agentic-mermaid-diagram-workflow
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted Mermaid source code containing labels or metadata that could influence the agent when it reads generated diagram summaries or semantic facts. Ingestion points: Mermaid source code provided to verify, describe, and mutate tools (SKILL.md). Boundary markers: Absent; source code is treated as literal input. Capability inventory: JavaScript execution via the execute tool, diagram mutation, and rendering. Sanitization: Absent; the system prioritizes lossless round-trips over filtering.
- [DYNAMIC_EXECUTION]: The hosted MCP and Code Mode channels include an execute tool (SKILL.md, references/hosted-mcp.md) that runs JavaScript code in a node:vm sandbox. While this is a core functionality for the diagram workflow SDK, it provides a surface for dynamic code execution.
- [EXTERNAL_DOWNLOADS]: The skill requires communication with a hosted MCP service at https://agentic-mermaid.dev/mcp. This domain is specific to the project and is not recognized as an established or whitelisted technology service.
Audit Metadata