← Back to Skills Marketplace
Skill Test
by
Weiwei Fan
· GitHub ↗
· v0.1.1
· MIT-0
149
Downloads
0
Stars
1
Active Installs
2
Versions
Install in OpenClaw
/install skills-test
Description
Evaluate and QA a skill before release on ClawHub, skills.sh, and similar directories. Includes the bundled static evaluator `scripts/eval_skill.py` plus gui...
Usage Guidance
This package appears coherent and non-malicious, but treat it like any tool that reads and parses local files: 1) Inspect scripts/eval_skill.py before running (there is a visible typo/truncation that may crash the script). 2) Run it in an isolated disposable workspace or container so the evaluator can only read the target skill directory. 3) If you plan to automate evals in CI, run the script on a small sample first and fix the code bug (the 'metada' typo / truncated function) to avoid false failures. 4) If you need runtime rubric grading that contacts external model providers, confirm and provide only the intended API keys to those providers — this package itself does not require credentials.
Capability Analysis
Type: OpenClaw Skill
Name: skills-test
Version: 0.1.1
The skill bundle is a legitimate utility designed for the static analysis and quality assurance of other OpenClaw skills. The core logic in `scripts/eval_skill.py` is a safe Python script that performs regex-based checks on Markdown files and verifies local file existence without any network activity, shell execution, or obfuscation. Furthermore, the documentation in `SKILL.md` and `references/sandbox-testing.md` explicitly promotes security best practices, instructing the AI agent to treat third-party code as untrusted and to use isolated environments for testing.
Capability Assessment
Purpose & Capability
Name, description, SKILL.md instructions, and the bundled scripts/evidence files align: the skill is a skill-evaluator and bundles scripts/eval_skill.py to perform static checks. Required binary is only python3 and there are no unrelated environment variables or config paths.
Instruction Scope
SKILL.md explicitly instructs the assistant to read the target SKILL.md and run python3 scripts/eval_skill.py against a target skill and to prefer isolation. Those instructions stay within the stated purpose. The bundled evaluator performs filesystem reads and regex parsing of frontmatter and docs (expected). Note: the included scripts/eval_skill.py listing contains a clear coding issue (a typo 'metada' / truncated code near platform-readiness checks) which may cause runtime exceptions — this is a bug, not malicious behavior, but you should inspect/fix before running.
Install Mechanism
No install spec and no external downloads; this is instruction-only with a bundled Python script. Risk is low from install mechanics. The only code executed if you follow instructions is the local Python script; there are no network calls or archive extracts shown in the provided files.
Credentials
No environment variables, primary credentials, or config paths are requested. That is proportionate for a static skill-evaluator.
Persistence & Privilege
The skill does not request always:true, does not declare persistence, and provides no indication it will modify other skills or system-wide agent settings. It performs read/analysis of files in the target skill directory only (as expected).
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install skills-test - After installation, invoke the skill by name or use
/skills-test - Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.1.1
- Added bundled static evaluator script (`scripts/eval_skill.py`) for out-of-the-box quality and readiness checks.
- Added support for user-defined deterministic and LLM rubric grading via optional workflows, clarified that only the static evaluator is bundled.
- Introduced an example `eval.yaml` to guide skill regression test suite creation.
- Updated documentation for clearer separation of bundled vs. optional grading tools, including revised prerequisites and notes about LLM/API key handling.
- Clarified typical use cases, evaluation modes, and workflow steps for skill QA prior to publication.
v0.1.0
Initial release of skill-test — a toolkit for evaluating and QA'ing agent skills before release on ClawHub, skills.sh, etc.
- Provides static and rubric-based evaluation of agent skills via `python3 scripts/eval_skill.py <skill-path>`
- Supports trigger tests, deterministic graders, LLM scoring, and regression tracking
- Offers two core modes: Trial Mode (first-pass trust) and Publish Evaluation Mode (release readiness)
- Details clear workflows, assistant responsibilities, and next test recommendations
- Explicitly documents when to use (and not use) this skill for skill QA
- Lists prerequisites and offers sample prompts and output/report structure
Metadata
Frequently Asked Questions
What is Skill Test?
Evaluate and QA a skill before release on ClawHub, skills.sh, and similar directories. Includes the bundled static evaluator `scripts/eval_skill.py` plus gui... It is an AI Agent Skill for Claude Code / OpenClaw, with 149 downloads so far.
How do I install Skill Test?
Run "/install skills-test" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Skill Test free?
Yes, Skill Test is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Skill Test support?
Skill Test is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Skill Test?
It is built and maintained by Weiwei Fan (@fwwdn); the current version is v0.1.1.
More Skills