Does one unanswered question stop the agent?
No. It pauses only that task and continues unrelated work. A later session collects the answer with get_answers.
Never answer your coding agent twice.
Your team’s decisions, in one playbook every agent works from. When something is not covered, the agent asks the right person.
npx @policylayer/setup One command. No code access. Nothing to migrate.
Prefer manual setup? Open the panelfirst session · your machine
> ship it
The tests have not run. Should I deploy?
> no. never deploy without a green run.
record_decision({ rule_body, decision, reference, … })
✓ Approved by you · Playbook v17
next week · your teammate’s machine
> ship it
Playbook v17 requires a green run. Running tests.
✓ 214 passed · deploying
I’m a
Works with 
Claude Code 
Codex 
Cursor 
GitHub Copilot 
VS Code 
Gemini
Claude Code Codex Cursor GitHub Copilot VS Code Gemini 01 / The playbook
Team rules, versioned and signed.
Every connected agent works from the same decisions. Change a rule once and every agent gets it. Agents make fewer avoidable mistakes, and you spend less time correcting them.
acme/api · playbook v17 ▪ 4 agents connected
Rule Approved by Since
Never deploy without a green run Approved by You
Since v17
Never force-push shared branches Approved by D. Ives
Since v14
Use pnpm, never npm Approved by P. Shah
Since v12
Waiting on you
Which licence should new source files use?
claude-code-4127 · 12m ago
History
Can I force-push to clean up main’s history?
Never force-push shared branches. Open a pull request instead.
answered by dana@acme.com · 2d ago
✓ Answer applied · case resolved
02 / The ask
Anything not covered becomes a question for its owner.
PolicyLayer routes the question to whoever owns that decision. The agent continues with work the playbook already covers.
03 / Connect
One command. One click. Your agents start from the same playbook.
Run it in your repository and approve the pairing link in your inbox. PolicyLayer connects Claude Code, Codex and Cursor, imports your existing rules as playbook v1 and verifies everything.
Prefer to control each step? Use manual setup Terminal The whole setup
# run in your repository
› npx @policylayer/setup
# pairing code
PL-7K2M · check your inbox
✓ key stored · 3 agents connected ✓ playbook v1 imported · verified
The MCP interface ▪ human approval required
01
get_playbook Load the latest signed playbook 02
ask_policy Route a question to its owner 03
get_answers Collect human answers 04
record_decision Propose a reusable rule
✓ Approved by you · new playbook published
04 / Common questions
Before you connect.
Three practical answers about how PolicyLayer behaves.
What does PolicyLayer see?
The playbook you import, the questions your agents ask and the answers people give. It never reads your repository or source code.
What happens if PolicyLayer is unavailable?
Agents keep using the last signed playbook. They receive no new authority until PolicyLayer is back.
Every agent needs to know what it can decide. PolicyLayer is how company judgment becomes machine authority.
Connect your coding agent ↗
Prefer to talk first?
Talk to us about a team rollout