peekaboo-hardened

Pass

Audited by Gen Agent Trust Hub on Apr 21, 2026

Risk Level: SAFEDATA_EXFILTRATIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill accesses sensitive system data, including full-screen captures, window lists, and the system clipboard. It also manages application configuration and credentials through the peekaboo config command.- [EXTERNAL_DOWNLOADS]: The skill installs the underlying Peekaboo tool via a third-party Homebrew tap at steipete/tap/peekaboo.- [COMMAND_EXECUTION]: Subprocess calls are used to drive the peekaboo CLI for various UI interactions (clicking, typing, scrolling) and system management (launching/quitting apps).- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its ability to ingest untrusted data from the screen and clipboard while possessing high-privilege execution capabilities. Evidence:
  • Ingestion points: see, capture, and clipboard commands in SKILL.md
  • Boundary markers: Absent for screen-captured text or UI element descriptions
  • Capability inventory: Full UI automation (clicks, typing, app control) and local script execution via peekaboo run
  • Sanitization: Absent.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 21, 2026, 06:31 PM
Security Audit — agent-trust-hub — peekaboo-hardened