Knuckles-Team Multi-Perspective Inquiry Workflow (STORM, native)
Implements the native perspectival-inquiry engine (agent-utilities concepts KG-2.127 / KG-2.128 / KG-2.129).
Run Stanford's STORM method as a single delegated call to the native perspectival-inquiry
engine in agent-utilities — distinct expert lenses ask different questions, the KG answers
each, and the result is a contradiction/agreement/blind-spot map plus a peer-review
self-critique whose frontier question is submitted as the next research loop.
This workflow is the on-demand twin of the loop's default behaviour: it does not
re-implement the STORM prompts, it calls research_artifact action=inquire (which the loop
also uses), so the two surfaces never drift.
Steps
Step 1: Inquire
Agent: inquiry-agent
Tools: research_artifact
Call research_artifact with action=inquire and topic=<the topic>. The engine derives
the expert lenses, fans acquire_for_topic across each lens's questions, builds the
contradiction/agreement/blind-spot map, runs the peer-review, and (by default) materializes
the inquiry as typed KG nodes. Expected: inquiry_briefing — perspectives + agreements +
divergences + blind_spot + peer_review (dominant/missing lens, confidence, frontier question).
Step 2: Read Back [depends_on: inquire]
Agent: inquiry-agent
Tools: graph_query
Query the materialized inquiry subgraph (research_inquiry / perspective / agreement /
contradiction / blind_spot / peer_review nodes) to present the briefing and surface
the frontier question that was queued as the next research loop. Expected:
perspectival_briefing.
Execution
This is a thin delegating workflow. Execute it via graph-os rather than re-prompting:
- MCP:
research_artifact(action="inquire", topic="<topic>") - REST:
POST /api/research/inquire {"topic": "<topic>"} - Or run the whole workflow through
graph_orchestrate(action="execute_workflow", agent_name="multi-perspective-inquiry", task="<topic>").
The native engine is the single source of truth (knowledge_graph/research/perspective.py);
this skill only orchestrates the call and reads back the result.