Repo

sepia

Provides write, review, refactor, and recreate operations that de-AI prose using research-backed fiction passes and domain rules for professional writing.

GitHub ↗Aug 28, 2026Paper ↗

  • llm
  • writing-tools
  • prompt-engineering
  • agent-skills
Full pipeline diagram showing method from narrative representation to feature classification
Full pipeline diagram showing method from narrative representation to feature classification

What it is

Portable agent skill for de-AI writing that targets narrative structure in fiction and venue-specific rules in professional prose across Claude Code, Codex, Grok, and Antigravity. Diagnose, minimally edit, or fully rewrite prose to remove AI tells by fixing fiction's narrative architecture first and applying venue-matched rules for docs like PR replies, postmortems, and release notes.

Key results

  • Works across Claude Code, Codex, Grok Build, and Antigravity
  • Uses 4 operations: write, review, refactor, recreate
  • StoryScope basis: 93.2% macro-F1 from narrative features alone

Shows AI-writing detection can hinge more on narrative structure than surface style, so effective revision must operate above wording.

One useful ML email a week.

Worth your time, not all of your time.

We track new ML work across papers, code, and data, then keep only the pieces worth a closer look. Selected, not collected.

Papers.Data.Code

The Weekly ML Signal

Selected, not collected.

One useful ML email a week: the papers, repos and datasets worth your attention.

Worth your time, not all of your time.