External Model

Shells out to a different AI model via one of three external agentic CLIs — opencode, cursor-agent, kiro-cli — and returns its answer. Four common requests all reduce to this same mechanism, just framed differently:

1. Second opinion — "what would GPT-5/Gemini say about this?"
2. Delegate a task — "have cursor handle this", "let kiro do it"
3. Pick the best model — "which model is best for this, then run it"
4. Raw run — "run this prompt through opencode"

Don't overthink which framing applies — resolve a CLI/model (see Resolution below) and invoke the dispatcher.

Invocation