researching-anthropic-guidancelisted

Research a topic by fanning out parallel web-search agents restricted to **official Anthropic domains only**, then synthesizing their findings into a single briefing that merges agreement into consensus and calls out contradictions. **No code, config, or test files are written during this skill** unless the user explicitly asks to persist a snapshot (Step 5) — it otherwise ends with an inline briefing. **This skill must only run in the main (hub) agent, never inside a subagent.** It dispatches subagents via the Agent tool; spokes carry `disallowedTools: Agent` and cannot do this themselves. If you find yourself executing this skill as a subagent inside a larger task, stop and surface the research request back to the hub instead. ## Steps ### 1 — Scope the topic Read the research topic from the user's invocation or from the surrounding task context (e.g. a design decision an audit or plan is blocked on). If the topic is ambiguous or spans multiple unrelated subjects, ask **one** focused clarifying question before proceeding — otherwise infer reasonable scope and proceed immediately. Identify 3–5 facets of the topic that a thorough research pass should cover. Each facet becomes one Explore agent brief in the next step. Good facets are orthogonal and independently searchable (e.g. for "subagent design": architecture/composition patterns, tool-grant philosophy, model selection guidance, prompt-writing conventions, context/token management). Derive a short kebab-case topic s