Winbda/agent-designer

AI Agent Designer

Overview

Designs AI agent architectures — single agents or multi-agent systems — with clear roles, tools, memory, and orchestration patterns.

Workflow

Step 1: Define the Agent

1. Objective: What should the agent accomplish?
2. Scope: What's in and out of bounds?
3. Tools needed: What APIs/services should it access?
4. Autonomy level: Fully autonomous, human-in-the-loop, or supervised?
5. Platform: LangChain, CrewAI, AutoGen, custom, or conceptual?

Step 2: Choose the Architecture

| Pattern | When to Use | |---------|-------------| | Single agent + tools | One task, multiple steps | | Sequential chain | Linear workflow, each step feeds the next | | Router agent | Different tasks need different specialists | | Parallel agents | Independent tasks that can run simultaneously | | Hierarchical | Manager delegates to worker agents | | Collaborative | Agents discuss and refine outputs |

Step 3: Design the System

For each agent:

## Agent: [Name]

### Role
[One sentence — what this agent does]

### System Prompt

[Full system prompt]

### Tools Available
| Tool | Purpose | Input | Output |
|------|---------|-------|--------|
| [tool] | [why] | [format] | [format] |

### Memory
- **Short-term**: [conversation context]
- **Long-term**: [persistent storage approach]
- **Shared**: [what other agents can access]

### Decision Logic
- IF [condition] → [action]
- IF [condition] → [escalate to human]
- IF [error] → [fallback]

### Output
[What this agent produces and where it goes]

Output Format

# AI Agent System: [Name]

## Architecture Overview
**Pattern**: [architecture type]
**Agents**: [count]
**Orchestration**: [how they coordinate]

## System Diagram

[Text-based architecture diagram]

## Agent Specifications

### Agent 1: [Name]
[Full specification per template above]

### Agent 2: [Name]
[Full specification]

## Orchestration Logic
[How agents communicate and coordinate]

## Error Handling
[What happens when things go wrong]

## Human Touchpoints
[Where and when humans intervene]

## Implementation Guide
- **Platform**: [recommendation]
- **Models**: [which LLMs for which agents]
- **Estimated cost**: [per-run cost estimate]
- **Setup steps**: [how to build it]

Quality Checklist

• [ ] Each agent has a clear, single responsibility
• [ ] Tools defined with input/output formats
• [ ] Orchestration logic handles all paths
• [ ] Error handling and fallbacks defined
• [ ] Human oversight points identified
• [ ] Cost estimate included