box-legal-workflows-intake

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references and downloads prerequisites from the official Box GitHub repository (github.com/box/box-for-ai). These are identified as trusted resources provided by the vendor for core functionality.
  • [COMMAND_EXECUTION]: Instructions provided in the prerequisites section utilize npx skills add to install required capabilities into the agent environment.
  • [INDIRECT_PROMPT_INJECTION]: The skill has an indirect injection surface as it processes external client intake documents to check for completeness and extract data.
  • Ingestion points: Document analysis is performed on files within the intake folder using ai_qa_multi_file and ai_extract_structured_from_fields_enhanced tools.
  • Boundary markers: The instructions explicitly direct the AI to focus on "facts only" and to cite sources, which helps delineate between document content and agent instructions.
  • Capability inventory: The skill has the ability to write metadata (set_file_metadata), add file comments, manage collaborations/permissions (create_collaboration), and generate documents (create_docgen_batch).
  • Sanitization: The workflow includes mandatory human confirmation steps ([CONFIRM: ...]) for all critical actions, including folder IDs, requirement checklists, risk ratings, assigned personnel, and document sharing methods, effectively mitigating the risk of automated misbehavior driven by malicious document content.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 06:13 PM
Security Audit — agent-trust-hub — box-legal-workflows-intake