add

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 1.00). The skill explicitly fetches SKILL.md from arbitrary GitHub repositories using the command "gh api repos///contents/skills//SKILL.md" and then analyzes that untrusted, user-generated content to auto-propose conditions that will change agent behavior.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill runtime fetches remote skill descriptions using the GitHub API endpoint gh api repos/<owner>/<repo>/contents/skills/<skill>/SKILL.md --jq '.content' | base64 --decode, and those fetched SKILL.md contents are directly analyzed and used to generate the agent's proposed "condition" (i.e., control prompts/behavior), so this is a runtime external dependency that can control prompts.