Web Novel Studio

Problem: Continuity and platform compliance in web-novel writing

Writing a long web novel is not just “generate one chapter”: premise, character state, foreshadowing, pacing, and target-platform rules all have to be maintained across sessions. A single model call tends to drift into contradictions, obvious AI phrasing, or platform red lines. This skill frames web-novel creation as an executable workflow: read the book config from doc.md, then lazily load references/ modules for creation memory, character relations, foreshadowing, worldview, outline, drafting, polish, platform rules, and style learning.

How it works: controller + lazy-loaded modules

• Step 0 collects configuration: genre, platform, protagonist, word count, selling point, and other missing fields.
• Steps 1–8 build continuity infrastructure: creation memory, character relations, foreshadowing, general rules, topic planning, worldview, character building, and a 30-chapter outline.
• Steps 9–11 produce the draft: read memory and craft techniques before writing, update memory after writing, then polish, run platform compliance checks, reduce AI flavor, and emit a report.
• Step 12 closes the style loop: uploaded snippets are distilled into emotion, tone, phrasing, and scene patterns in style-library.md for later chapters.

It behaves more like an engineering pipeline: report first, draft second, with step-level status instead of claiming a publish-ready final version.

Boundaries

• The output is positioned as a high-completion draft plus revision guidance, typically requiring about 70% author revision.
• Platform rules differ: the material stresses that Jinjiang restricts AI-generated narrative prose, so use it for outlines, characters, and scene setup; other platforms still need manual review.
• It does not make legal judgments for the user; when content crosses a line, it flags rewriting, manual handling, or skipping.