alibabacloud-agentloop-experience

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

• Third-party content exposure detected (high risk: 0.75). Outsider-authored free text can enter the agent’s LLM context via the runtime path where scripts/search_context.js calls the configured external AgentLoop SearchContext endpoint and returns results[].content (from upstream) as JSON, which the agent then uses as “context”.

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

• Potentially malicious external URL detected (high risk: 0.90). The runtime CLI (scripts/search_context.js) issues POST requests to the configured AGENTLOOP_RECALL_ENDPOINT and the returned results are injected into agent prompts as context, so remote content from that endpoint can directly control agent behavior (AGENTLOOP_RECALL_ENDPOINT).