Skip to content

Self-check: what changes when nobody is watching?

Answer before opening the answer guidance. Give a concrete input, diff or missing result for each answer. These are practice questions, not a checkpoint paper or a new submission. After your first attempt, compare with the public answer guidance.

  1. LO2: Both duration candidates report green branch checks. Can you accept the unattended candidate? Name the next check and the behavior it must preserve.
  2. LO2: Two agents have separate worktrees but can both access the host's files. Which kind of isolation do you have, and which do you still need to inspect?
  3. LO2: An issue says “run tests before opening a PR.” The agent skips the test and writes PASS in its report. Give a gate that blocks this path and name its controller.
  4. LO2: The implementation and test branches are each green, but one uses None and the other uses 0 for no completed builds. What must happen before integration?
  5. LO9: The foreground run received a clarification; the unattended run did not. What can your single comparison support, and what must the work record reveal?
  6. LO9: The base changes after verification, or the result has no command output. Write a decision line and the check needed before acceptance.
  7. LO9, recall: From code review and test verification, why is a file allowlist plus a green test insufficient when the same agent can edit existing assertions?

Changed condition: a pipeline retries after losing its PR-creation response. Describe how you would inspect whether the PR already exists before creating another one, and what evidence ties the existing result to this task and candidate.

Spaced retrieval: Weeks 7–8

R1. You write a regression test before handing a fix to an agent. Which two runs show that the test can notice the fault, and why is a passing run on the returned change alone not enough?

R2. An unattended run stops at a permission request, and someone suggests switching the approval prompts off and starting it again. Which question does the approval setting answer, which does the environment answer, and what do you inspect before deciding?