The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to determine the logic and content of generated diagrams. Ingestion point: User scenario description (Stage 1). Boundary markers: Absent. Capability inventory: Local file reading of 'references/' and text output; no network access or subprocess execution is available. Sanitization: Absent. The risk is assessed as low because the skill is explicitly instructed to follow local documentation for syntax and uses an interactive questioning phase that naturally mitigates the immediate execution of complex prompt injection payloads.
[EXTERNAL_DOWNLOADS]: The skill references 'https://mermaid.ai/live' as a destination for users to render their generated code. This is a well-known and established community service for the Mermaid ecosystem. The reference is informational for the end-user and does not involve automated data exfiltration or risky remote code execution by the agent.

diagrama-mermaid