neurodivergent-visual-org
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to determine cognitive modes and populate Mermaid diagrams, creating a surface for potential indirect injection attacks.
- Ingestion points: User messages are ingested in the 'Mode Detection Algorithm' (
references/mode-detection-algorithm.md) and the 'Mermaid Diagram Selection Guide' (SKILL.md) to generate content. - Boundary markers: The instructions do not define specific delimiters or 'ignore embedded instructions' warnings for the data being placed inside Mermaid nodes.
- Capability inventory: The skill has access to
Bash(cmd:*), which allows it to read and write files on the host system. - Sanitization: While the skill mandates URL-encoding for playground links using Python's
urllib.parse.quote(), it lacks explicit instructions for sanitizing or escaping user-provided text before it is interpolated into the Mermaid diagram syntax itself, which could allow a user to break the diagram structure or influence the agent's behavior during generation.
Audit Metadata