What problem does it solve? When a user explicitly names a skill that is absent from the currently visible skill metadata, or when no visible skill semantically covers a specialized request, the host AI has no safe way to find and load that capability. This Skill provides a bounded, read-only fallback that retrieves candidate skills from a local catalog and deterministically validates the exact dependency closure before anything is loaded. ## Core Features & Use Cases - Cold Discovery: Retrieves candidate skills from a local capability catalog using one or two functional domain hints (e.g., decision, engineering, coding, frontend), while keeping semantic selection with the host AI. - Deterministic Validation: Verifies catalog schema and fingerprint, entrypoint and package SHA-256 hashes, path containment, and the full dependency closure before authorizing a read of the validated files. - Execution Admission: Generates admission contracts and handoff files for native subagent execution (one-shot, parent-mediated, or multi-turn interview), with revalidation checks immediately before any controlled write. - Use Case: A user asks for an evidence-based design interview, but no visible skill covers interactive questioning. The host runs the router with the 'decision' domain hint, selects one candidate, validates its closure, and hands it to the appropriate native agent with a strict admission contract. ## Quick Start Ask the assistant to find and validate the named local skill that is missing from the visible skill list, passing the complete request and one functional domain hint.
