pinecone-assistant
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is developed by the official service provider (pinecone-io) and implements standard administrative and interaction patterns for the Pinecone Assistant service.
- [SAFE]: Authentication is handled correctly by retrieving the PINECONE_API_KEY from environment variables rather than hardcoding credentials.
- [SAFE]: Network operations are restricted to the official Pinecone API infrastructure via the official Python SDK.
- [SAFE]: File system operations in upload.py and sync.py are scoped to documentation-specific extensions (.md, .txt, .pdf, .docx, .json) and include robust exclusion lists for sensitive directories like .git, .venv, and node_modules.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied documentation and messages which constitutes a standard ingestion surface for Large Language Model interactions. 1. Ingestion points: Documentation files read by upload.py/sync.py and user messages in chat.py. 2. Boundary markers: None implemented in the processing scripts. 3. Capability inventory: File reading and Pinecone API transmission. 4. Sanitization: No local filtering of content is performed before transmission to the API. This surface is evaluated as safe as the skill lacks dangerous local capabilities and the interaction is confined to the Assistant API environment.
Audit Metadata