cloakbrowser-fallback

Warn

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill relies on the cloakbrowser package, which is instructed to download a 535 MB custom Chromium binary from a remote source when initialized via python -m cloakbrowser install.\n- [REMOTE_CODE_EXECUTION]: The skill facilitates the execution of a custom-compiled Chromium binary with source-level modifications ("C++ source-level fingerprint patches"). This binary is downloaded from an unverified source and used to navigate websites, presenting a risk if the binary contains malicious patches.\n- [INDIRECT_PROMPT_INJECTION]:\n
  • Ingestion points: The skill fetches and processes content from arbitrary URLs provided by the user or agent via scripts/cloak_fetch.py and scripts/cloak_fetch.mjs.\n
  • Boundary markers: None present in the prompt instructions to prevent the agent from obeying instructions found on the fetched pages.\n
  • Capability inventory: The agent can execute shell commands, write files (screenshots and HTML dumps), and perform further network requests.\n
  • Sanitization: No visible sanitization of the extracted page text or HTML before it is returned to the agent context.\n- [COMMAND_EXECUTION]: The skill uses shell commands to manage the browser environment and fetch data, including python -m cloakbrowser and direct execution of local helper scripts.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 28, 2026, 05:22 PM