Agent Designer — Multi-Agent System Architecture
Agent Designer is a skill for designing multi-agent systems, generating provider-ready tool schemas, and evaluating agent execution logs. It provides deterministic scripts for architecture planning, schema validation, and performance analysis, producing JSON outputs with designs, diagrams, and recommendations.
✨ What it does
- Deterministic architecture planner that scores patterns (single agent, supervisor, pipeline, hierarchical, swarm) from r
- Tool schema generator that validates schemas and outputs Anthropic and OpenAI formats.
- Execution log evaluator that computes success rate, latency, cost, bottlenecks, and SLA compliance.
- Verification loop requiring zero invalid schemas and zero critical issues before design completion.
- Mermaid diagram generation for visualizing the architecture.
- Implementation roadmap and optimization recommendations.
🎯 When to use it
- When designing a new multi-agent system from requirements, including choosing an orchestration pattern and defining agent roles.
- When generating tool schemas in Anthropic or OpenAI formats from plain tool descriptions.
- When evaluating agent execution logs to identify bottlenecks, errors, cost issues, or SLA compliance.
🚀 How to use
Trigger by asking to design a multi-agent system, generate tool schemas, or evaluate agent logs. Provide a requirements JSON (copy sample_system_requirements.json) for design, tool descriptions JSON for schemas, and execution logs JSON for evaluation. Run the provided Python scripts in sequence. Example prompts:
Design an agent architecture for research automation.
Generate Anthropic tool schemas from these tool descriptions.
Analyze these agent run logs for bottlenecks.
📄 Output: JSON files: architecture design (arch.json), tool schemas (tools.json, tools_anthropic.json, tools_openai.json), and evaluation results (eval.json) with split files.
📦 Add this skill to Claude Code
# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills
# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/engineering/skills/agent-designer .claude/skills/agent-designerSkill source: engineering/skills/agent-designer/SKILL.md
⚠️ Good to know
Not for Claude Code workflow files (use workflow-builder) or single-agent prompt design (use agent-workflow-designer). Requires Python and the provided scripts.
❓ FAQ
What orchestration patterns does the planner consider?
The planner considers single agent, supervisor, pipeline, hierarchical, and swarm patterns, and scores them deterministically based on requirements.
What is the validation gate for tool schemas?
Every tool must print '✓ Valid' when running tool_schema_generator.py --validate; if any are invalid, fix them before proceeding.
How do I know when the design is complete?
The design is complete when there are 0 invalid schemas and the evaluator reports 0 critical issues on a pilot run; otherwise, apply recommendations and re-run.
🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.