| name | brainstorm |
| description | The assistant should use this when the user has a vague concern, hunch, symptom, unknown failure, or asks what to investigate before any Detective case exists. It produces a read-only, approval-ready brief. Do not use when the user explicitly asks to create/open a case (use open-case), continue an existing case (use investigate), review status (use review-board), resolve a known blocker with user guidance (use discuss-case), or close/finalize findings (use close-case). |
Brainstorm — Pre-Case Exploration
Turn an unclear concern into a concise case brief. Follow the shared protocol in ../references/detective-core-protocol.md for enums, evidence labels, handoffs, and orchestration boundaries.
<HARD_GATE name="pre-approval-no-mutation">
Brainstorm is pre-approval only. Do not call detective_open_case, detective_add_node, detective_add_edge, detective_add_intent, detective_add_action, or any other case-mutating MCP tool until the user explicitly approves the brief and asks to proceed. Read-only code/file/log inspection is allowed when it helps frame the problem.
</HARD_GATE>
<HARD_GATE name="explicit-open-approval">
User approval must be explicit and current-turn clear, for example "open it", "proceed", or "create the case". Ambiguous interest, agreement with the brief, or continued brainstorming is not permission to mutate Detective MCP state.
</HARD_GATE>
Do not call SetGoal or emit `` markers during unapproved brainstorming. After approval, hand off to `/detective:open-case` when available; open-case handles durable MCP creation, SetGoal, and fallback markers.
<HARD_GATE name="no-stop-fallback-before-approval">
Do not emit <DETECTIVE-STOP-FALLBACK ...> markers during brainstorm pre-approval. Stop fallback activation is allowed only after a durable MCP case exists and only as a fallback when no SetGoal or equivalent goal tool is available.
</HARD_GATE>
Workflow
- Restate the symptom, suspected area, and why it matters.
- Do read-only pre-work when it can answer obvious questions without mutating case state.
- Ask exactly one clarifying question at a time; prefer concrete choices.
- Classify inputs using current protocol terms: observation, clue, hypothesis, constraint, success criterion, open question.
- Agree on scope, out-of-scope boundaries, budget, and checkpoint cadence.
- Produce an approval-ready brief.
- Ask whether to open it; on explicit approval, invoke
detective:open-case or directly follow that skill's MCP procedure if skills are unavailable.
Case Brief Template
## Case Brief: <short title>
**Crime Scene**: <observable problem>
**Investigation Intent**: <what the user wants to learn or decide>
**Success Criteria**: <what counts as solved>
**Scope**: <in scope / out of scope>
**Budget**: <max actions or time box; checkpoint cadence>
**Initial Observations**:
- <known fact or symptom>
**Initial Hypotheses**:
- <hypothesis> — confidence: <low|medium|high>; basis: <why>
**Initial Clues**:
-
:
:
:
End by asking one question: "Does this brief capture the case? If yes, should I open it with /detective:open-case now?"
