The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/render.ts script uses child_process.execSync to dynamically run package managers like npm or bun to install the beautiful-mermaid package if it is not found in the environment.
[EXTERNAL_DOWNLOADS]: The skill attempts to download and install the beautiful-mermaid package from the NPM registry during execution if it is missing, as seen in the ensurePackage function in scripts/render.ts.
[REMOTE_CODE_EXECUTION]: The scripts/render.ts script utilizes dynamic imports (await import(name)) to load the beautiful-mermaid package at runtime after potentially installing it via shell commands, which constitutes dynamic loading of remote code.
[DATA_EXFILTRATION]: The skill presents an indirect prompt injection surface. User-supplied Mermaid diagram syntax is rendered into an SVG, which is then embedded without sanitization into an HTML wrapper by scripts/create-html.ts. This file is subsequently opened by the agent-browser tool, creating a risk where malicious diagram content could execute JavaScript (XSS) in the browser context to access sensitive data.
Ingestion points: Mermaid diagram code provided by the user via the --code or --input arguments to scripts/render.ts.
Boundary markers: No delimiters or warnings are used when the SVG content is interpolated into the HTML template in scripts/create-html.ts.
Capability inventory: The skill possesses file-write capabilities (writeFileSync), command execution via execSync, and browser automation via the agent-browser tool mentioned in SKILL.md.
Sanitization: The SVG content is read and injected directly into the HTML <body> in scripts/create-html.ts without any sanitization or verification of its content.

beautiful-mermaid