GitHub radar
Sepia: De-AI Writing Skill for Claude Code and Codex
A portable agent skill for Claude Code, Codex, Grok Build, and Antigravity that fixes the structural patterns that give AI-generated text away — narrative architecture in fiction, and venue-matched rules in professional documents.
Sepia is a portable agent skill for Claude Code, Codex, Grok Build, and Antigravity that repairs AI-generated text at the layer research shows actually gives it away. A 2026 study (StoryScope, 61,608 stories) found that a classifier using narrative-structure features alone detects AI fiction at 93.2% accuracy — swapping word choice barely moves that number. Sepia targets the structural tells: narrator explaining the theme, single-track causal plots, emotions rendered only as bodily sensation, resolved endings, linear time. For fiction it runs a three-pass protocol: narrative architecture first, then discourse flow, then surface style. For professional documents — release notes, PR replies, postmortems, tickets, technical articles — each type gets a dedicated thin rule file on top of a shared checklist. Four operations: write, review (diagnose only), refactor (minimal edits), recreate (full rewrite). Installs via the Claude Code or Codex plugin marketplace in one command.
Why a vibe-coder should care
If you use AI agents to write fiction or professional documents and want the result to sound like a person, not a language model, Sepia works at the level that actually matters. Word-choice swapping is table stakes; Sepia goes to the structure underneath.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install the sepia skill from https://github.com/Nanako0129/sepia into Claude Code at user scope via /plugin marketplace add Nanako0129/sepia, then show me how to run /sepia review on one of my files.
▌ More finds