critical-skillgovern

A cross-platform skill quality governance framework — battle-tested on 128 skills across 40 categories.

中文版

The Problem

The more skills you have, the harder it is for an LLM agent to pick the right one. This isn't a model limitation — it's a routing problem. Skill descriptions lack the signal LLMs need for accurate selection.

Critical SkillGovern gives you a battle-tested workflow to keep your skill library accurate at any scale.

Why You Should Care

The third-party skill packs you downloaded — Superpowers workflow pack, Baoyu creative pack, Matt Pocock TypeScript pack, design skill packs — likely have routing issues:

📖 Full case study → examples/third-party-packs.md

Before governance:

【不适用场景】不需要生成创意内容/可视化时    ← Too vague, no alternative

After governance:

【不适用场景】For simple charts/diagrams → use excalidraw;
For AI art rather than structured infographics → use image generation
            ← Points to specific alternatives

The difference: before, the model guesses wrong ~30% of the time. After, misfires drop to ~5%.

Core Pipeline

Routing Diagnosis → Critical Questioning → User Approval → Execution → Verification → Final Report

Supported Platforms

Quick Start

# 1. Review all your skills
python scripts/batch-review.py --path ~/.hermes/skills --format hermes

# 2. Review a specific category (e.g., third-party packs)
python scripts/batch-review.py --path ~/.hermes/skills/creative --format hermes

# 3. Export CSV report
python scripts/batch-review.py --path ~/.hermes/skills --format hermes --csv report.csv

What to Fix (Priority Order)

Project Structure

critical-skillgovern/
├── README.md                         # English (this file)
├── README.zh-CN.md                   # 中文版
├── LICENSE                           # MIT
├── docs/
│   ├── methodology.md                # Core methodology (6-stage pipeline)
│   ├── routing-optimization.md       # Routing optimization guide
│   └── prompts/
│       └── classify-skills-v2.txt    # Skill classification prompt v2.1
├── adapters/
│   ├── hermes.md                     # Hermes Agent guide
│   └── codex.md                      # Codex CLI guide
├── dsh-plugin/                       # DeepSeek Harness plugin
│   ├── package.json                  # DSH bundle manifest
│   ├── cordis.patch.yml              # Cordis configuration
│   └── index.js                      # Plugin entry (3 tools)
├── templates/
│   ├── diagnostic-card.md            # Diagnostic card template
│   └── skill-audit-report.md         # Audit report template
├── scripts/
│   ├── batch-review.py               # Batch review script
│   └── validate-routing.py           # Routing validation script
├── examples/
│   ├── before-after.md               # Before/after optimization examples
│   ├── full-audit-example.md         # Full 128-skill audit walkthrough
│   └── third-party-packs.md          # Third-party pack case study
└── CONTRIBUTING.md                   # Contribution guide

Real-World Impact

Tested on a 128-skill, 40-category Hermes Agent library with Superpowers, Baoyu, and 16 design skill packs:

Why Critical SkillGovern?

DeepSeek Harness Plugin

Install the DSH plugin to get 3 skill governance tools directly in your agent:

# From GitHub
dsh plugin --profile web add github:Xiaopeng212321414321413231/critical-skillgovern#path:dsh-plugin

# Or from local clone
dsh plugin --profile web add ./dsh-plugin

Listed on DSH 1024Store.

Contributing

PRs and issues welcome! See CONTRIBUTING.md.

License

MIT — free to use, modify, and distribute.