adobe-firefly-image
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of user-provided prompt text which is then used to construct requests to external Adobe Firefly API endpoints. This introduces a potential surface for indirect prompt injection.
- Ingestion points: Prompt and negativePrompt input parameters defined in the generation and edit requests within SKILL.md.
- Boundary markers: The instructions explicitly require following Adobe's Generative AI User Guidelines and implementing human review for high-impact or sensitive imagery.
- Capability inventory: The included Python scripts possess capabilities for network communication via urllib.request and file system writes via pathlib.
- Sanitization: The skill provides structured art-direction guidance and technical exclusions to constrain model behavior.
- [SAFE]: The skill implements secure credential management by using environment variables for client secrets and access tokens. All network operations are directed at official, well-known Adobe domains, and the instructions provide extensive guidance on privacy, data retention, and content provenance.
Audit Metadata