A Reddit post details how Claude Code (since v2.1.277, September 18) reads AGENTS.md as project instructions when no CLAUDE.md exists, letting repos set up for Codex or Cursor work unchanged. The file-precedence rules fail silently: adding a personal CLAUDE.local.md makes Claude Code stop reading AGENTS.md with no error. The author recommends /config's 'claude-md-and-agents-md' option or a CLAUDE.md importing AGENTS.md to get both loaded.
A developer found that Claude Code stops loading a repo's AGENTS.md once any CLAUDE.md, .claude/CLAUDE.md, or CLAUDE.local.md is present, per Claude Code's own memory docs. The workaround is an @AGENTS.md import on line 1 of CLAUDE.md, then verifying with /memory or /context all. The post offers guidance on structuring shared agent instructions across Cursor, Claude Code, and Codex.
A developer found that Claude Code stops loading a repo's AGENTS.md once any CLAUDE.md, .claude/CLAUDE.md, or CLAUDE.local.md is present, per Claude Code's own memory docs. The workaround is an @AGENTS.md import on line 1 of CLAUDE.md, then verifying with /memory or /context all. The post offers guidance on structuring shared agent instructions across Cursor, Claude Code, and Codex.
Practitioners on Reddit discuss how to maintain one set of instructions, skills, and MCP servers across AI coding agents like Claude Code, Codex, OpenCode, Cursor, and Copilot. Recurring pain points include differing config formats (Cursor .mdc rules vs CLAUDE.md vs AGENTS.md) and uneven loading behavior. Claude Code can now point CLAUDE.md at AGENTS.md, which eases cross-tool reuse but doesn't fully resolve the fragmentation.
Practitioners on Reddit discuss how to maintain one set of instructions, skills, and MCP servers across AI coding agents like Claude Code, Codex, OpenCode, Cursor, and Copilot. Recurring pain points include differing config formats (Cursor .mdc rules vs CLAUDE.md vs AGENTS.md) and uneven loading behavior. Claude Code can now point CLAUDE.md at AGENTS.md, which eases cross-tool reuse but doesn't fully resolve the fragmentation.
A user tested Claude Code 2.1.312 and confirmed that when a repo contains both AGENTS.md and CLAUDE.md, the tool loads only CLAUDE.md by default. The post documents workarounds and edge cases from Anthropic's memory docs that teams relying on AGENTS.md may miss.
A user tested Claude Code 2.1.312 and confirmed that when a repo contains both AGENTS.md and CLAUDE.md, the tool loads only CLAUDE.md by default. The post documents workarounds and edge cases from Anthropic's memory docs that teams relying on AGENTS.md may miss.
An AI engineer building agentic systems shared a proposed repository structure meant to standardize how coding agents like Claude Code, Cursor, and Codex work in one codebase. It combines AGENTS.md and CLAUDE.md files, Claude Code hooks/rules/skills, per-feature specs, and review guidelines, drawing on official docs from Anthropic and OpenAI. The engineer solicited community feedback on alternatives, calling the draft a reasonable starting point rather than settled best practice.
An AI engineer building agentic systems shared a proposed repository structure meant to standardize how coding agents like Claude Code, Cursor, and Codex work in one codebase. It combines AGENTS.md and CLAUDE.md files, Claude Code hooks/rules/skills, per-feature specs, and review guidelines, drawing on official docs from Anthropic and OpenAI. The engineer solicited community feedback on alternatives, calling the draft a reasonable starting point rather than settled best practice.
A GitHub project, agents-contract, packages a single-file AGENTS.md contract for AI coding agents together with a Python linter that validates it and a SessionStart hook that checks the environment before work. It targets Claude Code, Codex, Cursor, and GitHub Copilot, aiming to reduce agent guesswork and broken builds. Claude Code support works via a CLAUDE.md that imports @AGENTS.md.
A GitHub project, agents-contract, packages a single-file AGENTS.md contract for AI coding agents together with a Python linter that validates it and a SessionStart hook that checks the environment before work. It targets Claude Code, Codex, Cursor, and GitHub Copilot, aiming to reduce agent guesswork and broken builds. Claude Code support works via a CLAUDE.md that imports @AGENTS.md.