The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx to download and run @probelabs/maid@0.0.29 and @mermaid-js/mermaid-cli from the standard npm registry. These are pinned versions used specifically for diagram validation and rendering. The skill instructions include a check for network access permissions before proceeding with these downloads.
[COMMAND_EXECUTION]: The workflow executes local Python scripts (mermaid_lint.py, render.py) located in a sibling directory (../diagram-review/). These scripts are part of the vendor's coordinated set of diagramming tools.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied descriptions to generate diagram code. It mitigates potential injection risks by enforcing structural constraints, such as quoting all labels and capping the number of nodes, while providing specific instructions to avoid inventing entities.
[DATA_EXFILTRATION]: The skill explicitly documents the privacy risks of third-party diagram editors like kroki.io, warning users that standard access controls may not apply, and provides a 'fail closed' mechanism if validation scripts cannot be run.

mermaid-draw