The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes untrusted data from ingestion points such as external URLs, brand documents, and application archives. This presents a potential surface for indirect prompt injection. The skill implements boundary markers by wrapping content in protective tags and providing instructions to ignore embedded commands. The capability inventory includes file writing, network requests, and system command execution. Sanitization is present in the form of pre-filtering for common injection patterns and size validation.
External Data Retrieval: The skill utilizes network capabilities to fetch design specifications and brand assets from remote URLs. To address potential risks like Server-Side Request Forgery (SSRF), it implements strict outbound network policies, including the blocking of private, loopback, and link-local IP ranges, along with DNS rebinding defenses and redirect limits.
System Command Execution: The skill uses shell commands for project initialization, file management, and extracting theme configurations from codebases. These capabilities are supported by path safety hooks that restrict operations to the project directory and block access to sensitive system locations like SSH keys or environment configuration files.
Credential Management Considerations: The skill supports integration with external services which require authentication tokens. It incorporates logic to manage these secrets via environment variables rather than hardcoding them, and includes mechanisms to mask these tokens in logs and prevent their accidental persistence in project files.

design-system