What problem does it solve? Writing short-form Chinese web fiction (盐言故事, 七猫短篇) with AI often produces flat emotion, template-feeling prose, AI-flavored sentence patterns, and inconsistent formatting. This Skill turns a vague short-story request into a disciplined, emotion-first production workflow with genre-specific style packs, reversal design rules, and deterministic quality checks. ## Core Features & Use Cases - Pipeline Adapter Enforcement: Inside a novel-studio repository it never writes prose directly; it converts requirements into run-prompts/*.md files and invokes novel-studio --pipeline, keeping canon production in the verifiable pipeline. - Emotion-First Four-Phase Workflow: Phase 1 locks the target emotion (意难平/反转震撼/爽感释放 etc.), Phase 2 builds the core framework with reversal design and benchmark-book deconstruction, Phase 3 writes scene-by-scene in batches with per-section word-count gates (≥800 chars/section, ≥8000 total), and Phase 4 polishes with AI-pattern and degeneration scanners. - Genre Style Packs & Reference Library: Ships dedicated style packs for 追妻火葬场, 复仇打脸, 总裁豪门, and 宅斗宫斗 plus 20+ reference files covering hooks, dialogue, reversals, villains, and banned AI-flavor phrases. - Deterministic Quality Scripts: check-ai-patterns.js flags high-risk AI sentence patterns and em-dashes, check-degeneration.js detects repetition/truncation/meta-leakage, and normalize-punctuation.js finalizes punctuation. - Use Case: A user says "帮我写一篇盐言故事，追妻火葬场题材，1万字". The Skill loads the 追妻火葬场 style pack, builds the core framework with a reversal and three planted clues, writes the story in 2-3 section batches into 第01章_正文.md, then runs the checker scripts until all blocking findings are resolved. ## Quick Start Ask the AI to write a short web story with a target emotion and genre, for example: use story-short-write to write a 10000-character 盐言 story in the 追妻火葬场 genre with a 意难平 ending.
