integrate-openclaw

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime consult path (Patter consult → local adapter → POST /v1/chat/completions to OpenClaw) forwards the caller’s free-form speech/transcript as request_text into the LLM context via messages[].content (i.e., outsider-authored caller input), creating an indirect prompt-injection surface.

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

• Potentially malicious external URL detected (high risk: 1.00). This skill calls a runtime consult URL (e.g. http://127.0.0.1:8000/consult) which forwards to the OpenClaw chat-completions gateway (e.g. http://127.0.0.1:18789/v1/chat/completions) and injects the returned choices[0].message.content directly into the voice agent as the spoken reply, so remote content fetched at runtime directly controls agent output.