The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The run.py script uses the mlx_lm library to download and load LLM weights from Hugging Face (mlx-community/Qwen2.5-7B-Instruct-4bit and mlx-community/Qwen2.5-Coder-7B-4bit). These models are hosted on a well-known service and are standard for the MLX ecosystem.
[COMMAND_EXECUTION]: The skill utilizes the clawdbot CLI tool to execute the Python inference script on designated MacBook nodes via the system.run command. This is consistent with the skill's stated purpose of model orchestration.
[INDIRECT_PROMPT_INJECTION]: The run.py script ingests untrusted data from the agent context (prompts) through command-line arguments or STDIN. While this creates an ingestion point for prompt injection into the local LLM, the skill's capabilities are limited to text generation and do not include high-risk side effects like filesystem writes or network exfiltration within the Python script itself.

mlx-brain