trellis-session-insight

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

• Third-party content exposure detected (high risk: 0.85). Runtime path: the skill calls the local CLI trellis mem, which indexes and then dumps cleaned dialogue from outsider-authored platform JSONL logs (e.g., Claude/Codex/Pi conversation logs written by other participants), and that dumped free text is inserted into the agent’s LLM context for summarization/quoting.