Nanako0129/sepia

De-AI writing skill for any Agent Skills-compatible agent (77+ via the Skills CLI), with native plugins for Claude Code, Codex, Grok Build, and Antigravity....

View on GitHub ↗Jump to charts ↓Open shareable report →

Nanako0129/sepia has 3k+ stars and 200+ forks on GitHub, is written in Python, and was last pushed on Oct 15, 2026. Data as of . Signed-in members get hourly updates — create a free account.

Summary Information

Updated 60 minutes ago
Added to GitGenius on October 6th, 2026
Created on August 28th, 2026
Open Issues & Pull Requests: 5 (+0)
GitHub issues: Enabled
Number of forks: 210
Total Stargazers: 3,221 (+0)
Total Subscribers: 5 (+0)

Repository Insights (GitGenius)

Median issue/PR response: 0.5 hours
Mean response time: 3.6 days
90th percentile: 15.9 days
Tracked items: 18

Charts & Analytics

Fetching additional details & charts...

Issue Activity (beta)

Open issues: 4
New in 7 days: 0
Closed in 7 days: 0
Avg open age: 18 days
Stale 30+ days: 0
Stale 90+ days: 0

Recent activity

Opened in 7 days: 0
Closed in 7 days: 0
Comments in 7 days: 0
Events in 7 days: 0

Top labels

  • enhancement (2)
  • bug (1)

Most active issues this week

Sign in to see which issues are moving.
Sign in

Detailed Description

Sepia is a de-AI writing skill for agent-compatible systems that repairs AI-generated text by addressing narrative architecture rather than surface-level word choice.

The tool solves the problem that popular humanizers focus on syntax and vocabulary, yet research shows AI fiction remains detectable through structural patterns: unexplained themes, overly tidy causality, limited emotional range, absence of real-world grounding, and linear time progression. Sepia applies a three-pass protocol targeting narrative architecture first (loosening causal chains, mixing emotion modes, adding real references), then discourse flow (varying rhythm, fixing pacing), and finally surface style. It diagnoses AI fingerprints across narrative and sentence-level layers for known model families including Claude, GPT, Gemini, and DeepSeek, using measured features from research and vendor guidance where available.

The tool works as a portable Agent Skill compatible with any agent supporting the standard specification, with native plugin packaging for Claude Code, Codex, Grok Build, and Antigravity. It offers four operations: write, review (diagnosis only), refactor (minimal edits), and recreate (full rewrite). Professional documents receive venue-matched rules for release notes, pull request replies, postmortems, tickets, and technical articles. Developers should choose this if they need AI-generated fiction or professional prose that passes structural scrutiny rather than just surface polish, and if they work within agent-compatible systems. The single canonical skill definition avoids per-platform forks.

Development activity shows sustained engagement with the research foundation underlying the tool's approach, as evidenced by the research directory containing digested studies that inform the three-pass protocol. The project maintains language support across English and Chinese variants, indicating attention to international usability. Native plugin implementations for multiple agent platforms demonstrate commitment to practical integration beyond the base specification.