One File, Two AI Code Reviews: GitHub Copilot Now Reads Your CLAUDE.md

TL;DR: Copilot code review now reads CLAUDE.md and AGENTS.md from your PR's head branch — teams using both Claude Code and GitHub Copilot code review can enforce code conventions from a single file.

GitHub shipped code review customization improvements on July 17. Two changes matter for Claude Code shops:

CLAUDE.md is now a code-review input

Copilot code review scans CLAUDE.md, AGENTS.md, GEMINI.md, and any *.instructions.md files in your repo alongside .github/copilot-instructions.md. Claude Code already uses CLAUDE.md as its behavior spec — now Copilot code review reads the same file. One file drives consistent conventions in both tools.

Instructions come from the HEAD branch

Copilot now reads instructions from the PR's feature branch, not the default branch. You can iterate on new review instructions in a feature branch and test them before merging — no more needing to merge instruction changes to main first.

Setting it up

# .github/copilot-code-review.yml — optional: install deps the review needs
name: Copilot Code Review Setup
steps:
  - run: pip install -r requirements.txt
  - run: npm ci

If your CLAUDE.md is already well-structured (coding style, test expectations, commit conventions), Copilot code review picks it up automatically. Add a ## Code Review section for review-specific guidance — for example, flagging when coverage drops or when public APIs lack docstrings.

The Copilot code review firewall is also now independently configurable from the Copilot cloud agent, giving you separate network access controls for each. And both agents can now be assigned different runner types in org settings — useful when your review agent needs a larger machine than your cloud agent.

Sources: GitHub Changelog: Copilot code review customization (Jul 17)