The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes bash commands to manage configuration files and telemetry logs within the user's home directory (~/.superstack/). It also utilizes platform-specific commands such as open, xdg-open, and start to display generated HTML previews in the browser.- [DATA_EXFILTRATION]: Usage statistics are transmitted to a remote endpoint via curl. This telemetry activity is restricted to metadata (skill name, platform, duration) and is only performed after an explicit interactive consent prompt has been accepted by the user.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its data ingestion points. 1. Ingestion points: idea-context.md, build-context.md, and package.json. 2. Boundary markers: Absent. 3. Capability inventory: Shell execution (telemetry and browser launch), file system modification (CSS and Next.js layout files). 4. Sanitization: Absent. The potential impact is limited to the manipulation of design output tokens.

brand-design