Research Scoping

A single-context triage pass that maps a question before you decide to spend real tokens on it. It runs the multi-perspective move from Stanford's STORM method compressed into one prompt: discover the perspectives the question demands, surface where they fight, sketch how a full answer would be structured, then say whether to handle it inline or escalate. It retrieves nothing and cites nothing. The output is a map and a plan, framed as questions to pursue, not answers.

The value is economic and analytical at once. A full COMPLEX /ann run costs hundreds of thousands of tokens; this pass costs a fraction of that in one context with no subagent fan-out. So it either saves you the big run, or sharpens the roster so the big run is better aimed.

Lane — when this skill, not another

• research-scoping (this): map a question's perspectives and tensions in one cheap pass, then triage. No retrieval, no citations, no subagents, no deliverable.
• /grill-mel: interview Ane one question at a time to sharpen a design she already holds. research-scoping runs the opposite direction, handing Ane a map without interviewing her.
• /ann: full orchestration that produces the deliverable. research-scoping is the front door you walk through before deciding to open it.
• /researcher, /evidence-synthesis: retrieve and cite real sources. research-scoping never does; it hands off to these when the question needs evidence.
• /decision-debate: token-heavy multi-agent debate on one contested decision. research-scoping is single-context and spawns nothing.

If the question clearly warrants the deliverable, skip scoping and go straight to /ann or the right builder.

The honest-by-construction rule

This skill must never print a citation or assert a fact about IPPF, an MA, a partner, a contact, a figure, or a date. Everything it surfaces is a hypothesis to verify, by design. A fast, un-sourced map that looks authoritative is the failure mode the whole MEL system exists to prevent, so the guardrail is structural, not stylistic: