amazon-nova-canvas
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied prompts and images through the Amazon Bedrock API. While the provided Python script includes validation for prompt length and image properties, the underlying AI model remains susceptible to prompt injection attacks embedded in user input.
- Ingestion points: The main function in the Python script accepts prompt and negative text as command-line arguments.
- Boundary markers: The script enforces a character limit (1024) on prompts and validates image dimensions and byte sizes before processing.
- Capability inventory: The skill can execute network requests to AWS Bedrock (boto3.client.invoke_model) and write manifest and image files to the local file system (atomic_write) within SKILL.md.
- Sanitization: Input prompts are used in the JSON payload without further escaping, though generated images are validated for format and dimensions after retrieval to ensure they match expected outputs.
Audit Metadata