The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from design-system.md, brand-philosophy.md, and user-provided inputs to generate HTML files. This data flow creates a surface for indirect prompt injection where malicious instructions inside those files could influence the agent's output.
Ingestion points: design-system.md, brand-philosophy.md, and user content descriptions.
Capability inventory: The skill uses the Write tool to create files and Bash to execute commands.
Boundary markers: Metadata comments such as  are used for framework mapping but do not serve as safety boundaries for content validation.
Sanitization: No specific instructions are provided to sanitize or escape input data during the generation process.
[DYNAMIC_EXECUTION]: The skill directs the agent to execute a local Node.js script using a path determined by an environment variable ($BRAND_CONTENT_DESIGN_DIR/scripts/html-icons.js). Executing scripts from computed paths is a medium-risk practice that depends on the integrity of the environment.
[COMMAND_EXECUTION]: The skill instructions include the use of the Bash tool to run node-based commands for searching and retrieving SVG icons for integration into the HTML output.

html-generator