eywa

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly calls an external Eywa MCP endpoint (EYWA_URL / eywa-call.sh) and instructs the agent to read room snapshots, messages, tasks, and knowledge (e.g., eywa_start, eywa_status, eywa_knowledge, eywa_msg), which are untrusted/other-agent content that the agent is expected to interpret and could materially influence subsequent actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The eywa-call.sh helper makes runtime POST requests to the MCP endpoint (default: https://eywa-mcp.armandsumo.workers.dev) and extracts .result.content[0].text from the response to feed agent behavior, meaning remote content from that URL can directly control prompts/instructions and is required for the skill to operate.