What problem does it solve? When an AI architect agent needs expert guidance on unfamiliar domains, ad-hoc answers risk being unverified or stale. This Skill provides a structured consultation protocol that routes questions to subject-matter-expert agents, enforces read-before-cite discipline, and marks uncertainty explicitly so advisory answers stay grounded in the current codebase. ## Core Features & Use Cases - Bounded SME Consultation: Identifies 1-3 relevant domains per task and calls the swarm's SME agent once per domain, serially, with a cap of 3 SME calls per project phase. - Guidance Caching: Checks .swarm/context.md for cached guidance first and stores new guidance there to avoid redundant consultations. - Read-Before-Cite Discipline: Requires agents to read actual files or tool output before citing source code, line ranges, API behavior, or CLI flags, and mandates confidence: LOW marking for unverifiable claims. - Use Case: During a multi-phase software project, the architect encounters a database schema question outside its expertise; it invokes this protocol to consult the SME agent, caches the guidance, and cites only verified file contents in its final answer. ## Quick Start Activate MODE: CONSULT to answer an advisory question by consulting the swarm's SME agents and caching the verified guidance in context.md.
