auditing

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill's Input Normalization step explicitly performs runtime fetches of external GitHub repositories and release archives (e.g., GitHub URLs like https://github.com// and .zip/.tar.gz release URLs) to shallow-clone / download the target, and those fetched SKILL.md/agent files are then parsed and used to drive agent behavior (prompts/dispatch), so remote content can directly control prompts and must be treated as a runtime external dependency.