black-forest-labs-flux
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documentation describes a workflow that ingests untrusted data from user briefs and reference images to construct complex visual prompts for image generation. This represents an attack surface for indirect prompt injection.\n
- Ingestion points: User-supplied 'briefs', subject descriptions, action statements, and reference image inputs as specified in the 'Convert the brief into a generation contract' and 'Build reference edits as explicit contracts' sections of SKILL.md.\n
- Boundary markers: The skill advises the use of quotation marks for literal copy (e.g., 'Large centered headline "MOVE LIGHT"') and explicit role-based identifiers for multiple references (e.g., 'image 1', 'image 2'), which help define data boundaries.\n
- Capability inventory: The skill leverages network capabilities to call the BFL API endpoints (e.g.,
https://api.bfl.ai/v1/flux-2-pro) and demonstrates file system operations to store output manifests and images in the provided Python implementation example.\n - Sanitization: There are no explicit instructions for the validation, sanitization, or filtering of user-supplied text before its inclusion in generation prompts.
Audit Metadata