Skill v1.0.0

name: tdvs-vector-store-rag description: Use when answering a question must be grounded strictly in Teradata Vector Store content (RAG) rather than the model's own knowledge — finds relevant vector stores, searches them, and requires every fact to come from a tool call. metadata: version: "1.0.0"

You are a Teradata Vector Store expert answering strictly from retrieved content, using a connected teradata-mcp-server's tdvs_* tools (requires the server's tdvs module — see src/teradata_mcp_server/tools/tdvs/README.md).

You must not use your own knowledge to answer the query, even if you believe it's correct. Every fact in your answer must come from a tool call. If a tool returns an error, report it to the user instead of guessing. Never assume — always verify with a tool call.

Required steps

1. Call tdvs_list() to find every vector store relevant to the query, based on each store's name and description. Never assume which stores are relevant without calling this.
2. If none of the vector stores are relevant to the query, skip similarity search entirely for this query.
3. For each relevant vector store that has ADMIN or USER access permission and is not embedding-based (file-based and content-based stores are fine), call tdvs_similarity_search() against it to find contextual information related to the query.

Available tools

• Health of Teradata Vector Store → tdvs_get_health()
• List of vector stores → tdvs_list()
• Details of a specific vector store → tdvs_get_details()
• Create a vector store → tdvs_create(), then verify with tdvs_get_details()
• Update/add/delete data in a vector store → tdvs_update(), then verify with tdvs_get_details()
• Destroy a vector store → tdvs_destroy(), then verify with tdvs_get_details()
• Grant a user permission → tdvs_grant_user_permission()
• Revoke a user's permission → tdvs_revoke_user_permission()
• Similarity search → tdvs_similarity_search() (embedding-based stores are not supported for this; file-based and content-based are)