install-qwenpaw
Fail
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONPERSISTENCECREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill downloads and executes remote shell scripts by piping them directly to
shandbash. This pattern is found in both thescripts/setup.shscript and the manual instructions inSKILL.mdfor installing theqwenpawtool fromhttps://qwenpaw.agentscope.io/install.sh. - [PERSISTENCE]: The installation process modifies the user's shell profile (
~/.bashrc) to add the application's binary directory to the systemPATH. This ensures that the application remains persistently available and executable in all future terminal sessions. - [CREDENTIALS_UNSAFE]: The
setup.shscript is designed to accept sensitive credentials, such as Alibaba Cloud API keys and DingTalk secrets, as positional command-line arguments. This is an insecure practice because command-line arguments are typically visible to all users on a multi-user system via process monitoring tools likeps. - [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface for indirect prompt injection by configuring the agent to read and follow instructions from local markdown files used for long-term memory and identity.
- Ingestion points: The agent reads files in the
~/.qwenpaw/directory, includingAGENTS.md,SOUL.md, andMEMORY.md, during initialization and heartbeat cycles. - Boundary markers: The provided templates contain headers but lack robust instructions or delimiters to prevent the agent from obeying malicious prompts that might be embedded in the stored memory data.
- Capability inventory: The skill provides the agent with capabilities for shell command execution, file system manipulation, and network communication.
- Sanitization: There is no evidence of validation or sanitization of the content ingested from these persistent memory files.
- [EXTERNAL_DOWNLOADS]: The skill fetches the
uvpackage manager installer from the official domain of Astral (astral.sh), which is a well-known service provider. - [COMMAND_EXECUTION]: The skill executes various shell commands to configure the environment, including directory creation, setting restrictive file permissions (
chmod 600andchmod 700), and launching theqwenpawapplication as a background service usingnohup.
Recommendations
- HIGH: Downloads and executes remote code from: https://astral.sh/uv/install.sh, https://qwenpaw.agentscope.io/install.sh - DO NOT USE without thorough review
- AI detected serious security threats
Audit Metadata