cloudflare

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.90). At least two URLs match high-risk patterns: a direct .sh installer (can be abused for "curl | sh" malware installs) and a GitHub URL embedding a token (credential leakage / supply-chain risk), so these sources should be treated as suspicious until verified.

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

• Third-party content exposure detected (medium risk: 0.65). SKILL.md’s “Retrieval Sources” section directs the system to fetch Cloudflare public docs and changelog pages at runtime (e.g., via “Search https://developers.cloudflare.com/ directly”), so the LLM context can include outsider-authored free text from those web pages.

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

• Potentially malicious external URL detected (high risk: 0.90). The Agents SDK examples register and call an external MCP server at https://mcp.example.com (env.MCP_SERVER_URL) at runtime to fetch tool definitions which are then passed into the model (controls agent prompts/tools), making this a runtime dependency that can change agent behavior.