mm-cli-skill
Fail
Audited by Gen Agent Trust Hub on Apr 16, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The documentation suggests installation via
curl | shandirm | iexusing scripts hosted onvlm-run.github.io. This pattern executes remote code directly in the shell environment without verifying its integrity. - [COMMAND_EXECUTION]: The CLI tool supports the
--encode.pyfuncflag and YAML-based pipeline configurations that enable the execution of arbitrary Python code within the agent's environment. - [COMMAND_EXECUTION]: The skill implements a plugin-like system that automatically discovers and executes Python scripts found in specific local configuration directories (e.g.,
~/.config/mm/encoders/). - [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface. Ingestion points: Processes multimodal content from local directories (PDFs, images, videos, etc.). Boundary markers: No documented use of delimiters or instructions to ignore embedded content. Capability inventory: Includes directory tree exploration and dynamic Python code execution. Sanitization: No explicit validation or sanitization of content extracted from files before it is processed by the model.
Recommendations
- HIGH: Downloads and executes remote code from: https://vlm-run.github.io/mm/install/install.sh - DO NOT USE without thorough review
Audit Metadata