debug

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

• Third-party content exposure detected (medium risk: 0.30). In /debug Phase 0-2, the skill ingests the user-provided problem description (free text) and then the LLM uses it to plan and run probe commands (Bash/curl/grep/ls/trace), while also passing a “finding packet” to /seek-verdict; this means outsider/authored text is directly used at runtime rather than requiring the agent to first select a specific trusted item.