GitHub has added new controls for Copilot code review, including a default firewall, custom setup steps, broader instruction-file support, and runner settings that are separate from the Copilot cloud agent.
The July 17 changelog is a follow-up to a busy month of Copilot governance updates. On July 14, GitHub moved a /security-review command into the Copilot app and added Visual Studio MCP trust checks. The new code-review changes put more of the pull-request review surface under explicit repository and administrator control.
Instructions now come from the branch being reviewed
GitHub says Copilot code review now reads custom instructions from the pull request’s head branch rather than the base branch.
That matters for teams testing review rules. A developer can update instructions in a feature branch and validate how Copilot responds before merging those instructions into the main branch. GitHub says this includes copilot-instructions.md, *.instructions.md, agent skills, and AGENTS.md.
Copilot code review also now reads REVIEW.md, GEMINI.md, and CLAUDE.md files. That is a practical nod to how teams are already storing review guidance for multiple agent tools.
The review runtime is becoming its own environment
The runtime changes are more important than the file-name list.
GitHub says Copilot code review now uses a firewall by default. It also supports custom setup steps through copilot-code-review.yml, so teams can install dependencies or prepare the repository before review. Runner settings are now independent from the Copilot cloud agent, which lets administrators configure the review environment separately.
That separation is the key governance point. Code review is not the same task as a coding agent changing files. The review agent needs enough access to understand the change, but not a blank check to reach every dependency, network endpoint, or secret.
