The Agent Skills Directory

[SAFE]: The skill incorporates a formal "Exact write-plan protocol" (referenced in references/exact-write-protocol.md) that enforces strict safety boundaries. It requires explicit user intent, unique target identification, and verification of results before and after every modification, effectively preventing autonomous or accidental data corruption.\n- [SAFE]: Instructions distinguish clearly between "read-only" status requests and "exact-write" commands. The agent is explicitly prohibited from inferring business values (such as prices or quantities) or making autonomous business decisions, ensuring that all actions are driven by direct user input.\n- [SAFE]: The skill utilizes a modular architecture, interacting with a dedicated Yandex KIT MCP server and delegating complex catalog audits to a specialized doctor skill. This minimizes the risk of scope creep and maintains clear operational boundaries.\n- [SAFE]: Communication guidelines and reporting formats are well-defined, focusing on factual data presentation and explicit coverage notes, which helps prevent user confusion regarding the state of the store.

a1-yandex-kit-operator