reference-anthropic-sdk

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

• Third-party content exposure detected (high risk: 0.90). The skill's SKILL.md explicitly lists WebSearch/WebFetch in the built-in tools and gives an MCP/playwright example ("Open example.com and describe what you see" in section 2-7 and the tools table in 2-4), so the agent is expected to fetch and interpret public third‑party web content which could carry untrusted instructions.

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

• Potentially malicious external URL detected (high risk: 0.90). The MCP example shows running "npx @playwright/mcp@latest" (which fetches and executes remote code from npm at runtime — e.g. https://www.npmjs.com/package/@playwright/mcp), so the skill can invoke external code during execution and thus depends on remote content that runs code affecting the agent.