Feature request: support AGENTS.md as a native context file alongside CLAUDE.md

Summary

Claude Code currently reads CLAUDE.md as its native per-repository instruction file. This works well, but AGENTS.md is a widely adopted, tool-agnostic convention used by Codex and other AI coding agents for the same purpose. Supporting AGENTS.md natively would make Claude Code first-class in multi-agent workflows without requiring duplicate files.

Current behavior

Claude Code reads CLAUDE.md at the repository root. AGENTS.md is not automatically loaded. Users who follow the AGENTS.md convention (common in repos that also use Codex) must either:

• Maintain a separate CLAUDE.md that just points to AGENTS.md, or
• Duplicate instructions across both files

Desired behavior

Claude Code should read AGENTS.md natively — either as a fallback when CLAUDE.md is absent, or in addition to CLAUDE.md with a defined precedence order (e.g., CLAUDE.md wins on conflict).

A reasonable reading order would be:

1. CLAUDE.md (Claude-specific overrides, if present)
2. AGENTS.md (shared cross-tool instructions)

Why this matters

AGENTS.md is tool-agnostic by design. Teams standardizing on it get consistent agent behavior across Codex, Cursor, and other tools for free. Requiring a separate CLAUDE.md creates friction and risks instruction drift when the two files diverge.

Supporting AGENTS.md would make Claude Code a better citizen in repos that are already following this convention — without breaking any existing CLAUDE.md behavior.

Workaround

Currently working around this with a minimal CLAUDE.md in every repo that simply lists the reading order and points to AGENTS.md:

Read these files before taking any action in this repository:

1. `AGENTS.md` — behavioral rules and operating instructions
2. `rules/repo_rules.md` — binding structural constraints
3. `DEPLOYMENT.md` — deploy process
4. `.ai_context.md` — supplemental context

This works but requires maintaining an extra file in every repository.