source-driven

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

• Third-party content exposure detected (high risk: 0.70). The SKILL.md "Fetch Official Docs" workflow (step 2) explicitly instructs the agent to use WebFetch to retrieve public third‑party documentation pages and to read/interpret those pages to drive implementation decisions, which exposes it to untrusted public web content that can materially influence actions.

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

• Potentially malicious external URL detected (high risk: 0.70). This skill explicitly requires fetching dependency-specific documentation pages at runtime via WebFetch (e.g., https://docs.example.com/v3/api#method), and those fetched docs directly guide the agent's prompts/instructions for implementing code, so the external URL content controls runtime behavior.