The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill recommends executing npx getdesign@latest add <brand> in references/design-reference.md to fetch design presets. This command downloads and runs code from a third-party package registry (NPM) at runtime. The use of the @latest tag makes it susceptible to supply chain attacks, as it fetches the most recent, unverified version of the package. While the skill specifies that this command should only be executed with explicit user approval, the underlying pattern of executing unversioned code from a non-predefined trusted source is a security concern.
[EXTERNAL_DOWNLOADS]: The npx command involves downloading external software from the NPM registry. The specific package referenced (getdesign) is not provided by a designated trusted organization or a widely recognized well-known service, representing an external dependency risk.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands, notably the npx utility for design presets and potentially other developer tools to verify terminal output or project states (as mentioned in references/mode-quick-fix.md). This increases the agent's interaction with the host system's shell.
[INDIRECT_PROMPT_INJECTION]: The skill is highly focused on ingesting external data to perform its UI polish tasks, creating a surface for indirect prompt injection.
Ingestion points: The skill reads source code from user-provided repository URLs and pastes (SKILL.md) and processes visual information from screenshots of existing UI (references/mode-screenshot-iteration.md).
Boundary markers: There are no explicit instructions or delimiters provided to ensure the agent ignores potentially malicious instructions embedded in the external source code or visual data it is analyzing.
Capability inventory: The agent has the capability to execute shell commands (npx), read file contents, and utilize vision tools for screenshot analysis.
Sanitization: The skill does not outline any procedures for sanitizing or filtering instructions from external inputs before they are incorporated into the agent's context.

ui