box-legal-workflows-ma
Pass
Audited by Gen Agent Trust Hub on May 6, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill documentation includes a prerequisite step to install a core 'box' skill from the vendor's official GitHub repository at
github.com/box/box-for-ai. - [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it is designed to process untrusted data (documents uploaded for M&A due diligence) and has the capability to modify system state, such as file permissions.
- Ingestion points: File content analyzed via
ai_qa_multi_file,ai_qa_single_file, andai_extract_structured_from_fields_enhancedin SKILL.md. - Boundary markers: None present. The instructions do not specify delimiters or warnings for the model to ignore instructions found within the processed files.
- Capability inventory:
create_collaboration(grants access),create_folder,copy_file, andset_file_metadatain SKILL.md. - Sanitization: None present. There is no evidence of data validation or sanitization before document content is interpolated into AI prompts.
Audit Metadata