agent-browser

Pass

Audited by Gen Agent Trust Hub on Jul 20, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill directs the user to install the agent-browser package globally via npm (npm i -g agent-browser).
  • [REMOTE_CODE_EXECUTION]: The skill uses agent-browser skills get core to fetch its primary logic and instruction set from a remote source at runtime. This allows the tool to load executable workflows that are not present in the static skill file.
  • [COMMAND_EXECUTION]: The skill requires the Bash tool to execute agent-browser commands, including installation, skill retrieval, and browser automation tasks.
  • [PROMPT_INJECTION]: By fetching instructions dynamically from a remote CLI (agent-browser skills get), the skill introduces an indirect injection surface where the agent's core behavioral logic is provided by an external source at runtime.
  • Ingestion points: agent-browser skills get commands (SKILL.md).
  • Boundary markers: None identified in the discovery stub.
  • Capability inventory: Full shell access via Bash tool, browser automation (CDP), and access to local Electron application contexts.
  • Sanitization: Not specified in the current stub file.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 20, 2026, 11:05 AM
Security Audit — agent-trust-hub — agent-browser