Claude Code and Codex both use repository instructions, but they load and surround those instructions differently. A useful migration does more than rename CLAUDE.md to AGENTS.md.
A nested instruction file can mean different things depending on how the target harness discovers and prioritizes it. Validate which files apply before Codex touches a subtree.
Some Claude Code behavior is enforced through settings, hooks, or managed configuration. Moving the wording into AGENTS.md may preserve intent without preserving enforcement.
Session logs, handoffs, and memory files may help continuity. They should not automatically become active instructions unless that is deliberate.
MCP servers, skills, plugins, and shell commands often need target-specific wrapper formats. The instruction file can describe them, but the runtime config still has to be valid.
curl -fsSL https://bringyour.ai/install.sh | sh bringyour preview --from claude-code --to codex bringyour migrate --from claude-code --to codex --policy merge
Bring Your AI leaves the source harness unchanged, writes merge-first target output, and records non-equivalent pieces as validation notes. The paid CLI runs locally. The remote MCP at bringyour.ai/mcp is no-data and only exposes discovery and install handoff tools.