lynx-simulator

Warn

Audited by Gen Agent Trust Hub on Nov 20, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill uses npx --yes agent-lynx@latest to download and execute the latest version of the core CLI tool from the npm registry.
  • [REMOTE_CODE_EXECUTION]: The agent-lynx simulator run command accepts --jsb and --native-extension flags which are used to download, install, and execute arbitrary npm packages from the public registry based on task requirements.
  • [DYNAMIC_EXECUTION]: The native extension loader in examples/extensions/native-package/index.cjs uses require to dynamically load compiled .node binary files from computed paths based on the host platform and architecture.
  • [DYNAMIC_EXECUTION]: The skill executes dynamic JavaScript code through page-env.mjs factories and JSB handler packages to mock bridge calls and manage page state at runtime.
  • [COMMAND_EXECUTION]: The skill frequently invokes shell-based commands including agent-lynx, npx, and npm pack to manage the simulator lifecycle and package development.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data sources including compiled Lynx bundles (template.js), page-env configurations, and recorder artifacts. This presents an attack surface where malicious instructions could be embedded in the processed data.
  • Ingestion points: SKILL.md (bundle URLs/files), references/recorder-fixture.md (recorder artifacts).
  • Boundary markers: Absent; the skill does not specify delimiters or instructions to ignore embedded content in processed bundles.
  • Capability inventory: shell command execution (agent-lynx), dynamic loading of native binaries (require), and network requests via the host fetch API.
  • Sanitization: Not implemented for the ingested bundle or fixture content.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 20, 2026, 03:18 PM