research-sweeplisted

# Skill: /research-sweep [topic list] Produce a brief whose every number you would defend in public — by researching each topic independently, then paying a second agent to destroy the first one's claims. **Don't:** dispatch a researcher without the preamble from [research-preamble.md](../_workflow-lib/research-preamble.md) — an agent researching from memory looks identical to one researching from sources until you check a citation. **Don't** let the same agent research a topic and verify it; self-verification confirms, it does not test. **Don't** merge unrefuted and refuted claims into one brief without marking which is which. ## Steps | Step name | Stage | Runs | |-----------|-------|------| | `scope` | Topic list + shared context | orchestrator, with Shane | | `research` | One agent per topic, parallel | researcher subagents | | `verify` | One agent per load-bearing claim | fact-checker subagents | | `merge` | Single decision-grade brief | orchestrator | ### Stage 1: Scope — step `scope` Settle three things before dispatching anything: 1. **The topic list.** Each topic gets its own agent. A topic is one question with a stated angle, not a subject area — "which trigger signals empirically correlate with closed-won, and on what evidence" beats "buying signals". 2. **The shared context block.** Product, positioning, and audience, pasted identically into every researcher prompt so findings come back usable rather than generic. 3. **The counter-position requirement.** Ev