lightpanda-browser

Fail

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONPERSISTENCE
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill instructs the user to install components by piping a remote script directly to a shell, a pattern that allows execution of untrusted code without prior review.
  • Evidence: curl -fsSL https://github.com/nichochar/install-lightpanda/raw/main/install.sh | bash found in SKILL.md and references/lightpanda-setup.md.
  • Risk: The script is hosted in a repository (nichochar/install-lightpanda) that is not identified as a trusted or verified source.
  • [EXTERNAL_DOWNLOADS]: The skill facilitates the download and compilation of software from external, untrusted sources.
  • Evidence: git clone https://github.com/lightpanda-io/browser.git followed by zig build in SKILL.md.
  • Evidence: docker pull lightpanda/browser:latest in references/lightpanda-setup.md.
  • [DYNAMIC_EXECUTION]: The skill documentation describes and enables the execution of arbitrary JavaScript within the browser context, which could be leveraged to perform unauthorized actions on websites.
  • Evidence: Documentation for agent-browser --cdp 9222 eval in SKILL.md and references/commands.md.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a broad attack surface for indirect prompt injection due to its ability to ingest and interact with untrusted web content using powerful capabilities.
  • Ingestion points: agent-browser open and snapshot commands (found in SKILL.md) that pull external DOM data into the agent's context.
  • Capability inventory: eval (JavaScript execution), fill (input), click (interaction), and network route (traffic manipulation) documented in SKILL.md.
  • Boundary markers: None identified in the prompt instructions to isolate external data.
  • Sanitization: No evidence of input validation or output escaping in the provided shell templates.
  • [PERSISTENCE]: The setup documentation includes templates for creating persistent background services that allow the browser daemon to survive system reboots.
  • Evidence: Configuration files for macOS launchd (io.lightpanda.browser.plist) and Linux systemd (lightpanda.service) in references/lightpanda-setup.md.
Recommendations
  • HIGH: Downloads and executes remote code from: https://github.com/nichochar/install-lightpanda/raw/main/install.sh - DO NOT USE without thorough review
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Sep 17, 2026, 09:09 AM