codex-autoresearch-skill
Fail
Audited by Gen Agent Trust Hub on May 16, 2026
Risk Level: HIGHCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill explicitly instructs users to start the agent with the
--dangerously-bypass-approvals-and-sandboxflag. This removes the security boundaries intended to protect the host environment, allowing the agent to execute any shell command, access sensitive files, or perform network operations without human oversight or technical restriction. - [EXTERNAL_DOWNLOADS]: The documentation provides installation instructions that fetch code from an external GitHub repository (
github.com/leo-lilinxiao/codex-autoresearch). This source is not verified as part of the official vendor infrastructure or trusted organization list. - [REMOTE_CODE_EXECUTION]: The primary function of the skill is an autonomous 'modify-verify' loop. It generates code changes and then executes them via inferred shell commands (e.g.,
pytest,npm test,grep). When combined with the sandbox bypass, this creates a mechanism where the AI can autonomously execute arbitrary and potentially malicious code on the user's machine. - [DATA_EXPOSURE]: By encouraging the removal of the sandbox, the skill places all local user data, including environment variables, SSH keys, and cloud credentials, at risk of being read or exfiltrated by the agent during its autonomous operations.
Recommendations
- AI detected serious security threats
Audit Metadata