Skill Dash

September 20, 2026 · View on GitHub

Local dashboard that uses TypeSafe Jev to judge Claude Code and Codex skills (usefulness, redundancy, clarity, action) with transcript evidence and overlap audit; also the pipeline behind the public whichskills.dev census.

At a glance	Details
Source	Source
Maintainer	48Nauts-Operator / 48Nauts. Independently curated; this entry is not an upstream submission or endorsement. TypeSafe did not commission the whichskills.dev run.
Format	Python stdlib + SQLite local server (`server.py`); browser UI on loopback.
Requirements	Python 3. Loader/evidence/static pre-scan need no key. Judgments and overlap audit need `TYPESAFE_API_KEY` (or macOS Keychain `xnaut` / `plugin/typesafe/TYPESAFE_API_KEY`).
License	MIT.
Disclosure	AI-assisted catalog review; no affiliation with the maintainer. Listing is not an endorsement. Source and offline unittest inspected. Live TypeSafe judgments and whichskills.dev regeneration were not run. Distinct from SkillRanker / Skillbox / jev-skill-gate.

When to use

Use it to review your own ~/.claude/skills tree (or any SKILL.md roots) with inspectable Jev distributions and keep/merge/delete decisions stored locally. Prefer SkillRanker for next-step ranking in a live session, or jev-skill-gate for Claude Code skillOverrides gating.

How it works

engine.py posts typed Score/Choice questions to https://api.typesafe.ai/v1/systemone (default jev-1.13.0). Local code owns skill loading, transcript evidence scans, pairwise overlap, SQLite persistence, and the loopback UI. Nothing on disk is deleted by the app; only judgment payloads leave the machine when you run judges.

Get started

git clone https://github.com/48Nauts-Operator/skill-dash.git
cd skill-dash
git checkout fc13176636934faa251bf9b8e57674cc134fcf96
python3 -m unittest discover -s tests -v
# Live dashboard (charges TypeSafe when judging):
# export TYPESAFE_API_KEY=… && python3 server.py --port 3345

Examples and demos

README describes the local skill tree UI and --roots mode for third-party skill repos.
Public census methodology notes for whichskills.dev in scripts/build_site.py (author-reported costs).

Limits and data handling

Binds to 127.0.0.1. Skill bodies and descriptions are treated as data to judge, not instructions (policy text in engine.py). Token/cost figures in upstream docs are author-reported. Live judgment runs send skill state to TypeSafe.

Review and maintenance

Reviewed on 2026-09-21 at commit fc13176: MIT. AI-assisted source review of engine.py, server.py, README, and LICENSE. python3 -m unittest discover -s tests -v: 4 passed. No live TypeSafe calls.