First public release of tagore — a skill for making AI-generated prose sound human.
What's in this release
- SKILL.md (649 lines): full pipeline (voice calibration → draft rewrite → 12-item checklist → 8-dimension score → self-audit → final rewrite), 29-pattern catalog, 8 core principles, worked example.
- 8-dimension scoring rubric with subtotal floors. Mechanics (Directness, Rhythm, Trust, Authenticity, Density) cover how sentences land. Substance (Specificity, Restraint, Voice) covers whether the text actually says something. A piece must clear 56/80 overall AND clear both subtotal floors — no passing on mechanics alone.
- 8-harness compatibility: Claude Code, OpenCode, GitHub Copilot CLI, OpenAI Codex CLI, Gemini CLI, Goose, Cursor, Windsurf.
- install.sh: POSIX-compatible auto-detecting installer with `--all`, `--platform NAME`, `--dry-run`, and `--uninstall` flags. Drops a universal symlink at `~/.agents/skills/tagore/` for cross-tool discoverability.
- examples/essays-with-scores.md: three before/after rewrites (product launch, internal status email, README intro) with full scoring breakdowns.
Install
```bash
curl -fsSL https://raw.githubusercontent.com/apurvrdx1/tagore/main/install.sh | bash
```
Credits
Built from two excellent prior works:
- humanizer 2.5.1 — based on Wikipedia:Signs of AI writing, maintained by WikiProject AI Cleanup.
- stop-slop by Hardik Pandya.
Both MIT-licensed; this merge is too.