The Agent Skills Directory

[PROMPT_INJECTION] (LOW): The skill exposes an Indirect Prompt Injection surface (Category 8) through its RAG (Retrieval Augmented Generation) implementation examples. 1. Ingestion points: Document metadata and text retrieved via index.query in the RAG and Vercel Function examples. 2. Boundary markers: Absent; the example code joins metadata directly into a context string for the LLM without using delimiters or protective instructions. 3. Capability inventory: The skill has database write/delete capabilities via index.upsert, index.delete, and index.reset. 4. Sanitization: No sanitization of retrieved database content is demonstrated before it is passed to the LLM.
[EXTERNAL_DOWNLOADS] (SAFE): The skill installs the @upstash/vector package. This is an official SDK from a trusted organization (Upstash), qualifying for a downgraded severity under the TRUST-SCOPE-RULE.
[CREDENTIALS_UNSAFE] (SAFE): API credentials like UPSTASH_VECTOR_REST_TOKEN are managed correctly through environment variables and placeholders. No hardcoded secrets were detected.

upstash-vector-db-skills