Nanako0129/sepia
De-AI writing skill for Claude Code, Codex, Grok Build, and Antigravity — narrative-architecture repair for fiction, venue-matched rules for professional prose. Based on StoryScope (arXiv:2604.03136).
What it solves
sepia addresses the problem of "AI-sounding" text, specifically the structural and architectural patterns that make AI-generated content easy for classifiers to detect. Unlike typical humanizers that only change word choice or syntax, sepia targets the deeper narrative architecture of fiction and the specific professional norms of technical documentation to make AI writing indistinguishable from human writing.
How it works
It functions as a portable Agent Skill that can be integrated into AI agents like Claude Code, Codex, Grok Build, and Antigravity. The tool uses a three-pass protocol for fiction and a domain-specific rule set for professional prose:
- Narrative Architecture (Fiction): Fixes high-level tells such as over-explaining themes, overly tidy causal chains, and linear time.
- Discourse Flow (Fiction): Adjusts paragraph sequences and rhythm to remove robotic templates.
- Surface Style (Fiction): Refines vocabulary and syntax to remove clichés.
For professional documents (like release notes, PR replies, and postmortems), it applies a shared checklist to remove "slop" and matches the writing to the specific venue's expectations.
Who it’s for
- Writers and authors using AI to draft fiction who want to avoid AI detection and improve narrative quality.
- Software engineers and technical writers who need to produce professional documentation that doesn't look like a generic chatbot output.
- Users of AI agent platforms (Claude Code, Codex, Grok Build, Antigravity) who want a standardized writing refinement skill.
Highlights
- Architecture-First Approach: Focuses on narrative structure rather than just surface-level word swapping.
- Multi-Platform Support: A single canonical skill that works across multiple AI agent environments.
- Research-Backed: Based on a digest of eleven studies on AI writing patterns and detection.
- Four Core Operations: Provides specific modes for writing, reviewing (diagnosis), refactoring (minimal edits), and recreating (full rewrite).
- Domain-Specific Rules: Tailored guidelines for release notes, PR replies, postmortems, tickets, and technical articles.
Related
- Project
- Project
- Project
- Project