What problem does it solve? Multi-step work breaks down when it must span multiple chat sessions: context is lost, decisions get re-litigated, and authority boundaries blur. This Skill turns a complex task into self-contained, cold-start prompts that carry the full work frame, decisions, and permissions into every fresh chat. ## Core Features & Use Cases - CHAIN mode: Decomposes sequential work into 2-6 stages where each stage executes its slice and emits the next stage's prompt with updated context, ending in a CHAIN COMPLETE report. - FAN-OUT mode: Generates one independent prompt per piece of work sharing identical stable context, with optional dispatcher and collector prompts for combined results. - Authority-safe propagation: Keeps operator decisions, observed facts, and stage proposals labeled separately so no stage silently widens permissions or promotes a proposal into a decision. - Use Case: You need to refactor a stylesheet and then validate responsive behavior in separate clean chats. The Skill produces a Stage 1 prompt you paste into a new chat; that chat does the refactor and emits the Stage 2 validation prompt automatically. ## Quick Start Ask the agent to break your multi-step task into a prompt chain where each stage runs in a fresh chat and generates the next stage's prompt.
