chroma

Pass

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches metadata from the official PyPI registry (pypi.org) to check for package updates in scripts/check-updates.py. The skill also documents the automatic download of pre-trained models from Hugging Face for local embedding generation as described in references/06-embedding-functions.md.
  • [COMMAND_EXECUTION]: Provides standard CLI commands for package installation and server management (chroma run) across multiple environments.
  • [INDIRECT_PROMPT_INJECTION]: \n
  • Ingestion points: The skill facilitates indexing external data from files, images, and remote sources like GitHub or S3 as described in references/08-multimodal.md and references/10-chroma-cloud.md. \n
  • Boundary markers: Not explicitly defined in the provided implementation snippets, which is expected for database documentation. \n
  • Capability inventory: Includes file system interaction via data loaders and network communication for cloud/API operations. \n
  • Sanitization: Relies on default agent behavior; no explicit sanitization logic is present in the provided code samples.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 4, 2026, 10:30 AM
Security Audit — agent-trust-hub — chroma