Write one AGENTS.md that every coding agent reads
Added ⧉ Copyable prompt below
Our verdict
The cheapest upgrade to any vibe-coded or agent-built project. AGENTS.md is an open format read by Cursor, Codex, Gemini CLI, Aider, goose and others, and Claude Code reads it when there is no CLAUDE.md. Keep it short and universal: practitioners report agents start ignoring bloated files.
- Who it's for
- Anyone using more than one coding agent, or handing a repo to agents regularly
- Leverage
- Every agent session starts with the same project context
- Effort
- Low
- Date added
What it does
AGENTS.md is plain Markdown at the repo root, a README for agents with no required fields. The format site recommends a project overview, build and test commands, code style, testing instructions and security notes, with nested files in monorepo packages (the closest file wins). HumanLayer's guide adds: cover what, why and how; keep it to universally applicable instructions; point to separate docs instead of pasting them in; leave style to linters; and write it by hand rather than auto-generating it.
How builders use it
- Create AGENTS.md at the repo root with the build, test and lint commands an agent must run before finishing.
- If you use Claude Code with an existing CLAUDE.md, symlink one to the other (ln -s AGENTS.md CLAUDE.md) so both stay in sync.
- Move task-specific detail into an agent_docs/ folder and list those files with one-line descriptions.
- For Aider add read: AGENTS.md to .aider.conf.yml; for Gemini CLI set context.fileName to AGENTS.md in .gemini/settings.json.
- Edit it whenever an agent makes the same mistake twice.
What to expect
Agents run your real test commands and follow project conventions without being told each session, and switching tools doesn't mean rewriting your instructions.
Copy this: AGENTS.md starter
# AGENTS.md
## Project
<one paragraph: what this app does and who it is for>
## Map
- app/: <what lives here>
- lib/: <what lives here>
## Commands
- Install: <cmd>
- Dev: <cmd>
- Test: <cmd>
- Typecheck/lint: <cmd>
Run test and typecheck before you say a task is done; fix failures.
## Rules
- Keep diffs small; don't refactor unrelated code.
- Never commit secrets; use env vars.
## More docs (read only if relevant)
- agent_docs/database.md: schema and migrations
- agent_docs/deploy.md: how releases workRelated discoveries
- ToolVibe-Coding & Prompt-to-App
Context7: up-to-date docs for AI editors
Feeds current, version-specific library docs into your coding assistant.
Effort: Low
- SetupVibe-Coding & Prompt-to-App
Cursor Rules and AGENTS.md
Persistent instructions so the AI follows your stack every time.
Effort: Low⧉ Prompt
- SetupVibe-Coding & Prompt-to-App
Reusable skills and slash commands in Claude Code
Package your repeated prompts as skills you can call with one command.
Effort: Low⧉ Prompt
- WorkflowVibe-Coding & Prompt-to-App
Give your agent a spec-to-ship lifecycle with agent-skills
Install a set of open-source engineering skills so your agent specs, plans, builds, tests, reviews and ships in order.
Effort: Low⧉ Prompt
Want more like this?
Practical discoveries, without the noise.