentences why each agent needed the information it was given, and nothing more.
Reflection
Write half a page: where did the agent save you the most time, where did it mislead you, and which check caught it? That half page is worth more than the code.