visio-automation

Pass

Audited by Gen Agent Trust Hub on Jun 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes PowerShell scripts to automate Microsoft Visio via COM objects (Visio.Application and Visio.InvisibleApp). It includes a C# helper compiled at runtime using the Add-Type command to interface with the Windows Running Object Table (ROT), allowing the agent to interact with and verify existing Visio document instances.
  • [PROMPT_INJECTION]: The skill processes various external data formats, including Mermaid diagrams, Draw.io XML, and natural language descriptions, to construct diagram models. This creates an indirect prompt injection surface where instructions could potentially be embedded in the processed data.
  • Ingestion points: The functions Convert-MermaidToVisio, Convert-DrawIOToVisio, and Convert-NaturalLanguageFlowToVisio in scripts/visio_helpers.ps1 process external strings and files.
  • Boundary markers: The skill instructions emphasize normalizing input into an intermediate Graph Model structure, which acts as a logical boundary, though no explicit string delimiters are enforced in the instructions.
  • Capability inventory: The skill possesses local file system read/write access and the ability to execute PowerShell commands and manipulate the Visio COM API.
  • Sanitization: Inputs are parsed using specific regular expression patterns and XML parsing logic before being rendered into the final diagram, reducing the risk of accidental instruction execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 22, 2026, 06:03 PM