agent-reach
Fail
Audited by Gen Agent Trust Hub on Mar 16, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill instructs the user to execute
pip install https://github.com/Panniantong/agent-reach/archive/main.zip, which downloads and installs code from an unverified GitHub repository not belonging to a trusted organization. - [CREDENTIALS_UNSAFE]: The command
agent-reach configure --from-browser chromeis designed to automatically extract sensitive session cookies from the user's local browser storage. - [EXTERNAL_DOWNLOADS]: The
agent-reach installcommand performs secondary downloads of various third-party binaries and CLI tools (Node.js, mcporter, xreach CLI, gh CLI, yt-dlp) from remote servers at runtime without integrity verification. - [COMMAND_EXECUTION]: The skill makes extensive use of subprocesses to call external command-line utilities (
xreach,yt-dlp,gh,mcporter), creating a broad attack surface if external inputs are not strictly sanitized. - [DATA_EXFILTRATION]: The skill ingests data from a wide range of external platforms (Twitter, Reddit, Bilibili, YouTube) and utilizes third-party markdown conversion services (
jina.ai), which may expose the content of requested URLs to those services.
Recommendations
- AI detected serious security threats
Audit Metadata