CLAUDE.md Generator for AI Coding Agents Prompt

Improved Prompt

Your task as a CLAUDE.md architect is to craft a concise, production-ready project instruction file tailored for AI coding agents like Claude Code, Cursor, Windsurf, and Zed. Your goal is to create a high-impact, actionable guide that solves real problems and avoids generic advice.

Project Description

I will provide you with some or all of the following details about my project:

Core technology stack, including programming languages, frameworks, databases, and tools.
A high-level overview of the project structure, including key directories and their contents.
A list of non-obvious coding conventions and file organization rules that my team follows.
Common pain points or errors that AI agents tend to make when working on projects of this type.
Description of the deployment and testing workflows specific to this project.

If I provide minimal details, you have permission to ask up to 5 targeted questions to clarify the project requirements before generating the CLAUDE.md file.

CLAUDE.md Specification

Your output should be a markdown file with the following sections, adhering to these strict constraints:

Section 1: Project Overview (3-5 lines maximum)

Project name (exact)
One-line purpose statement
Core technology stack (2-3 key components)

Section 2: Architecture Map (5-10 lines maximum)

Key directories and their primary contents, using either a compact tree representation or a flat list.
Entry points into the system (main files or functions).
Critical data flows or paths between components.
Avoid verbose descriptions and focus on what each directory or file does.

Section 3: Common Commands

List of essential commands, formatted as a simple reference table or bullet points:
- Build command (single target)
- Test command (unit and full suite)
- Lint command
- Development server command
- Deployment command
- Include the exact command syntax, flags, and target files where applicable.

Section 4: Code Conventions (only non-obvious ones)

Naming patterns for variables, functions, and directories.
File organization rules (e.g., domain-specific layers, feature folders).
Import ordering or dependency management rules.
Skip anything already enforced by linters or formatters.
Include no more than 3-5 critical conventions.

Section 5: Gotchas & Warnings

Project-specific traps or quirks that have bitten the team in the past.
Things AI agents tend to get wrong or misunderstand in this type of project.
Known workarounds or fragile areas of the codebase.
Limit this section to 5-7 specific items.

Section 6: Git & Workflow (only if team has specific conventions)

Branch naming convention (e.g., feature/issue-123, hotfix/breaking-change).
Commit message format (e.g., Subject: <72 chars> / Body).
Pull request process (e.g., required approvals, checks).
Limit this section to 3-5 key points.

Section 7: Pointers (Progressive Disclosure)

List of files or directories that Claude should read for deeper context when needed:
- For API patterns, see @docs/api-guide.md
- For database migrations, see @prisma/migrations
- Use "see" followed by the exact file path.
- Do not embed entire files or use "@import" syntax.
- Limit this section to 5-7 relevant files.

Quality Criteria

Before finalizing the CLAUDE.md, verify the following:

The file is under 150 lines total, including headers and blank lines.
Every "don't do X" has a clear "do Y instead" or workaround.
The file solves real problems and avoids generic advice.
Test, build, and lint commands are included with exact syntax.
No "@-file" imports or embedded files; only "see path" references.
"IMPORTANT" or "YOU MUST" is used sparingly (2-3 times maximum).
The file would be equally useful to a new team member and an AI agent.

Please review the project description I provide and generate a CLAUDE.md that meets these specifications and quality criteria. If you need clarification, ask no more than 5 targeted questions before proceeding with the generation.