The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/render.ts utilizes child_process.execSync to execute shell commands for package management, specifically npm install or bun add, to fetch dependencies at runtime.
[REMOTE_CODE_EXECUTION]: The skill dynamically installs and loads external code by pulling the beautiful-mermaid package from a public registry during execution if it is not already present.
[PROMPT_INJECTION]: The skill exhibits a vulnerability to indirect prompt injection in scripts/create-html.ts. The script embeds raw, unsanitized SVG content (derived from user-provided Mermaid code) directly into an HTML template. When this HTML is subsequently processed by the agent-browser skill, any malicious scripts or event handlers within the SVG could be executed in the agent's browser context.
Ingestion points: User input provided via the --code or --input flags in scripts/render.ts, which is then passed to scripts/create-html.ts via the --svg flag.
Boundary markers: The skill does not implement boundary markers or instructions to ignore embedded commands within the processed data.
Capability inventory: The skill has the capability to execute shell commands (execSync), read/write to the file system, and interact with a web browser through a secondary skill.
Sanitization: No sanitization or validation of the SVG content is performed before it is interpolated into the HTML wrapper.

beautiful-mermaid