fal-models-catalog

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests untrusted content from external sources to perform media generation and analysis tasks.\n
  • Ingestion points: Modality reference files, such as references/audio-to-text.md and references/image-to-text.md, define parameters like audio_url, image_url, and prompt that ingest content from external locations.\n
  • Capability inventory: The skill leverages the genmedia CLI tool for model operations (genmedia run, genmedia status), which includes a local file-writing capability via the --download flag.\n
  • Boundary markers: There are no explicit instructions or delimiters within the reference templates to mitigate potential instructions embedded in media content (e.g., hidden text in images).\n
  • Sanitization: The instructions do not define any sanitization or validation routines for the external data ingested by the fal.ai endpoints.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 04:39 PM
Security Audit — agent-trust-hub — fal-models-catalog