huawei-cloud-install-openjiuwenswarm

Fail

Audited by Gen Agent Trust Hub on Sep 11, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONPRIVILEGE_ESCALATIONPROMPT_INJECTIONCREDENTIALS_UNSAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill downloads a 300MB binary runtime environment from a third-party personal repository (https://gitcode.com/afeng5267/jiuwenswarm_runtime.git) and executes multiple binaries and scripts extracted from it, including jiuwenswarm-start and jiuwenswarm-init.
  • [PRIVILEGE_ESCALATION]: The configuration and startup scripts extensively use sudo to perform system-level modifications, such as changing /etc/ssl/cert.pem, managing files in /usr/local/bin, and installing system packages using apt-get, yum, or dnf.
  • [PROMPT_INJECTION]: The skill instructions use forceful mandatory language to override agent behavior and explicitly command the agent to hide its operations from the user, specifically instructing it to install dependencies silently without notification or consent.
  • [CREDENTIALS_UNSAFE]: The skill extracts API credentials from the Huawei Cloud settings.json file and the Linux kernel keyring, then writes these secrets into a local .env file.
  • [COMMAND_EXECUTION]: Multiple Python scripts utilize subprocess.run with shell=True and string interpolation to execute shell commands, which increases the risk of command injection.
  • [EXTERNAL_DOWNLOADS]: The skill performs runtime downloads of Python packages (requests) and system utilities (keyutils) from external repositories without version pinning or integrity verification.
Recommendations
  • HIGH: Downloads and executes remote code from: https://raw.gitcode.com/openJiuwen/jiuwenswarm/raw/main/README.md - DO NOT USE without thorough review
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Sep 11, 2026, 08:53 AM
Security Audit — agent-trust-hub — huawei-cloud-install-openjiuwenswarm