Week 3 self-check answer guidance

Write your own record first. These are bounded checks, not a report of a real agent run.

Retrieval 1. should_trigger is the intended boundary, predicted is the student's expectation before the run, and boolean observed records the selection evidence the CLI showed. If the CLI omits selection, the transcript says selection: not shown; the YAML field remains boolean and observer_detail explains the basis.

Retrieval 2. Run uv run --locked python lectures/week-03/fixture/skill_check.py with the two student paths. It checks file structure and consistency; it cannot observe hidden skill selection or prove that a model ran.

Q1

The answer names the hand-run commands, their inputs and decisions, and the exact focused check command. The saved output and revision make the baseline reproducible.

Q2

The three positive prompts describe the same workflow in different words. Each has a prediction made before the run; the student cites the description words that set the boundary rather than copying the description into every prompt.

Q3

The near miss asks for implementation or a new feature rather than the selected workflow. “Stays quiet” means the CLI reported no selection; when selection is not shown, the transcript records selection: not shown and the YAML keeps a boolean supported by the record. It never means inventing a positive observation.

Q4

The five fields are the exact request, repository revision, selection line (or a selection: not shown entry in the transcript), command with exit status and output, and verification command with its result.

Q5

skill_check.py proves front matter, sections, links, prompt counts, and any required mismatch revision. It reads data only; it does not run a model, execute a skill, or certify that a host selected one.

Q6

The revision mapping states what description changed and lists exactly the mismatched prompt IDs with their rechecked observations. The original prediction remains in the file.