The Agent Skills Directory

[SAFE]: The skill implements strong local security for its visual companion server. It uses unique session tokens for authentication, restricting access to the local user. It also correctly prevents directory traversal attacks by validating file paths against the intended content directory.
[COMMAND_EXECUTION]: The server script includes functionality to automatically open the user's browser using platform-specific commands (like open, xdg-open, or rundll32.exe). These calls are gated by environment variables and user opt-in, representing standard behavior for local developer tools.
[EXTERNAL_DOWNLOADS]: The browser companion references a branding logo from a remote site (https://primeradiant.com). This is a resource from the author's infrastructure and is used only for UI branding within the companion tool.
[DATA_EXFILTRATION]: The skill records user interactions (selections and clicks) from the visual companion into a local events file. This data is consumed only by the agent to inform the brainstorming process and is stored within the local project's session directory.
[SAFE]: The skill ingests user interaction data from the browser companion via a local JSON-formatted file. This design manages the risk of indirect prompt injection by using structured data originating from the user's own local environment.

brainstorming