Yoder Skill Auditor

## Quick judgment **Yoder Skill Auditor** is a **specialized, high-utility security scanner for OpenClaw/ClawHub skills**. Based on the description, it appears well-designed for technical teams that need structured, repeatable security checks and trust scoring around skills. It is best suited as a **security gate and ongoing audit tool**, not as a sole source of truth for security. **Overall stance:** Recommended for security-conscious users and teams working heavily with OpenClaw skills, especially in automated or policy-driven environments. --- ## Key strengths - **Deep, domain-specific coverage**: 18 tailored checks that match common real-world threats (credential harvesting, exfil, obfuscated payloads, download-and-execute, privilege escalation, prompt injection, etc.). - **Prompt-injection detection**: Explicit coverage of agent manipulation in documentation ("ignore instructions", role hijacking, hidden HTML), which is often overlooked. - **Trust scoring framework**: 5 dimensions (Security, Quality, Structure, Transparency, Behavioral) with numeric score + letter grade, making it easy to set organizational policies (e.g., block <60 or <40). - **Rich workflows**: Pre-install inspection, single-skill audit, diff audits, trend tracking, benchmarking, and batch scans across all installed skills. - **Automation-ready**: CLI tooling, JSON output, and clear exit codes (0=PASS, 1=REVIEW, 2=FAIL, 3=Error) integrate well into CI/CD and fleet management. - **Validation effort**: 12 test fixtures (8 malicious, 4 clean) and 28 automated assertions give some evidence of reliability and attention to false positives. --- ## Risks and limitations - **Potential overreliance / false sense of security**: Marketing claims such as "zero false positives" and "best-in-class detection" should be treated cautiously; manual review is still needed for high-risk deployments. - **Coverage is pattern-based**: The checks are mainly heuristics (e.g., patterns for curl|bash, base64, eval/exec, sensitive paths). Novel or well-obfuscated attacks may evade detection. - **Ecosystem-specific**: Highly tailored to OpenClaw/ClawHub skills; it is not a general-purpose code security scanner and may miss non-skill-related issues in broader projects. - **Operational fit required**: Designed for users comfortable with shell and Python CLIs; less suitable for non-technical operators without tooling or automation around it. --- ## Recommended scenarios Use **Yoder Skill Auditor** when: - You are **installing or updating skills from ClawHub**, especially from untrusted or third-party authors. - You want a **policy gate in CI/CD** that enforces minimum trust scores and fails on CRITICAL findings. - You manage **many skills across environments** and need batch audits, benchmarking, and trend tracking. - You need to **compare multiple skills** for the same purpose and want security/quality structure to influence the selection. It is less appropriate as a standalone solution for organizations that need **comprehensive, language-level or infrastructure-level security scanning** beyond OpenClaw skills; in those environments it should be combined with broader application security tools.