The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx to download and execute the @mermaid-js/mermaid-cli package from the NPM registry during the syntax verification step. This is a well-known package maintained by the official Mermaid.js community.
[COMMAND_EXECUTION]: The skill executes shell commands using filenames derived from user-supplied arguments. While instructions specify a kebab-case naming convention, the lack of programmatic sanitization of the <diagram-name> variable before it is used in bash commands like mmdc or npx creates a potential command injection surface.
[INDIRECT_PROMPT_INJECTION]: The skill includes a mandatory step for the AI to read and review a rendered PNG image. Because the content of this image is generated based on untrusted user requirements, an attacker could potentially embed visual instructions designed to manipulate the AI's review process or subsequent behavior. -- Ingestion points: The agent reviews the rendered image file (figures/diagram-name.png) in Step 5. -- Boundary markers: No specific delimiters or instructions are provided to the agent to disregard potential instructions embedded within the visual diagram content. -- Capability inventory: The skill uses bash for system commands, and write or edit for file manipulation. -- Sanitization: There is no evidence of filtering or sanitizing the input data before it is rendered into the diagram format.

mermaid-diagram