The skill allows arbitrary shell command execution
Claims to do
TweakTune Synthesizer: You are an interactive assistant that helps users design and build **tweaktune pipelines** for synthesizing training data for large language models (LLMs). TweakTune is a Rust-powered, Python-facing library that provides a pipeline-based architecture for generating synthetic text, structured JSON, conversations, and function calling datasets using LLM APIs.
Actually does
The skill uses `Read`, `Write`, `Glob`, `Grep`, and `Bash` tools to interactively gather user requirements. It then generates Python `tweaktune` pipeline code, `requirements.txt`, and other supporting files. The generated code is designed to read data from local files, HuggingFace datasets, or databases, interact with LLM APIs (OpenAI, Azure OpenAI, generic APIs) for data synthesis, and write the results to local files.
Shell command execution function detected
system(
The skill explicitly allows the `Bash` tool, enabling the agent to execute arbitrary shell commands. This poses a critical risk for system compromise, resource abuse, persistence, and privilege escalation.
allowed-tools: [...] - Bash
The combination of `Read`, `Write`, and `Bash` tools grants the agent the ability to read sensitive files from the system and exfiltrate them to external destinations or write them to accessible locations.
allowed-tools: - Read - Write - Glob - Grep - Bash
The skill offers an option for users to directly input API keys, which the agent will then embed into the generated Python code. Although a warning is provided, the agent facilitates the creation of code with hardcoded sensitive credentials.
API key source? ... c) Direct input (will be in code - warn about security)
The skill allows for 'Custom Validation' and 'Custom Steps' using `lambda` functions or `CustomStep` classes. If user input is directly incorporated into these generated code blocks without sanitization, it could lead to arbitrary code execution within the generated pipeline.
.validate(lambda data: your_validation_logic(data))
The skill references specific internal file paths (e.g., `/home/jovyan/SpeakLeash/...`) for documentation and test files. This discloses internal system structure and potentially sensitive project details.
Reference Files: - /home/jovyan/SpeakLeash/tweaktune/tweaktune-python/tests/test_basic.py
[](https://mondoo.com/ai-agent-security/skills/github/qooba/tweaktune/tweaktune-synthesizer)<a href="https://mondoo.com/ai-agent-security/skills/github/qooba/tweaktune/tweaktune-synthesizer"><img src="https://mondoo.com/ai-agent-security/api/badge/github/qooba/tweaktune/tweaktune-synthesizer.svg" alt="Mondoo Skill Check" /></a>https://mondoo.com/ai-agent-security/api/badge/github/qooba/tweaktune/tweaktune-synthesizer.svgSkills can read files, run commands, and access credentials. Mondoo helps organizations manage the security risks of AI agent skills across their entire fleet.