Sepia Repairs AI Writing’s Structural Tells

Sepia targets a meaningful weakness in generic AI “humanizers”: fluent wording cannot fix predictable story structure. Its strongest idea is treating human-like writing as calibrated variation, not a checklist of anti-AI tricks.

Fiction passes move from narrative architecture to discourse flow, then surface style. The repository includes a 30-feature diagnostic rubric and model-specific corrections. Professional writing rules are matched to document type instead of applying one generic tone. Its research digest grounds the workflow in StoryScope and related studies, though the skill’s real-world effectiveness still needs independent evaluation. Support for Claude Code, Codex, Grok Build, and Antigravity makes it unusually portable for an instruction-only tool.