Skill

通过 python3 脚本调用 clawhub 搜索技能候选，安装到临时目录后用 skill-shield 做安全扫描，再按安全 40%、质量 30%、相关性 30% 评分排序，输出 Markdown 对比报告和结构化 JSON，并卸载未推荐的技能；前置条件是 clawhub CLI 已安装并认证、skill-shield 的 scan.py 可访问，且需 Python 3.8+。

yx2601816404-sys

热度 76.92

最近更新 2026-04-29 12:17:35

分类：搜索依赖：代码执行 / 读写文件 / 第三方 API应用场景：推荐 / 研究 / 报告撰写

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add skill-store

Metadata

name: skill-store
description: Smart skill installation advisor for ClawHub. Searches for skills matching your needs, evaluates candidates on security (via skill-shield), code quality, and documentation, then produces a comparison report with a recommendation. Use when: looking for a skill to do something specific, comparing similar skills, or wanting a safety-checked recommendation before installing. Zero external dependencies.

Skill Store — Smart Installation Advisor

Find the right skill without the guesswork. Searches ClawHub, evaluates candidates on security + quality, and recommends the best one.

Usage

python3 scripts/evaluate.py "your search query" [options]

Examples

# Find a weather skill
python3 scripts/evaluate.py "weather forecast"

# Evaluate top 3 only, save reports
python3 scripts/evaluate.py "image generation" --top 3 --output-dir ./reports

# Use a specific skills workspace for clawhub
python3 scripts/evaluate.py "security audit" --workdir /path/to/workspace

Options

Option	Default	Description
`--top N`	5	Number of candidates to evaluate
`--output-dir DIR`	(stdout)	Save report.md and report.json to DIR
`--workdir DIR`	/tmp/skill-store-eval-*	Working directory for clawhub install
`--scanner PATH`	auto-detect	Path to skill-shield's scan.py
`--keep-all`	false	Don't uninstall non-recommended candidates

Output

Prints a Markdown comparison report to stdout. With --output-dir, also writes:

report.md — human-readable comparison
report.json — structured evaluation data

How It Works

Searches ClawHub for skills matching your query
Installs top N candidates into a temporary directory
Runs skill-shield security scan on each candidate
Evaluates code quality (lines of code, documentation, tests, structure)
Scores and ranks candidates (security 40%, quality 30%, relevance 30%)
Generates comparison report with recommendation
Uninstalls non-recommended candidates (unless --keep-all)

Scoring

Dimension	Weight	What's measured
Security	40%	skill-shield rating, findings count, permission audit
Quality	30%	Code lines, SKILL.md completeness, README, tests
Relevance	30%	clawhub search score (normalized)

Requirements

clawhub CLI installed and authenticated
skill-shield scan.py accessible (auto-detected from sibling directory or via --scanner)
Python 3.8+

Exit Codes

Code	Meaning
0	Recommendation made successfully
1	No candidates found or all evaluations failed