story-deslop

Detect and rewrite AI-flavored prose patterns in Chinese web novel manuscripts.

116|21|Updated Jul 6, 2026
One-click install
npx skills add https://github.com/Xiaoyangy/novel-studio --skill story-deslop-xiaoyangy
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: story-deslop
Source: https://github.com/Xiaoyangy/novel-studio/tree/main/skills/story-deslop
Command: npx skills add https://github.com/Xiaoyangy/novel-studio --skill story-deslop-xiaoyangy

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? AI-generated fiction often reads as overly polished, formulaic, and explanatory, with banned phrases, rigid sentence patterns, and summary-style endings that break immersion. This Skill scans Chinese web novel text for these AI writing fingerprints and rewrites them into natural, colloquial prose while preserving plot, characters, and foreshadowing. ## Core Features & Use Cases - AI-Flavor Detection and Grading: Scans text against a banned-word list and seven gates (banned words, sentence patterns, told emotions, uniform rhythm, dialogue tone, ending elevation, narrator explanation), then classifies severity as light, medium, or heavy with quantitative metrics. - Gate-Based Rewriting: Applies minimal, deletion-first edits that replace abstract emotion with concrete action, break up parallel structures, and remove authorial explanation, with deletion-ratio caps (15%/25%/35%) to prevent over-editing. - Deterministic Script Checks: Runs Node.js scripts to detect high-risk comparison sentence patterns, em-dashes, stuttered short sentences, long paragraphs, model degeneration (verbatim repetition, truncation, placeholder leakage), and to normalize punctuation. - Use Case: Paste a chapter that reads "too AI" or point to a manuscript file; the Skill produces a detection report, rewrites the file in place, and outputs a change log with before/after comparisons and character-count deltas. ## Quick Start Ask the assistant to run /story-deslop on your chapter file to detect AI writing patterns and rewrite the text naturally in place.

Frequently Asked Questions about story-deslop

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I remove AI writing style from Chinese web novel text?

Run the detection phase to scan for banned words, formulaic sentence patterns, and told emotions, then apply gate-based rewriting that replaces abstract description with concrete action. Severity grading (light, medium, heavy) determines how many of the seven gates are applied.

What AI writing patterns does this detection cover?

It covers banned vocabulary, the "not A but B" comparison pattern, "带着" adverbial clauses, direct emotion statements, uniform paragraph rhythm, formulaic dialogue tags, summary-style endings, and narrator explanation or god-view spoilers. Scripts also flag em-dashes, stuttered short sentences, and overlong paragraphs.

Can I run detection only without modifying my manuscript?

Yes. Requesting check-only mode runs Phase 1 and Phase 2, producing a problem-marker table with gate classifications and severity grading without editing the original text or writing any files.

Does the rewriting preserve plot and character details?

Yes. The Skill only changes how things are said, never what is said, and forbids deleting foreshadowing, hooks, character traits, or plot turns. Deletion ratios are capped at 15%, 25%, or 35% depending on severity, with uncertain items flagged for manual review.

How do I whitelist terms that trigger false positives?

Create a .deslop-whitelist file in the project root with one exempt term per line, using # for comments. Matching substrings are skipped during banned-word counting, which protects worldbuilding terms, nicknames, and intentional rhetoric.

What are the limitations of automated AI-flavor removal?

Blocking degeneration findings like verbatim repetition or truncation cannot be fixed by rewriting style and require regenerating the affected passage. After three full rescan rounds with ten or more remaining changes, the text is flagged for manual review instead of further automated edits.