Decide once. Every agent obeys.
One shared playbook for all your coding agents.
Make a call once: “always use pnpm here”, “never push without asking”. Approve it, and Claude Code, Codex and Cursor all follow it, starting with the session you are in.
Works with
-
Claude Code -
Cursor -
Codex -
GitHub Copilot -
VS Code -
Gemini
You keep telling them the same things.
You told Claude Code to stop pushing without asking. Then you told Codex. Then a fresh session forgot, and you typed it again. Your rules live in a separate file for every agent, hand-edited, and no two of them say quite the same thing. Every agent you add makes it worse.
One playbook, read by every agent, at the start of every session.
PolicyLayer holds your rules in one place. Each agent connects to it and fetches the current playbook when a session starts, so they all work from the same instructions. Update the playbook once and you have updated every agent.
You decide. Your agent does the paperwork.
- 01
Start with the file you already have.
Paste in your CLAUDE.md or AGENTS.md. It becomes version one of your playbook. Nothing to rewrite, nothing lost.
- 02
New decisions are captured as you make them.
When you settle something mid-session (your agent asks “tabs or spaces?”, you answer), it writes your answer up as a one-line rule and hands you an approval link.
- 03
You approve with one click.
The rule joins the playbook under your name, and the agent you are working with picks it up straight away. Agents cannot write rules themselves. A person’s click is what makes a rule real.
Say it once and it holds: every agent, every session, without repeating yourself.
Why this beats a file.
- Every decision you approve is one you never make again.
- Rules are captured when you decide them, not when you remember to edit a file.
- One playbook instead of a CLAUDE.md for Claude Code, an AGENTS.md for Codex and rules files for Cursor, each quietly drifting its own way.
- Agents propose, people approve. Nothing enters the playbook on a model’s say-so.
- Yours to take anywhere. Export everything and switch agents without losing a rule.
One person answers. Everyone’s agents follow.
Solo, PolicyLayer stops you repeating yourself. On a team, it stops everyone repeating and contradicting each other. A question is answered once, by whoever owns that call, and every teammate’s agents follow the same rule from then on.
The playbook becomes your team’s source of truth for how agents work here. Someone joining tomorrow starts with agents that already know all of it.
Every approved decision is one the whole team never makes again.
Team workspaces are rolling out. Write to us and we will set yours up.
Every decision is on the record.
Every question, answer, approval and version is logged. For any rule you can see who asked, who decided, what they said and when it took effect. When an agent does something surprising, you check what it was actually told instead of guessing.
QuestionAnswerApprovalVersion
Four steps, one sitting.
- 1
Create a workspace for your repository. A workspace holds one repo’s playbook, the same scope as the CLAUDE.md it sits behind.
- 2
Import your rules file. It becomes playbook version one, word for word. The original stays downloadable.
- 3
Connect your agents over MCP. One configuration works across your agents. Your key lives in local config, never in a tracked file.
- 4
Point your CLAUDE.md at the playbook. Add a short block at the top of the file that tells agents to fetch the playbook at the start of every session. Keep anything repo-specific in the file; nothing is deleted or replaced.
Privacy: PolicyLayer receives the playbook you import, never your repository or source code.
Common questions.
- Does PolicyLayer see my code?
- No. It receives the playbook you import and the questions your agent asks. It never reads your repository or source code, and it can write to nothing of yours.
- What happens to my existing CLAUDE.md?
- It stays in your repo, untouched. You add a short block pointing agents at the playbook and keep anything else you like in the file. The imported copy becomes playbook version one, and you can export the full playbook at any time.
- Which coding agents work with it?
- Anything that speaks MCP: Claude Code, Codex, Cursor, Grok Build, Muse Code and whatever ships next. The playbook itself is plain text any agent can follow.
Decide it once.
Every agent, every session, every teammate, from now on.
Connect your agent ↗