What problem does it solve? Large research-writing tasks often collapse into one-pass generation that produces placeholder chapters, weak argumentation, and unauditable output. This Skill enforces a controlled workflow with stage detection, persistent task packets, review gates, and multi-agent chapter ownership so full-paper drafting stays structured and verifiable. ## Core Features & Use Cases - Stage Detection: Classifies work into scope, evidence, method, experiments, drafting, and review stages, routing each to the appropriate writing skill. - Task Packets & Planning Artifacts: Requires persistent files like plan/outline.md, plan/progress.md, and per-task packets before any manuscript text is written. - Multi-Agent Chapter Gate: Dispatches a separate agent per major chapter with disjoint file ownership and records provenance in plan/chapter-agent-provenance.md. - Review Gates & Audits: Enforces spec-compliance and quality reviews plus a capability-use audit before any medium task is marked complete. - Use Case: When asked to draft a full SCI paper, the Skill first locks the chapter architecture, creates one task packet per chapter, dispatches independent agents, and only claims progress after controller review. ## Quick Start Ask the assistant to draft a full research paper on your topic and let it set up the plan files, chapter architecture, and per-chapter task packets before writing any manuscript text.
