camofox-browser

Warn

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The script scripts/setup.sh installs the @askjo/camofox-browser package from the NPM registry and executes it using node server.js. This package is not from a recognized trusted vendor.- [EXTERNAL_DOWNLOADS]: The installation process involves downloading third-party code and a Firefox-based browser binary (~300MB) from external servers during the setup phase.- [COMMAND_EXECUTION]: The skill executes local shell scripts and leverages system binaries like curl, python3, and node to manage the browser server and perform interactions.- [PERSISTENCE]: The skill starts a Node.js server in the background using nohup, which remains active across agent sessions to maintain browser state and sessions.- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes content from external websites to create accessibility snapshots for the agent. This represents an attack surface where a malicious website could embed instructions targeting the agent.
  • Ingestion points: The snapshot command in scripts/camofox.sh fetches and displays page content.
  • Boundary markers: No explicit delimiters or warnings to ignore embedded instructions are present in the scripts.
  • Capability inventory: The skill has the ability to perform network requests (curl), write files (screenshots in /tmp), and interact with web elements.
  • Sanitization: The skill does not appear to sanitize or filter the text content retrieved from web pages before presenting it to the agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 17, 2026, 06:44 PM