command-center

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

• Third-party content exposure detected (high risk: 0.75). At runtime, lib/server.js reads OpenClaw session transcript JSONL files from the user’s workspace (e.g., agents/main/sessions/*.jsonl) and extracts message content text into API responses like /api/state and /api/sessions (via src/sessions.js → getSessionDetail() / getSessions()), which can include outsider-authored free text (e.g., Slack/Telegram user messages) that is then ingested into the agent’s LLM context when the agent consumes those API responses.