learn

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

• Third-party content exposure detected (high risk: 0.85). The required workflow can ingest outsider-authored free text when the user supplies a public URL or pasted notes (“/learn ” via WebFetch, or “/learn + pasted notes” used directly), and that fetched/pasted prose is then distilled into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs the agent to use WebFetch to fetch a user-supplied URL at runtime and to pull content from that page to produce the SKILL.md, so the user-provided URL (fetched via WebFetch) can directly control the agent's generated instructions.