Week 4 self-check

Answer without an agent, then compare with the public answer guidance. These questions are formative. Use the studio records as evidence rather than guessing what an agent “must have” done.

1. Write one instruction-file rule for a parser change. Name the failure it prevents and the evidence a reviewer can inspect.
2. What is the difference between the request “run the focused test” and a pre-commit hook that exits 2 for an unused import? Give one way the request can be skipped.
3. A hook prints a finding but exits 0. Did it gate the commit? What observation settles it?
4. Your staged paths are src/report.py and pyproject.toml; the rule forbids dependency changes in a parser task. Which path should the hook report, and why?
5. A passing hook run is recorded. What application behavior does that record leave unverified?
6. Give a researcher subagent two allowed tools, one forbidden action and a hand-back field. Why does the parent still inspect the cited file?
7. The implementer returns a diff but no test output. Which contract item is missing, and should the parent continue?
8. A reviewer says “looks safe” without a file, input or command. Mark the hand-back confirmed, refuted or cannot tell, and state the missing evidence.
9. When should a role stop and ask for a narrower task? Give one concrete boundary.
10. End with a decision line for your studio: accept, narrow, recheck or stop. Name the record that supports it.

Transfer

A project specification says “the agent should follow repository rules.” Rewrite it as one observable instruction and one event-time gate. State what each check proves and what it cannot prove.

R1. A request says “run the focused test,” but the agent reports only a green hook. What observation is still missing?

R2. A subagent returns a file name and a confident summary. What evidence lets you decide whether its hand-back is confirmed?