Lightweight Novel Collaboration

Problem

In long-form novel collaboration, models can ingest too many setting files, summaries, and history snapshots at once, causing context pollution, state drift, and hard-to-audit conclusions. When every continuation pass relies on the full source of truth, the AI can mix old snapshots, temporary discussion, and current text, making it hard to know which setting is actually valid.

How It Works And Where It Fits

• Route first, then execute: choose the matching workflows/... for new novels, takeover, continuation, or chapter audit instead of loading every agent upfront.
• Workflows own boundaries; agents execute steps: a workflow defines entry conditions, required files, project structure, initialization, execution order, and gates. Roles such as planner, writer, observer, settler, auditor, and reviser are read only when needed.
• Traceable state: current text outranks older snapshots, conclusions should trace to a fact_id or chapter number, and the workflow avoids fake “already synced” claims.

This skill fits iterative novel projects that need continuation, auditing, and state maintenance. It is not intended for one-shot short drafts, and rule changes should stay in the skill documentation rather than the novel-writing flow.