The Agent Skills Directory

[SAFE]: The skill performs its stated purpose of auditing local skill metadata without any suspicious or hidden behaviors. It adheres to the principle of least privilege by only accessing the specific directory where skills are expected to be installed.
[DATA_EXPOSURE]: The skill accesses the .openclaw/extensions/superpowers directory to read skill definitions. This access is limited to the skill's own operational scope and does not target sensitive user files, credentials, or system configurations. No network operations are present, precluding data exfiltration.
[DYNAMIC_EXECUTION]: The Python script correctly uses yaml.safe_load() for parsing skill frontmatter. This is a security best practice that prevents potential arbitrary code execution vulnerabilities associated with unsafe YAML deserialization.
[INDIRECT_PROMPT_INJECTION]: The skill processes descriptions from other potentially third-party skills. It mitigates the risk of indirect prompt injection by tokenizing the text and applying a Jaccard similarity model, which effectively strips the instructional structure required for adversarial prompts to influence the agent reviewing the conflict report.

skill-conflict-detector