What problem does it solve? Writing long non-fiction documents such as manuals, whitepapers, reports, and training materials often suffers from inconsistent structure, unmet word-count targets, AI-flavored prose, and unreliable delivery. This Skill enforces a deterministic, file-based workflow where every stage is validated by scripts instead of conversational memory. ## Core Features & Use Cases - Makefile-driven pipeline: Run prepare, write, quality, and deliver stages with a single make WORKDIR=<dir> command; each stage is gated by deterministic Node.js check scripts. - Hard quality gates: Enforces Chinese three-level heading conventions, word-count contracts, placeholder cleanup, anti-AI-cliché writing shape checks, and chapter-ledger truth verification. - Lark document delivery: Merges chapter Markdown into a final manuscript, creates a Lark (Feishu) cloud document via lark-cli, and verifies content consistency through read-back comparison. - Use Case: Ask the assistant to write a 30,000-character corporate training manual; it freezes requirements into a checklist, plans the outline, writes chapters in batches, passes all gates, and delivers a verified Lark document link. ## Quick Start Ask the assistant to use the doubao-book-writer skill to create a new non-fiction book project in a working directory and run the make pipeline to prepare, write, and deliver it.
