CLAUDE.md Generator Prompt

Generate a concise CLAUDE.md repository instruction file from verified project commands, conventions, boundaries, and existing documentation.

Prompt Template

You are a repository-instructions editor generating the smallest useful CLAUDE.md for Claude Code from verified project facts. Inspect the supplied repository tree and existing documentation before drafting. Treat README files, contribution guides, package manifests, scripts, CI workflows, formatter and linter configs, AGENTS.md files, existing CLAUDE.md files, and nested directory instructions as evidence with explicit scope. Use current dated official Claude Code documentation supplied by the user or available to you for any product-specific behavior. Never invent commands, paths, architecture, conventions, tools, versions, test results, deployment rights, secret names, or supported features. Do not copy volatile documentation into the file when a stable link or repository command is enough.

Repository root and sanitized file tree: [paste]
Project purpose and major components: [verified details]
Existing instruction and documentation files with scopes: [paste or list]
Exact install, development, format, lint, type-check, test, build, and preview commands: [verified commands]
Languages, runtimes, package managers, frameworks, and exact versions: [details]
Code style, naming, testing, documentation, and review conventions: [verified rules]
Generated, vendored, migration, lock, snapshot, and protected files: [details]
Architecture boundaries and directories requiring special handling: [details]
Allowed and forbidden mutations, network actions, commits, pushes, and deployments: [details]
Security, privacy, credential, customer-data, and logging boundaries: [details]
Known failure modes, environment requirements, and validation limits: [details]
Desired file location and nested-scope needs: [root, subdirectory, both]

Provide:
1. A verified, inferred, and unknown facts table with evidence paths.
2. A conflict and precedence review across existing instructions and documentation.
3. One concise proposed CLAUDE.md in Markdown containing only durable, actionable repository guidance.
4. Separate sections for project map, safe commands, coding conventions, tests and validation, protected areas, and security boundaries only when supported by evidence.
5. Recommendations for nested CLAUDE.md files only where a directory genuinely needs different instructions, with the intended scope stated.
6. A removal list for redundant, obvious, temporary, or unverified guidance that should not enter the file.
7. A validation checklist that runs only verified commands and never fabricates results.
8. Open questions that must be resolved before replacing an existing file.

Do not edit or overwrite a repository file unless explicitly authorized. Preserve stronger existing safety rules, keep secrets and internal data out of the draft, and use placeholders for every unresolved command or path.

Example Output

Proposed CLAUDE.md

# [Project]

Project map

- [source directory]: [verified purpose]

- [test directory]: [verified purpose]

Commands

- Install: [verified install command]

- Check: [verified lint or type-check command]

- Test: [verified test command]

- Build: [verified build command]

Working rules

- Follow [existing style guide path].

- Keep changes within the requested scope and preserve unrelated worktree edits.

- Do not edit [verified generated or protected paths].

- Add or update tests for behavior changes where the existing suite supports them.

Safety

- Never commit credentials, customer data, or local environment files.

- Do not push, deploy, migrate data, or run destructive commands without explicit authorization.

Still unresolved

- [command or convention requiring owner confirmation]

The draft contains no guessed scripts or paths; each bracketed item must be resolved from repository evidence before use.

Tips for Best Results

  • ๐Ÿ’กProvide the actual tree, scripts, configs, and existing instruction files so the draft can cite repository evidence instead of guessing.
  • ๐Ÿ’กKeep CLAUDE.md focused on durable project-specific rules; obvious coding advice and temporary task notes create noise.
  • ๐Ÿ’กUse nested instruction files only for directories with genuinely different commands, conventions, or safety boundaries.

Frequently Asked Questions

What is the CLAUDE.md Generator Prompt prompt?

Generate a concise CLAUDE.md repository instruction file from verified project commands, conventions, boundaries, and existing documentation. It's a free Claude prompt template from our Coding collection โ€” copy it, fill in the bracketed variables, and paste it into your AI tool.

Which AI tools work with this prompt?

It's written and tested for Claude, ChatGPT and Gemini. Any AI assistant that accepts free-form text prompts will handle it well.

How do I customize this Claude prompt?

Replace the bracketed variables โ€” such as [paste], [verified details], [paste or list] โ€” with your own details before running it. Provide the actual tree, scripts, configs, and existing instruction files so the draft can cite repository evidence instead of guessing.

Is this prompt free to use?

Yes. Every prompt on PromptAtlas is free to copy, customize, and use โ€” no signup required.