GPT-Image-2 Case · Multi-Agent LLM System Architecture

A featured GPT-Image-2 reference case from the “Research Paper Diagrams”, suitable for retrieving generation effects in the “Multi-Agent LLM System Architecture” domain.

Author: AI Plus Lab

✦Results

◌Case Background

来自 GPT-Image2-Skill README 的精选展示条目，适合作为“研究论文图示 / 多智能体 LLM 系统架构”方向的站内参考案例。

⌘Prompt Content

Horizontal 16:9 high-fidelity system diagram depicting a multi-agent LLM architecture, styled akin to detailed AutoGen / LangGraph / Anthropic Managed Agents figures. Subtle shadows, warm copper highlights, numbered process markers ①②③④.

Area 1 — "User interface": Rounded user box containing placeholder task "research question: summarize recent red-teaming attacks and reproduce the top three".

Area 2 — "Orchestrator layer": Central hexagonal node "Planner LLM", top edge in warm copper. Three satellite chips: "Task decomposition", "Agent routing", "Re-plan on failure". Small embedded chip "prompt cache hit ~98%".

Area 3 — "Specialised workers": 2×2 hexagons, labeled "Researcher" / "Coder" / "Critic" / "Writer", each with symbols and status strips ("idle", "running step 3/5", "done", "running step 2/4"). Central label "async message bus".

Area 4 — "Tools & memory": (a) "Tool registry" panel lists "web_search ×41", "python_exec ×27", "read_file ×18", "write_file ×12", "browser_use ×7"; (b) "Memory" panel includes "Short-term scratchpad" and cylinder "Long-term vector store — 1.8M episodes".

Bottom embed "Example trace": 8-step horizontal timeline from "User asks" to "Planner decomposes", "Researcher: web_search(...) ", "Coder: python_exec(...) ", "Critic: verify", "Re-plan" (circular arrow), "Writer: compose final answer".

Title: "Agentic LLM system: planner orchestrates specialised workers over a shared tool and memory layer". Subtitle: "adapted from AutoGen (Wu et al., 2023), LangGraph, and Anthropic Managed Agents patterns".

✎Outcome Notes

This case has been moved from the GPT-Image2-Skill README to AIPlusLab `/prompts` for convenient on-site retrieval, browsing, and reuse.