The Agent Skills Directory

[PROMPT_INJECTION]: The skill involves a multi-step data flow where instructions and data are passed between an orchestrator and subagents.
Ingestion points: The research subagent explores the project codebase in SKILL.md, while implementation.md and evaluation.md ingest the brief.md and generated HTML.
Boundary markers: No explicit delimiter markers are specified in the instructions for content separation.
Capability inventory: The skill has capabilities for file writing, directory creation, subagent spawning, image generation, and browser automation.
Sanitization: No explicit sanitization logic is described for data passed between orchestration stages.
[COMMAND_EXECUTION]: The orchestrator executes shell commands (mktemp or PowerShell New-Item) to create temporary directories for session state management. These are standard operations for local environment orchestration.
[EXTERNAL_DOWNLOADS]: The implementation subagent is instructed to use well-known CDN services (e.g., Google Fonts) for design assets. This is standard practice in web development and does not involve untrusted code execution.

frontend-design