| name | ghostcrab-json-answer-builder |
| description | Build stable JSON answers from GhostCrab MCP outputs for Personal workspaces. Use when the user wants JSON, API-ready payloads, or structured answers separating observed data, inference, missing evidence, and MCP tools used. |
GhostCrab JSON Answer Builder (Personal)
Convert MCP tool outputs into honest JSON.
References: ARTIFACT_KINDS.md, GAP_TAXONOMY.md, RUNTIME_QUERY_PIPELINE.md.
Delivery context (optional)
Resolve {starterkit} via STARTERKIT_PATHS.md — runtime Q&A tail of {starterkit}/personal-mcp/SOP_SEQUENCE.md after operator and evidence-discovery.
Template
{
"workspace_id": "",
"question": "",
"answer_status": "complete|partial|unsupported",
"observed_data": {},
"inferred_interpretation": {},
"missing_evidence": [],
"unsupported_claims": [],
"mcp_tools_used": [],
"recommended_next_tools": []
}
Rules
observed_data: only values returned by MCP tools in this session.
inferred_interpretation: label clearly; never merge into observed_data.
mcp_tools_used: tool names and key arguments.
- Do not treat
required_facets from a projection contract as proof rows unless search/pack returned matching facts.
Layer vocabulary (when explaining to the user)
- Facets —
agent_facts rows (ghostcrab_search)
- Graph —
graph_entity / graph_relation (ghostcrab_graph_search, ghostcrab_traverse)
- Analysis plan —
analysis_plan via ghostcrab_pack
- Snapshot —
answer_snapshot via ghostcrab_projection_get
See ARTIFACT_KINDS.md for the full table.
Guardrails
- Use
mcp_tools_used only (legacy Pro CLI JSON field names are deprecated).
- Route
unsupported answers to ghostcrab-gap-auditor for structured gap JSON.