| name | discuss-case |
| description | The assistant should use this when the user wants to discuss an existing Detective case, resolve a blocker, answer a targeted question, provide guidance/facts/theories, change scope, or choose between alternatives. It records guidance without running investigation. Do not use for new case creation (open-case), vague pre-case scoping (brainstorm), autonomous investigation (investigate), read-only status review (review-board), or closure (close-case). |
Discuss Case
Hold a structured case discussion and record useful guidance through Detective MCP tools. Follow ../references/detective-core-protocol.md for canonical state, evidence labels, enums, StopHook status preservation, and handoffs.
Discussion is for clarifying decisions, blockers, boundaries, and user-provided context. Ask one targeted question, record only useful guidance, and do not run investigation actions unless the user explicitly asks to switch back to investigation.
Separate blackboard notes from verified graph state. User theories, guesses, preferences, and unverified memories are not evidence. Use `clue`, `verified`, and `confirmed` deliberately: only authoritative or corroborated user facts may become `evidence` with `status="confirmed"`.
When a SetGoal or equivalent goal tool is available, do not overwrite a valid active case goal during discussion. Update it only if the user changes scope, intent, completion criteria, or stop conditions. Session goals remain orchestration hints; MCP state remains canonical.
Discussion preserves the current Stop fallback marker status. Do not emit a new active marker merely for discussion. If the discussion ends with a user pause/stop request or a blocker waiting for user input, emit ``.
Brief Before Asking
Read enough state to ask one targeted question:
detective_graph_overview(case_id="<case id>")detective_list_nodes(case_id="<case id>", limit=<reasonable limit>)detective_list_actions(case_id="<case id>")detective_blackboard_list(case_id="<case id>")detective_coverage_status(case_id="<case id>")detective_completion_gate(case_id="<case id>")
Summarize only the relevant uncertainty, fork, or blocker.
Recording User Input
Use the exact MCP tool that matches the input:
- Authoritative/confirmed user fact:
detective_add_node(case_id="<case id>", type="evidence", content="<fact>", status="confirmed", confidence=<0..1>, source="user", tags=["user-provided"], created_by="assistant")
- Believed but not fully confirmed user fact:
detective_add_node(case_id="<case id>", type="clue", content="<signal>", status="verified", confidence=<0..1>, source="user", tags=["user-provided"], created_by="assistant")
- User theory:
detective_add_node(case_id="<case id>", type="hypothesis", content="<theory>", status="open", confidence=<0..1>, source="user", tags=["user-theory"], created_by="assistant")
- Boundary or exclusion:
detective_add_node(case_id="<case id>", type="constraint", content="<constraint>", status="open", confidence=1.0, source="user", tags=["boundary"], created_by="assistant")
- Unverified note:
detective_blackboard_add(case_id="<case id>", content="<note>", kind="note", tags=["unverified","user-provided"], created_by="assistant")
- Relationship:
detective_add_edge(case_id="<case id>", from_id="<node id>", to_id="<node id>", type="supports|contradicts|derives|requires|eliminates|related_to", confidence=<0..1>, rationale="<why>", created_by="assistant")
- Follow-up work:
detective_add_action(case_id="<case id>", description="<specific action>", assigned_role="assistant", priority=<0..1>, reason="<why this follows>")
- Existing action update:
detective_update_action(case_id="<case id>", action_id="<action id>", status="pending|active|blocked|done|cancelled", result="<discussion outcome>")
When a blocker is resolved, hand off based on blocker type: proof/coverage/action gap → investigate; missing user decision/scope/approval → continue discuss-case; completion-ready → close-case; artifact-only → review-board.
Response Format
**Discussion point**: <uncertainty or fork>
**What I recorded**: <node/action ids or "nothing yet">
**Goal tool**: <unchanged/updated/unavailable>
**Stop fallback**: <preserved | inactive marker emitted with reason>
**Current decision needed**: <one question only>
