What problem does it solve? Teams that want local-first AI execution often assume a listed endpoint or model name guarantees capability, leading to failed tasks, privacy leaks, or fabricated results. This Skill provides a disciplined assessment protocol so an agent only relies on a local or remote model endpoint after verifying it against real evidence. ## Core Features & Use Cases - Evidence-based capability verification: Test reasoning, coding, extraction, or critique roles against the real endpoint with a small representative task before relying on it. - Honest scope boundaries: Clarifies that models/capabilities.yaml is a discovery reference, not a live router, and that Agentit does not implement automatic model routing. - Privacy and authorization guardrails: Prevents sending private code to new providers, spending money, or downgrading high-risk critics without explicit authorization. - Use Case: A user asks the agent to run code review on a local OpenAI-compatible endpoint. The agent verifies tools, context capacity, transport, and privacy boundaries on the actual endpoint, records observed capabilities and failures, and only then assigns the bounded role. ## Quick Start Assess whether my available local model endpoint can handle bounded code extraction tasks and record the verified capabilities.
