The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands to execute its internal Python logic. Instructions in references/diagram-generation.md specify the use of uv run to invoke scripts/generate_excalidraw.py and scripts/validate_excalidraw.py for processing diagram specifications and validating output files.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it processes untrusted user descriptions to generate diagram files and determines output filenames based on context.
Ingestion points: User prompts describing diagram requirements are ingested in SKILL.md and references/guided-design.md to build a JSON specification.
Boundary markers: The instructions do not define explicit delimiters or 'ignore' instructions for the user-provided content passed into the diagram builder.
Capability inventory: The skill uses scripts/generate_excalidraw.py to write files to the disk and executes these scripts via shell commands.
Sanitization: The skill employs a structured JSON intermediate format for diagram data, which provides a layer of structural validation before the generation script processes the input.

bmad-excalidraw