The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses osascript to programmatically control the macOS Finder and hide application windows (command down, option down + h). It also executes the screencapture utility to take screenshots of the user's desktop.
[DATA_EXFILTRATION]: The instructions mandate capturing a desktop screenshot (screencapture -x /tmp/widget-desktop-proof.png) as proof of completion. This process can inadvertently capture and expose sensitive information visible on the user's screen, such as open documents, browser tabs, or system notifications, which may then be processed or shared by the agent.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface. It ingests untrusted data from local repository files (Swift code, project configurations) through file system searches and a Python analysis script.
Ingestion points: Files are read in SKILL.md (via rg) and scripts/check_widget_design.py (via Path.read_text).
Boundary markers: None identified; the agent is not instructed to ignore embedded instructions in the codebase.
Capability inventory: The agent can execute shell commands, take screenshots, and modify files.
Sanitization: There is no evidence of sanitization or validation of the ingested code content before it influences the agent's decision-making process.

mac-widget-design