external-skill-acquisition

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

• Third-party content exposure detected (high risk: 1.00). The SKILL.md explicitly directs searching and installing community GitHub repos and other public marketplaces (Tier 2 and the curl/gh search examples) and instructs reading and injecting a newly acquired skill's SKILL.md into the current session (read(".claude/skills//SKILL.md")), meaning untrusted third-party skill docs are fetched and executed as agent instructions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs installing and reading external GitHub repos at runtime (e.g., https://github.com/obra/superpowers, https://github.com/vercel-labs/agent-skills, https://github.com/anthropics/skills) and then injects the fetched SKILL.md into the current session, so remotely fetched content would directly control agent prompts.