The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill requires an external binary installed via a third-party Homebrew repository (steipete/tap/peekaboo). This represents a dependency on code not hosted by the skill author or a trusted vendor list member.\n- [PRIVILEGE_ESCALATION]: The CLI requires high-level system permissions, specifically 'Accessibility' and 'Screen Recording', which grant the agent broad control over the operating system and visibility into other applications.\n- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its vision analysis features.\n
Ingestion points: Screen captures and UI metadata frames processed via the --analyze flag in the image and see commands.\n
Boundary markers: None provided in the documentation to separate UI text from agent instructions.\n
Capability inventory: Significant capabilities including click, type, paste, app launch, window management, and script execution (run).\n
Sanitization: No evidence of text filtering or instruction sanitization for visual content analysis.\n- [DYNAMIC_EXECUTION]: The peekaboo run command allows for the execution of .peekaboo.json scripts, providing a mechanism for running dynamically defined automation sequences.\n- [DATA_EXPOSURE_AND_EXFILTRATION]: The tool can access sensitive system data, including clipboard content via the clipboard command and screen content via capture and image commands.

peekaboo