hyv-evaluate-local

SkillAI & models

Run the optional deterministic local writing evaluation composite without model calls.

Available today. Use it from your connected AI after setup.

Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

Then ask your AI: use the hyv-evaluate-local skill

What this skill tells your AI

The instructions your AI receives, as published by shashank-sn/holdyourvoice in skills/hyv-evaluate-local/SKILL.md and read by ahel’s review.

Evaluate a caller-supplied candidate against an input and separate local user and AI-shadow paragraph fixtures. The module trains only on the paragraph IDs assigned to the train split. It does not generate prose, transmit samples, retain text, or establish authorship.

hyv evaluate-local input.md candidate.md user-paragraphs.json ai-shadow-paragraphs.json

Each JSON file is an array of { "paragraph_id": "stable-id", "text": "..." }. Keep variants under the same paragraph ID; HYV splits by ID before training. The report contains a train-only TF-IDF logistic-regression proxy, content F1, AI-tell change, and 9-d stylometric cosine. Treat every number as a local evaluation signal, not a publishing decision or identity claim.

Signals

GitHub stars
32
Last commit
Sep 2026
Advanced
Item type
skill
Key
hyv-evaluate-local
Source
github.com/shashank-sn/holdyourvoice