recursive-router

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

• Third-party content exposure detected (high risk: 0.80). This skill explicitly routes prompts to and ingests outputs from external CLIs/models (e.g., codex, kimi, opencode) via scripts like recursive-router-invoke.py and stores routed assistant output under run evidence (/.recursive/run//evidence/router/), so it clearly consumes untrusted third-party model/CLI output that could indirectly inject instructions.