---
title: "Govern GitHub Copilot CLI"
---

# Govern GitHub Copilot CLI

Enforce Neural Inverse Govern agent policy in GitHub Copilot CLI with a root-owned policy hook, its one uncovered case, and how Copilot usage is monitored.

## Control level

| What | Level |
|---|---|
| Tool calls | **Enforced** with a policy hook in `/etc/github-copilot/policy.d` |
| MCP rules that name a server | **Not covered**: the hook input does not name the MCP server separately |
| Model traffic | **Not routable**: no supported setting sends Copilot's requests elsewhere |
| GitHub Copilot usage | **Monitored** through the vendor connector (seats, usage and metrics) |

This is the most strongly proven of the vendor tools. Copilot CLI 1.0.98 was run for real, offline, with the compiled policy file and the real endpoint agent: a denied `rm -rf` was refused, also with `disableAllHooks` in the developer's settings and with the agent binary missing, and an allowed `ls` ran.

## 1. Write the policy and enrol the machine

Write rules in **Agent policy** ([Agent policy](https://neuralinverse.com/docs/govern/agent-policy/)) and enrol the machine with the endpoint agent ([Deploy](https://neuralinverse.com/docs/deploy/)). The agent fetches the policy compiled for the `copilot-cli` engine and installs `policy.d/keystone.json` in `/etc/github-copilot/policy.d`.

Policy hooks there must be root-owned, are not turned off by `disableAllHooks`, and run regardless of folder trust.

## 2. How failures and asks behave

- A crash or non-zero exit of the hook denies the call.
- A timeout lets the call through, so Govern's hook always answers before its deadline.
- Copilot CLI supports asking, so `ask` rules ask the developer.

Check the gaps on **Agent policy > Policy preview** with the engine `copilot-cli`.

A hook file in a repository (`.github/hooks`) is not locked and only advisory. Only the root-owned policy directory gives the Enforced level.

## Monitoring Copilot

GitHub Copilot seats, usage and metrics come in through a vendor audit connector; see [Evidence](https://neuralinverse.com/docs/evidence/).