← ClaudeAtlas

skill-verifylisted

Vet a skill or setup before installing. Combines lint + security in one pass. SAFE/CAUTION/UNSAFE verdict. Use when the user wants to check if a downloaded or cloned skill is safe to install.
redhat-community-ai-tools/harness-eval · ★ 25 · AI & Automation · score 79
Install: claude install-skill redhat-community-ai-tools/harness-eval
<!-- evaluator-ignore: content/allowed-tools-auto-approve --> # Skill Verify Vet a skill or setup for security and quality issues before installing. ## Hard Rules 1. **Run the CLI command.** Do not skip the scan. 2. **Report the verdict clearly.** SAFE, CAUTION, or UNSAFE. 3. **Show all findings with fix suggestions.** ## Step 1: Get the Path Ask the user for the path to the skill or setup directory to verify. If they provide a URL, clone it first: ```bash git clone --depth 1 <url> /tmp/skill-to-verify ``` ## Step 2: Run the Scan ```bash uvx --from harness-eval harness-eval skill-verify <path> --format json ``` If `uvx` is not available, fall back to `pip install harness-eval` and use `harness-eval` directly. Read the JSON output. ## Step 3: Present the Results Report the verdict prominently: **SAFE**, **CAUTION**, or **UNSAFE**. List all findings with their severity, rule, message, and fix suggestion. If UNSAFE, tell the user not to install without addressing the findings. If CAUTION, tell the user to review the warnings before installing. If SAFE, tell the user it's safe to install.