agents-md
Development standards for AI coding agents.
agents-md is an open-source, bilingual (English and Simplified Chinese) set of standards, templates and Claude Code skills that tell AI coding agents such as Claude Code, Cursor and Codex how to commit, branch, run CI/CD, migrate databases and ship safely.

agents-md at a glance
- What it is
- A specification, drop-in templates and installable skills for teams that let AI agents work in their repositories
- Works with
- Claude Code (CLAUDE.md and skills), Codex, Cursor and other agents that read AGENTS.md
- Languages
- English and Simplified Chinese, kept in step by a CI check
- Licence
- MIT, free and open source
- Version
- v0.2.0 (June 2026)
- Made by
- NGMOB Pte. Ltd., Singapore
What agents-md is
AI agents write code faster than Git workflows were designed to absorb. Wired in naively, they flood main with commits titled fix and wip, run up CI bills, trigger deployment storms and apply schema migrations through pooled connections. agents-md is the compatibility layer: rules that let an agent iterate quickly on its own branch while the shared system stays low-churn and safe for production.
The core of it fits in eleven rules. Agents may change code often but commit rarely, never push to main, ship one feature per pull request with squash merges, and deploy to production only from main. Migrations run on a direct database connection, retries are capped, and work is tracked in GitHub Issues that only a human can commit to.
There are four ways to adopt it: drop the CLAUDE.md and AGENTS.md templates into a repository, install individual Claude Code skills, add the whole standard as a Git submodule, or enforce the rules in CI with its lint scripts for Conventional Commits, branch names, bilingual docs and links.
- core rules
- 11core rules
- chapters in the standard
- 19chapters in the standard
- installable Claude Code skills
- 5installable Claude Code skills
- languages: English and Chinese
- 2languages: English and Chinese
What you can do with agents-md
- One-file standard
- STANDARDS.md covers workflow, branching, commits, CI/CD, databases, agent behaviour and platform engineering.
- Drop-in entry points
- CLAUDE.md and AGENTS.md templates give Claude Code, Codex and Cursor the minimum rules.
- Claude Code skills
- Git workflow, commits, CI/CD, database and agent-behaviour skills load when an agent needs them.
- CI lint scripts
- Check Conventional Commits, branch names, bilingual symmetry and links, plus a commitlint config.
- AI-specific chapters
- Secrets, prompts, evals, observability and machine-readable docs such as llms.txt.
- Issues-first workflow
- A Backlog-to-Kanban process for GitHub Issues and Projects with clear limits on agent authority.
Questions about agents-md
What is agents-md?
agents-md is an open-source set of AI-native development standards. It defines how AI coding agents such as Claude Code, Cursor and Codex should commit, branch, run CI/CD, handle database migrations and ship to production, and ships templates, skills and lint scripts that apply those rules.
Who makes agents-md and what does it cost?
agents-md is an open-source project from NGMOB, developed in the public GitHub repository maweis1981/agents-md. It is free under the MIT licence.
How do I add it to my project?
Copy templates/CLAUDE.md and templates/AGENTS.md into your repository root, or copy a skill folder into .claude/skills. You can also add the repository as a Git submodule, or reuse its lint scripts in your own CI.
Which agents does it support?
Claude Code reads CLAUDE.md and the skills; Codex, OpenAI agents and Cursor read AGENTS.md. The rules themselves are tool-neutral, so they apply to any agent that commits to a Git repository, including Devin and Windsurf.
Facts last checked on 2026-10-30 against github.com/maweis1981/agents-md. Built with Markdown, Claude Code skills, AGENTS.md, GitHub Actions, Bash, commitlint.
Writing about agents-md
- Cadre: Digital Employees You Can AuditCadre runs AI as role-based digital employees, and every job is source-checked, scored, approved and costed before it reaches a client. Here is how.
- Eleven Rules for Letting AI Agents Commit Codeagents-md is our open-source rulebook for AI coding agents: how Claude Code, Cursor and Codex should branch, commit, run CI and touch databases safely.
- Twenty-Two Products, One Stack: The NGMOB LineupEighteen live products and four labs, from Tapestra and Cadre to Duomi, GlowFrame, CarbClue and LaunchKit: what each one does and where agents fit in.