pentest-ai-killer-automation
Fail
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads a full automation suite and associated tools from an unverified third-party GitHub repository (github.com/vietjovi/pentest-ai-killer).
- [REMOTE_CODE_EXECUTION]: The installation process involves executing a shell script (install_pentest_tools.sh) from the downloaded repository using sudo bash, which grants administrative control to external, unvetted code.
- [COMMAND_EXECUTION]: The server and MCP integration expose a custom_command parameter that allows the agent or an external user to execute arbitrary shell commands on the host system. This creates a high risk of command injection, especially when processing untrusted input.
- [CREDENTIALS_UNSAFE]: The skill manages sensitive configuration such as X_API_KEY and AWS profiles (aws_profile). While placeholders are used in the documentation, the architecture encourages the transmission and storage of sensitive keys within environment variables and configuration files.
- [SAFE]: The skill has an indirect prompt injection surface. 1. Ingestion points: target domains and tool outputs from security scanners (nuclei, sqlmap, etc.). 2. Boundary markers: None provided to separate untrusted tool output from agent instructions. 3. Capability inventory: Arbitrary command execution via custom parameters, local file reading in reports directory, and network operations. 4. Sanitization: No evidence of input validation or output sanitization is mentioned in the documentation.
Recommendations
- AI detected serious security threats
Audit Metadata