alterlab-remote-compute

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

• Third-party content exposure detected (high risk: 0.70). High risk: the required runtime workflow for the generic REST backend (scripts/dispatch.py → _rest_request() → urllib.request.urlopen() → json.loads(resp.read().decode('utf-8'))) ingests free-form JSON fields from an outsider-controlled HTTP endpoint (e.g., RunPod-style /run and /status/<id> responses) into the agent’s LLM context via the tool’s returned status/handle text.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill issues runtime HTTP calls to RunPod endpoints (e.g., https://api.runpod.ai/v2/<ENDPOINT_ID>/run and https://api.runpod.ai/v2/<ENDPOINT_ID>/status/) to submit and poll jobs that execute user-provided payloads on remote compute, so these URLs are used at runtime to run remote code.