---

name: agent-tooling description: "Operate GPT Academic Void Terminal, Code Interpreter, Commandline Assistant, and natural-language plugin dispatch workflows." disable-model-invocation: true metadata: disco-role: operating license: GPL 3.0

Agent Tooling

Use this sub-skill when the user wants GPT Academic to choose tools, run generated Python, interpret uploaded files, modify configuration, or execute/suggest shell commands through agent-like plugins.

Trigger phrases

Read this sub-skill for “虚空终端”, “Void Terminal”, “CodeInterpreter”, “动态代码解释器”, “Commandline Assistant”, “use natural language to call a plugin”, “modify GPT Academic config”, “run a command”, “generate code to process this file”, or “dynamic function generation”.

First decisions

1. Classify the request as direct plugin use, Void Terminal, Code Interpreter, Commandline Assistant, or too risky without confirmation.
2. Run the request classifier for a first-pass safety route:

python sub-skills/agent-tooling/scripts/classify_agent_request.py "<user request>"

3. Ask before any action that mutates files/config, runs shell commands, deletes data, calls external services, or executes generated code.
4. If a deterministic domain plugin exists, prefer the owning sub-skill over a generic agentic route.

Route map

User goal	Surface	Read next
ask GPT Academic to choose a plugin	Void Terminal / natural-language dispatch	references/workflows.md
run generated Python on an uploaded file	Code Interpreter	references/workflows.md, references/safety.md
ask for shell command help	Commandline Assistant	references/safety.md
modify app config by instruction	Void Terminal, but only with explicit confirmation	references/safety.md, root references/configuration.md
deterministic PDF/code/media workflow	owning plugin/sub-skill	sibling sub-skill first

Boundaries

• Use ../academic-docs/SKILL.md, ../programming-code/SKILL.md, or ../multimodal-media/SKILL.md first when the user already knows the domain workflow.
• Use ../conversation/SKILL.md for provider setup, search, RAG, and normal chat.
• Treat Code Interpreter and shell routes as execution workflows, not just conversation.