claude-code-guide

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly enables fetching content from public sources—e.g., "claude skills install github:user/repo" and the MCP servers entries for "@anthropic-ai/mcp-server-github" and "@anthropic-ai/mcp-server-puppeteer"—which cause the agent to ingest and read untrusted, user-generated web/GitHub content that could carry indirect prompt injection.