← Back to Skills Marketplace
skill-usefulness-audit
by
gongyu0918-debug
· GitHub ↗
· v0.2.7
· MIT-0
292
Downloads
7
Stars
0
Active Installs
10
Versions
Install in OpenClaw
/install skill-usefulness-audit
Description
Finds unused, overlapping, risky, or under-evidenced agent skills and produces a cleanup report.
Usage Guidance
This skill appears to be a legitimate, local audit tool: it scans installed skill files, usage exports, and conversation history and produces a cleanup report. Before running it, review the included Python script (scripts/skill_usefulness_audit.py) yourself and run it in a trusted environment. Ensure the skills root and any usage/history files you provide do not contain secrets you don't want read by the script. If you require strictly manual-only runs, confirm your platform honors the skill's manual-invocation guidance (or set disable-model-invocation where possible) so the agent cannot call this skill autonomously.
Capability Analysis
Type: OpenClaw Skill
Name: skill-usefulness-audit
Version: 0.2.7
The skill bundle is a comprehensive utility designed to audit the usefulness, quality, and security risk of other installed skills. The core logic in `scripts/skill_usefulness_audit.py` performs static analysis, usage tracking, and risk scanning (detecting patterns like `curl|bash`, `eval`, or access to `.ssh` in other bundles). While the script requires broad read access to the skills directory and history logs to function, its behavior is strictly aligned with its stated purpose, and there is no evidence of data exfiltration, unauthorized execution, or malicious intent.
Capability Assessment
Purpose & Capability
The name/description (audit installed skills for usefulness/risk) matches the provided artifacts: a detailed SKILL.md, an ablation protocol, scoring rubric, and a Python audit script. The included script and references are proportionate to implementing a local audit tool.
Instruction Scope
Runtime instructions explicitly require collecting installed-skills metadata, usage files, history transcripts, and local references, then scoring and producing reports. Those actions are within scope for a skill-audit tool. The SKILL.md emphasizes manual trigger only and describes which files to read; it does not instruct uploading data to external services.
Install Mechanism
There is no install spec (instruction-only) and the repository includes a Python script the user runs locally. No external downloads or package installs are required by the bundle itself, which reduces supply-chain risk.
Credentials
The skill declares no required environment variables, binaries, or credential access. It reads local skill files, usage exports, and history transcripts for audit evidence — appropriate for this purpose. The code contains risk-detection regexes for protected-path patterns (e.g., .ssh, .aws) but does not itself request those credentials.
Persistence & Privilege
The skill is not always-enabled and is user-invocable. SKILL.md requests manual invocation only. Registry metadata shows model-invocation is allowed by default (disable-model-invocation: false), which could allow autonomous invocation depending on platform enforcement; agents/openai.yaml sets allow_implicit_invocation: false. This is a minor policy mismatch to be aware of but not an indication of malicious intent.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install skill-usefulness-audit - After installation, invoke the skill by name or use
/skill-usefulness-audit - Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.2.7
Fix script burden detection, refresh stale ablation candidates, and improve mixed-language context estimates
v0.2.6
Add quality burden scoring, cost-efficient ablation planning, and model-cost reduction estimates
v0.2.5
Improve README positioning, shorten ClawHub summary, and add score/report breakdowns
v0.2.4
Use lifetime community signals, improve history fallback recall, remove redundant risk reads, and validate bundle drift in CI
v0.2.3
Fix duplicate-name evidence routing, make bundle sync CRLF-safe, add missing-file warnings, and tighten risk scanning
v0.2.2
Reduce ClawHub scan false positives, rename protected-path risk labels, and keep audit behavior intact
v0.2.1
Fix duplicate-name evidence routing, doc-only risk false positives, host-prompt history inflation, and README quickstart
v0.2.0
Recency-weighted evidence, confidence score, risk scan, and offline community metrics
v0.1.1
Compatibility expansion and bilingual introduction
v0.1.0
Initial release
Metadata
Frequently Asked Questions
What is skill-usefulness-audit?
Finds unused, overlapping, risky, or under-evidenced agent skills and produces a cleanup report. It is an AI Agent Skill for Claude Code / OpenClaw, with 292 downloads so far.
How do I install skill-usefulness-audit?
Run "/install skill-usefulness-audit" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is skill-usefulness-audit free?
Yes, skill-usefulness-audit is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does skill-usefulness-audit support?
skill-usefulness-audit is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created skill-usefulness-audit?
It is built and maintained by gongyu0918-debug (@gongyu0918-debug); the current version is v0.2.7.
More Skills