New: watch your coding agents work, step by step, on a live board. Try the demo, no account

Free tools

CLAUDE.md and AGENTS.md generator

A good CLAUDE.md or AGENTS.md lists the commands to build and test, where things live, and the rules the agent keeps getting wrong, in under 200 lines. Fill in the form and copy the files, or keep one shared AGENTS.md and a CLAUDE.md that imports it with @AGENTS.md.

Runs in your browser. What you type stays on this page. By Jake Marsh. Checked against Claude Code docs: how Claude remembers your project (CLAUDE.md, imports, AGENTS.md) and AGENTS.md: a README for coding agents on 2026-11-16.

Only used for Next.js and Node starter commands.
Frameworks, database, hosting. Optional.
CommandsStarter values for your stack. Edit them to match the repo; blank ones are left out.
One per line as folder: what lives there.
Working rules
One per line. Specific beats general: say what to do and where.
Never let the agent
One per line.
Which files?

25 lines.

  • No project name yet; the heading will read "Project".
  • No one-line summary. A sentence on what the project does saves the agent a lot of guessing.

AGENTS.md

Repo root. Codex, Cursor, Copilot, Gemini CLI and most other coding agents read it. The one file you edit.

# Project

Stack: Next.js / React.

## Commands

- Install: `bun install`
- Run locally: `bun run dev`
- Build: `bun run build`
- Test: `bun test`
- Lint: `bun run lint`
- Type-check: `bunx tsc --noEmit`

## Working rules

- Add or update tests for every behavior change, and run them before saying the work is done.
- Before finishing, run the test, lint and type-check commands above and fix what they report.
- Keep each change small and on topic. Do not reformat or rename files you were not asked to touch.
- Read the neighboring files first and follow their patterns before introducing a new one.

## Do not

- Never commit secrets, API keys or .env files, and never print them in output.
- Never force-push, or rewrite history on a shared branch.
- Never delete, skip or weaken a test to make a run pass. Fix the cause.

CLAUDE.md

Next to AGENTS.md. Imports it, so Claude Code reads the same rules; add Claude-only notes below the import.

@AGENTS.md

How to use it

  1. Pick your stack and package manager. The command fields fill with starter values.
  2. Edit the commands to match the repo, then add a one-line summary and the folders an agent should know about.
  3. Tick the working rules and do-nots that apply, and add your own, one per line.
  4. Choose the file layout. One shared AGENTS.md with a CLAUDE.md that imports it is the least to maintain.
  5. Copy or download the files into the repo root, commit them, and run /context all in Claude Code to confirm CLAUDE.md is listed under Memory files.

What goes in the file, and what to leave out

Claude reads CLAUDE.md as context, not as enforced settings, so the more concrete a line is the better it holds. "Run `bun test` before committing" works; "test your changes" does not.

  • Put in: build, test and lint commands, the folders that matter, naming and style rules the code does not show, and things the agent got wrong twice.
  • Leave out: what the code already says, long tutorials, and rules that only apply to one folder. Those belong in a nested file, a path-scoped rule or a skill.
  • Keep it under about 200 lines. Longer files use more context and are followed less closely.
  • Never put secrets in it. The file is committed.

CLAUDE.md or AGENTS.md

Codex, Cursor, Copilot, Gemini CLI and many other agents read AGENTS.md. Claude Code reads CLAUDE.md, and falls back to AGENTS.md only when the repo has no CLAUDE.md. To share one file, keep the rules in AGENTS.md and make CLAUDE.md a single import line:

@AGENTS.md

## Claude Code

Use plan mode for changes under src/billing/.
CLAUDE.md

A symlink (ln -s AGENTS.md CLAUDE.md) also works, except on Windows checkouts, where Git may turn it into a plain text file. The import line is safer.

Questions

Where does CLAUDE.md go?
In the repo root as ./CLAUDE.md or ./.claude/CLAUDE.md, shared with the team through git. Personal notes go in CLAUDE.local.md (add it to .gitignore) or ~/.claude/CLAUDE.md for every project. Files in subfolders load when Claude works in them.
Do I need both CLAUDE.md and AGENTS.md?
Only if you use Claude Code next to other agents. Claude Code reads AGENTS.md when no CLAUDE.md exists, but once a CLAUDE.md is there it reads only that. The shared layout keeps the rules in AGENTS.md and imports them into CLAUDE.md.
How long should the file be?
Claude Code's docs suggest under 200 lines per file. The page counts lines as you edit and warns past that. If you need more, move folder-specific rules into nested files or .claude/rules.
Does the agent have to follow it?
No. It is context the agent reads, so a clear rule is followed most of the time, not every time. For something that must never happen, such as blocking a command, use a hook or a permission rule instead.
What does the Taskhold section do?
It tells the agent to file each request as a task on your shared list, report its steps and progress, and ask you questions there. It only works once the agent has the Taskhold connection; without it the agent skips the section.

Sources

Checked 2026-11-16.

Keep the agent's work on one list.

The file tells the agent how your repo works. Taskhold tells you what it is doing: each request becomes a task with steps, a progress bar and questions you answer from your phone. Tick the Taskhold box above to add the section, then try the live board or connect Claude Code from the Claude Code setup page. $12 a month or $120 a year, 7-day trial with a card.

Or connect your agent first, no sign-up: add https://taskhold.com/api/mcp as an MCP server. The first time it uses Taskhold, it gives you the trial link.

Related tools

  • Claude Code permissions builder

    Pick what Claude Code may run, what it must ask about and what it can never touch, and get the allow, ask and deny block for settings.json. Paste an existing file to check it.

  • Cursor rules generator

    Build .cursor/rules .mdc files from starter rules for your stack, with the frontmatter set correctly for each rule type.

  • Claude Code slash command generator

    Write the prompt once and get the .claude/commands file with the frontmatter, arguments, shell context and allowed tools wired in, plus a preview of what Claude receives.

All free tools