How to use the AI Agent Development Workflow agent skill

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file with instructions that your AI assistant reads when a task matches the skill's description. This skill is an index-style workflow. It tells the assistant how to structure a project for building AI agents and divides the work into seven phases: agent design, single agent implementation, multi-agent system, agent orchestration, tool integration, memory systems, and evaluation.

Each phase lists related skills to invoke, a set of actions and a copy-paste prompt. The skill does not run code itself. It shapes how the assistant plans and works through an agent-building task, and it points to other skills, such as CrewAI and LangGraph guides, for the detail of each phase.

• Phases cover tools, memory and evaluation, not only the agent loop.
• A simple quality checklist: agent logic working, tools integrated, memory functional, orchestration tested, evaluation passing.
• The skill says to choose only the phases that match your task.

How do you install the ai-agent-development skill?

Download the ZIP for this skill from this page. Then either unzip it into your personal skills folder, ~/.claude/skills/, or into a project's .claude/skills/ folder so that it applies only to that project.

If you use the Claude apps instead, open Settings, go to Capabilities, and upload the skill under Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload

What does the skill need before you start?

The skill lists no required tools, does not run scripts and does not use the network. Its only file is SKILL.md. The skills it points to, such as crewai and langgraph, are not included in this download, so you will need those separately if you want their implementation details.

• Requires: nothing listed
• Runs scripts: no
• Uses network: no
• Files: SKILL.md

How do you trigger the skill with a natural request?

Describe an agent-building task in plain language and name the part you need help with. The assistant will match your request to this skill's description, which covers autonomous agents, multi-agent systems and agent orchestration with CrewAI, LangGraph and custom agents. You can also call one of the phase prompts directly, such as the one for Phase 3.

The skill's own prompts use the @ form, for example asking the assistant to use the CrewAI skill to build a multi-agent system with roles.

• "Help me plan a single autonomous agent that answers support tickets and calls a lookup tool."
• "I need to choose between one agent and several agents with CrewAI for a research pipeline."
• "Use @agent-evaluation to evaluate agent performance"

What should you expect from the output?

The skill asks the assistant to define purpose, capabilities, tools, memory and success metrics before building. For evaluation it asks for criteria, test scenarios, measured performance and edge cases, and to record evaluation inputs, observed results and limitations.

The skill also says that multi-agent phases describe the product you are building. They do not authorise the assistant to spawn or delegate to other agents while doing the task, and a single-agent architecture is preferred when it fits. Cloud calls, host configuration, publishing, destructive operations and spending need separate authorisation.

• Use it to plan and structure the work, then check the result yourself.
• Expect the assistant to ask for clarification if inputs, permissions or success criteria are missing.

What are the safety notes?

The skill itself runs no scripts and makes no network calls. The risk comes from the work it guides. The skill states that cloud calls, host configuration, publishing, destructive operations and spending require authorisation, and that a checklist does not grant it.

The skill is not a substitute for environment-specific validation, testing or expert review. Do not rely on it for production deployment decisions.

• Review any commands or cloud actions the assistant proposes before approving them.
• Keep production decisions with your own team and testing.

What tips help you get the most from it?

Start with a clear purpose and a short list of tools the agent needs. Pick one phase at a time rather than running the whole workflow, because the skill is broad and adds process overhead to small tasks. When you choose between single-agent and multi-agent designs, ask the assistant to justify the choice against your task.

• Give a concrete task, not a vague goal.
• Ask for evaluation inputs and limitations to be written down.
• Skip this skill for simple one-off coding questions.