What problem does it solve? AI agent skills often silently pick how they execute — inline in the active agent, delegated to a sub-agent, or fanned out in parallel — without the caller knowing a choice existed. The Caller Options protocol (COP) makes these hidden routing decisions explicit so a human or agent caller decides instead of the skill guessing. ## Core Features & Use Cases - Five-step routing protocol: Triage, Enumerate, Assess, Surface, and Execute — with an anti-nag gate that skips prompting when one option clearly dominates. - Venue and Mode dimensions: Evaluates inline vs. delegated vs. fan-out execution venues plus any named modes a skill declares, using documented venue heuristics. - Depth-1 safety guard: A COP-RESOLVED marker block prevents nested delegation, and self-orchestrating skills are forced inline to avoid depth-2 sub-agent nesting. - Use Case: When running an investigation skill that performs heavy throwaway reading, COP detects the delegation option, recommends a primed sub-agent to protect the main context, and asks the caller to confirm before executing. ## Quick Start Invoke /caller-options followed by a task or skill name to get an ad-hoc routing recommendation with venue and mode options.
