| name | question |
| description | Grills product or engineering intent against existing docs and code, then records resolved choices for later QRDSPI work. Use when a request needs vocabulary, scope, edge cases, defaults, or user preferences clarified before research, design, structure, plan, or implementation. |
| disable-model-invocation | true |
| user-invocable | true |
Question
Quick start
Read enough vault context and code to avoid lazy questions, interview the user in small batches, then stop with a Question Summary that later work can carry forward.
No-edit, no-implementation phase. Do not call apply_patch, create tasks, write vault artifacts, or continue into another QRDSPI phase in the same turn. Stop after the Question Summary.
Workflow
-
Ground the interview
- Search/read relevant vault artifacts: domain docs, decisions, research, design, structure, plans, and related docs.
- Inspect the code for existing behavior, vocabulary, UI/API surfaces, state/persistence seams, and tests; inspect the code rather than asking when evidence is available.
- Answer anything the docs/code can answer. Ask humans only for intent, tradeoffs, priorities, or missing domain facts.
- Build a working evidence map with file/module/symbol references and label uncertainty explicitly.
-
Grill the design tree
- Walk the design tree branch-by-branch through vocabulary, user-visible behavior, scope, ownership/persistence, compatibility/migration, failure cases, non-goals, and acceptance boundaries.
- Prefer
ask_user for 1-3 focused questions at a time. Give a recommended answer for every question.
- Use concrete scenarios, not abstract prompts.
- After each answer, update the working decision tree and ask another batch if an unresolved branch could materially change evidence gathering or solution direction.
- Do not stop after one batch merely because you have plausible defaults.
-
Emit the handoff
- Produce a Question Summary as a context handoff packet, not a polished artifact.
- Include:
- intent in project vocabulary
- evidence map: docs/code read, relevant symbols, current behavior, uncertainty
- resolved answers with rationale
- decision tree covered
- blocking open questions only
- recommended defaults and explicit non-goals
- established answers plus factual claims later work should verify
Quality bar
- The output should preserve enough context that later work can continue from the conversation alone.
- Do not make the research handoff merely a list of generic questions.
- Do not create or update vault artifacts by default.