The Agent Skills Directory

[SAFE]: The skill utilizes placeholders such as your-api-key and your-key in SKILL.md and references/deployment.md, which is a safe practice for demonstrating API interaction without exposing real credentials.
[EXTERNAL_DOWNLOADS]: The skill references and installs several well-known and official libraries including pinecone-client, langchain-pinecone, langchain-openai, and llama-index. These references point to established services and follow standard integration patterns for vector databases.
[INDIRECT_PROMPT_INJECTION]: The skill documentation describes data ingestion and retrieval patterns that represent an indirect prompt injection surface:
Ingestion points: Data is ingested via index.upsert calls in SKILL.md and references/deployment.md, where arbitrary vector metadata is stored.
Boundary markers: The examples do not include explicit delimiters or instructions to ignore potential commands embedded in metadata.
Capability inventory: The skill uses the official Pinecone client to perform network operations for indexing and querying data.
Sanitization: There is no demonstration of data sanitization for the values stored in or retrieved from the database.
Assessment: These surfaces are inherent to the intended use-case of a vector database for RAG (Retrieval-Augmented Generation) and do not represent a malicious implementation in this context.

pinecone