What problem does it solve? Keeping project instructions, skills, subagents, and hooks consistent across both Claude Code and Codex is error-prone when each tool has its own native configuration format. This Skill defines a single shared layout under .agents/ so rules are written once and consumed by both harnesses without duplication or drift. ## Core Features & Use Cases - Shared instruction layout: Defines ownership rules for root and package AGENTS.md files, sibling CLAUDE.md symlinks, Serena memories, and shared skills discovered by both tools. - Compatibility rules: Documents how Codex reads AGENTS.md and inline TOML hooks while Claude Code reads CLAUDE.md links and .claude/settings.json, including subagent definitions via agents/claude.md and agents/codex.toml. - Verification guidance: Explains the agent-compat commit hook and PR lint workflow that validate symlink targets, config_file paths, and link resolution without running agents. - Use Case: When adding a new repository skill, you create it once under .agents/skills/, optionally add harness sidecars in its agents/ directory, and both Claude Code and Codex pick it up with no duplicated content. ## Quick Start Ask the agent to add a new shared skill or update the root AGENTS.md following the agent-files layout rules so both Claude Code and Codex load it correctly.
