reflex-browser

Pass

Audited by Gen Agent Trust Hub on Apr 7, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Configures the environment to fetch the @reflexautomation/reflex-cli package from a vendor-hosted registry at git.bqa-solutions.nl.
  • [REMOTE_CODE_EXECUTION]: Provides functionality to download the Reflex agent backend and its runtime using reflex agent download and reflex agent runtime install.
  • [COMMAND_EXECUTION]: Executes shell-based commands for session control and supports arbitrary Lua execution through reflex lua exec for advanced tasks.
  • [PROMPT_INJECTION]: Ingests untrusted web data through summary, text, and html commands (SKILL.md). Boundary markers are provided by a structured JSON response envelope (references/protocol.md). The skill allows high-impact actions like reflex lua exec and reflex browser fill (SKILL.md), but lacks explicit documentation for sanitizing external inputs (sanitization absent).
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 7, 2026, 08:41 PM