skills/qverisai/qverisbot/peekaboo/Gen Agent Trust Hub

peekaboo

Pass

Audited by Gen Agent Trust Hub on Mar 6, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill installs the peekaboo CLI binary using a third-party Homebrew tap (steipete/tap/peekaboo) during the installation phase.
  • [COMMAND_EXECUTION]: The skill provides extensive capabilities to control the host system, including launching/quitting applications (peekaboo app), interacting with system dialogs, and simulating mouse/keyboard input (click, type, drag, hotkey). It also includes a run command to execute .peekaboo.json scripts.
  • [DATA_EXFILTRATION]: The tool has the capability to access sensitive system information, including full-screen screenshots (peekaboo image), application window hierarchies and annotations (peekaboo see), and reading/writing the system clipboard (peekaboo clipboard).
  • [PROMPT_INJECTION]: The skill represents an indirect prompt injection surface as it ingests and processes untrusted data from the user interface.
  • Ingestion points: Screen captures, UI element maps (OCR/AI analysis), and window titles are ingested through commands like peekaboo see and peekaboo image.
  • Boundary markers: There are no explicit boundary markers or instructions to ignore embedded content in the captured UI data.
  • Capability inventory: The agent can use the results of these captures to perform high-privilege actions like click, type (including password entry), and app launch.
  • Sanitization: No sanitization or filtering of the captured screen content is documented before the data is returned to the agent context.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 6, 2026, 01:22 AM