(危险！)万能解释器工具(danger!)(omnipotent_interpreter_tool)

This pack's author didn't caption most nodes with a warning. This one gets two: the display name reads "(危险！)万能解释器工具(danger!)(omnipotent_interpreter_tool)" - "danger" stated once in Chinese and once in English, and "omnipotent" (万能, literally "all-purpose/all-powerful") describing what it actually does. When the person who wrote the node feels the need to say "danger" twice in its own name, that's worth taking at face value rather than reading as marketing.

Functionally, it sits in the same family as this pack's other code-execution tool, interpreter_tool: it registers a capability an LLM/agent node can call, and when the model decides it needs to compute something - math, text parsing, anything a language model shouldn't be trusted to eyeball - it writes code and this tool runs it, with the result flowing back into the conversation. The naming difference between the two isn't decorative. Plain interpreter_tool gets no warning at all; this one gets called dangerous and omnipotent twice over, which reads as the broader, more capable version of the same idea. Treat it as the one to be more careful with of the two, not less.

How it works

Wire the tool output into this pack's LLM/agent node's tool-list input. During a conversation, if the model decides code execution helps, it writes and runs code through this tool on its own - you don't write the code, and you don't see it happen unless you're logging the conversation.

The inputs and outputs that matter

• is_enable - the only input, an enable/disable dropdown rather than the plain boolean checkbox most of this pack's other toggles use, but functionally the same: it's the bypass switch.

Output: tool (STRING) - the serialized tool definition, meant to be wired into an agent node's tool-list input, not read directly.

How to install it

• ComfyUI Manager: search "comfyui_LLM_party", install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/heshengtao/comfyui_LLM_party, then pip install -r requirements.txt from inside the pack folder using ComfyUI's own Python, then restart.

No API keys of its own - it rides on whatever LLM node you've already got configured, since the LLM is what decides to invoke it.

Common issues & troubleshooting

Do not wire this into an agent that takes untrusted input. ComfyUI's custom-node ecosystem already runs arbitrary Python with full OS-level access and no sandbox by design - that's the baseline risk of installing any custom node at all. This tool adds a second layer on top: an LLM deciding what code to write and run, potentially based on a user's message rather than your own instructions. If you're building something a stranger can type into, don't give it this tool, or interpreter_tool, without a very deliberate reason.

You inherited a workflow with this node already wired in. Read what persona/system prompt is actually steering the agent before running it - the risk here isn't the node itself, it's not knowing what an LLM with code-execution has been told to do.

The LLM never uses it, or the call fails. Same as any tool-calling setup in this pack: the model needs a reason in its instructions to reach for it, and it needs to be running through an LLM node that actually supports tool/function calling - not every model or API configuration does, and one that can't will either skip the tool or hallucinate a plausible-looking result instead of really executing anything.