Agent Rules Kit: One AGENTS.md for Claude Code, Cursor & Codex (CLAUDE.md + Cursor Rules, 12 Stacks, Checker)

Stop maintaining three sets of agent instructions that drift apart.

This kit gives you one AGENTS.md as the source of truth for your coding agents, a CLAUDE.md that imports it, file-scoped rules for Cursor (.cursor/rules/*.mdc) and Claude Code (.claude/rules/*.md) generated from the same source, and a checker that finds the mistakes that make agents quietly ignore your rules.

Why it exists. Each tool reads different files. By default, Claude Code reads AGENTS.md only when the repo has no CLAUDE.md or CLAUDE.local.md, so adding a small CLAUDE.md can hide your AGENTS.md from it. Cursor ignores plain .md files in .cursor/rules. Codex reads only AGENTS.md files, and stops at 32 KiB by default. The kit is built around these rules, as documented by each tool in November 2026 (links included).

What's inside

• Core templates: an AGENTS.md covering commands, definition of done, boundaries ("ask first"), git, and a section for several agents (or sessions) working in one repo: worktree per task, named file scope, handoff notes. Plus a 3-line CLAUDE.md shim that imports it.

• 12 stacks, each prebuilt and ready to copy: Next.js (App Router), React + Vite, Node.js TypeScript API, FastAPI, Django, Python data/ML, Go, Rust, Rails, SvelteKit, React Native + Expo, Monorepo.

• 6 cross-cutting topics: testing, security, database migrations, API design, git & PRs, docs & comments.

• compose.py: pick any mix of stacks and topics and it writes AGENTS.md, CLAUDE.md, the Cursor .mdc rules (globs) and the Claude Code rules (paths) into your repo. It never overwrites files unless you pass --force. Use --inline to put the detailed rules into AGENTS.md too, for agents that read only that file.

• check_agent_rules.py: works on any repo. It flags unfilled placeholders, invalid .mdc frontmatter (including YAML that won't parse), .md files Cursor ignores, a CLAUDE.md that doesn't import AGENTS.md, a CLAUDE.local.md that hides AGENTS.md, nested AGENTS.md files Claude Code won't read, and files over each tool's size guidance. --json and --strict options make it usable in CI.

• A filled-in example (fictional monorepo with a Next.js app and a FastAPI service) that passes the checker with 0 errors and 0 warnings.

• Docs: a loading matrix (which tool reads which file, with sources), a tuning guide (how to write rules agents can follow, where each kind of rule belongs, a CI workflow snippet), and a 3-page Quick-Start PDF.

• Unit tests for both tools.

How it works

1. Copy a prebuilt folder into your repo, or run python3 tools/compose.py --stacks python-fastapi,react-vite --target your-repo.

2. Replace the {{PLACEHOLDERS}} with your real commands and conventions (about 15 minutes).

3. Run python3 tools/check_agent_rules.py your-repo.

Compatibility. Files follow the formats documented by Claude Code (CLAUDE.md, @imports, .claude/rules with paths), Cursor (AGENTS.md, .cursor/rules/*.mdc with description/globs/alwaysApply) and OpenAI Codex (AGENTS.md) as of November 2026. Other tools that read AGENTS.md can use the core file. Tools need Python 3.8+ (standard library only) on macOS, Linux or Windows. Everything is plain Markdown you can edit.

Honest limits. Instruction files guide agents; they don't enforce anything, and no template guarantees a particular result. Use hooks, linters, permissions and CI for hard rules. The stack rules are sensible starting points for you to edit, not a replacement for your team's conventions. Tool behavior changes often, so the docs link each source for you to re-check.

License. Use in any number of your own, your employer's or your clients' projects, commercial or not, and commit the files to those repos. Don't resell or redistribute the kit itself.

Digital download (ZIP). Not affiliated with or endorsed by Anthropic, Anysphere (Cursor), OpenAI, or any framework named. Product names are used only to describe compatibility.