pinecone-sync

Pass

Audited by Gen Agent Trust Hub on Apr 29, 2026

Risk Level: SAFEDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill facilitates indirect prompt injection by design. It reads and embeds untrusted data from the workspace (e.g., contents of inbox/) into a vector database (Pinecone) intended for future retrieval by other agent skills. If the ingested files contain malicious instructions, they could influence the agent's behavior when the content is recalled later.
  • Ingestion points: Files located in crm/, domain-notes/, research-briefs/, plans/, meetings/, and inbox/ (SKILL.md).
  • Boundary markers: None specified. Chunks are created by splitting on Markdown headers without adding delimiters to signal untrusted content.
  • Capability inventory: The skill uses Read and Bash to access the file system and Pinecone MCP tools (upsert-records) to store the data.
  • Sanitization: No sanitization or validation of the file content is performed before embedding.
  • [DATA_EXFILTRATION]: The skill is designed to move data from the local environment to an external service (Pinecone). It explicitly targets folders that are likely to contain sensitive information, such as crm/*.md and meetings/*.md (SKILL.md). While this is the stated purpose of a synchronization skill, it represents a data exposure surface if the target Pinecone index is not appropriately secured.
  • [EXTERNAL_DOWNLOADS]: The skill identifies a dependency on an external plugin (private-misc) and requires the usage of external Pinecone MCP tools for operation (SKILL.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 29, 2026, 05:16 AM
Security Audit — agent-trust-hub — pinecone-sync