AI Prompts for Agents
Write the files and rules that keep coding agents on your product, not inventing a new one every chat.
Prompts for designers or PMs help you draft the work. This page helps you lock that work into files agents reuse.
Start here
Fix a weak prompt, then turn a weekly job into a skill.
- AI Workflows
Optimize My Prompt
Diagnose a weak prompt and rewrite it with clearer task, context, constraints, and output format.
prompt-engineeringoptimize-promptchatgpt - AI Workflows
Create an Agent Skill (SKILL.md)
Turn a job you run often, like a UX audit or PRD critique, into a SKILL.md Cursor or Claude Code can load.
skill-mdagent-skillscursor
Project files for AI agents
Repo-root instructions agents actually read.
- AI Workflows
Create an AGENTS.md File
Write repo-root instructions so coding agents know the product, conventions, tools, and what needs approval.
agents-mdcoding-agentscursor - AI Workflows
Create a DESIGN.md File
Write a design-system file agents can follow for tokens, components, content, and accessibility.
design-mddesign-systemai-agents - AI Workflows
Create a CLAUDE.md File
Write Claude Code project instructions: stack, commands, design rules, tools, and what needs approval.
claude-mdclaude-codeproject-instructions
Before you give an agent tools
Permissions, review gates, and what happens when it fails.
From one-off prompt to a workflow that sticks
- 1Get one task working with a clearer prompt.
- 2Put lasting rules in AGENTS.md, DESIGN.md, or CLAUDE.md.
- 3If you run the same job every week, save it as a skill.
- 4Only then add tools and auto-actions, with approvals for anything you cannot undo.
- 5When the agent fails, update the file or skill. Do not just retry the same chat.
Related on AI UX Playground
Frequently asked questions
What are AI workflow prompts?
Prompts that produce setup files and operating rules for coding agents: AGENTS.md, DESIGN.md, CLAUDE.md, SKILL.md, a stronger task prompt, or a harness plan. Paste them into ChatGPT or Claude, then drop the output in your repo for Cursor or Claude Code.
What is the difference between AGENTS.md and CLAUDE.md?
AGENTS.md is meant to travel across tools. CLAUDE.md is for teams that live in Claude Code and need Claude-only commands, MCP tools, and habits. Put shared rules in one file and point the other at it. Do not keep two conflicting copies.
How is DESIGN.md different from AGENTS.md?
AGENTS.md covers how to work in the repo: conventions, tools, what needs approval. DESIGN.md covers how UI should look and behave: tokens, components, content, accessibility. Link DESIGN.md from AGENTS.md so agents do not invent a new visual system.
How are these different from prompts for designers or PMs?
Role pages help you write a journey map, PRD, or critique once. These prompts help you write the standing instructions so the next agent session follows the same bar without you re-explaining it.
What is an agent harness?
Everything around the model that makes an agent safe to run: instructions, context, tools, memory, who can approve what, how you check the result, and how a person steps in. Sketch that before the agent can send email, spend money, or merge code.
Where should I start?
If the output is junk, start with Optimize My Prompt. If you repeat the same job, make a skill. If the whole repo needs shared rules, write AGENTS.md and DESIGN.md. Use Design an Agent Harness when the agent gets tools and can take irreversible actions.