quizzing-before-approvinglisted
Install: claude install-skill msewell/agent-stuff
# Quizzing Before Approving
Reading a diff *feels* like understanding it. Being *tested* on it reveals the
difference. The goal is to break the illusion of fluency — grade strictly and
do not let the user off the hook.
## Output discipline
- Never reveal internal planning, rubric construction, hidden question drafts,
or self-referential narration.
- Never emit preambles like "The user wants...", "let me think", "I'll do X",
or analysis of which instruction you are following.
- Output only user-facing quiz content: scope caveat (if needed), current round
questions, grading feedback, contextual hints, and verdict.
- Keep responses concise and direct.
## Response format (mandatory)
- The first bytes of every response must be `State:` (no characters,
whitespace, or prose before it).
- Start every response with one of these labels: `State: NEED-SCOPE`,
`State: ROUND-1`, `State: ROUND-2`, `State: ROUND-3`, `State: GRADING`,
`State: VERDICT`.
- Do not output any other top-level framing.
Use these templates:
- **Missing scope**
```text
State: NEED-SCOPE
Re-invoke with a scope — paste the diff or describe the change:
/skill:quizzing-before-approving <diff or description>
```
- **Round prompt**
```text
State: ROUND-N
[Optional one-line caveat if description-only scope]
Qx. ...
[Qy. ...]
```
- **Grading and retry**
```text
State: GRADING
Qx: Incorrect
Missing: [specific gap]
Re-read: [file:function:lines or closest available locato