lossless-claw-use

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

• Third-party content exposure detected (high risk: 0.85). The skill's agent tools (lcm_grep, lcm_describe, lcm_expand_query) explicitly read summaries, raw conversation messages, and stored large file content from the LCM SQLite DB (see reference/agent-tools.md and reference/architecture.md), which are untrusted user-generated content across conversations and are used to drive sub-agent expansions and subsequent actions—exposing the agent to potential indirect prompt injection.