external-model-selection

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

• Third-party content exposure detected (high risk: 1.00). The SKILL.md explicitly instructs invoking external third-party models via claudish (e.g., "Task tool → golang-architect (PROXY MODE)" and Bash examples like claudish --model minimax/minimax-m2 writing outputs to ai-docs/sessions/.../output/*.md), so the agent ingests untrusted external model outputs as part of its workflow and those outputs can materially influence decisions and actions.