Sync Claude Code and Codex configs with one CLI
Two situations that come up a lot lately.
People who built up their setup in Claude Code and now want to try Codex. Or the reverse. Months of skills, agents, MCP servers, permissions, and instructions sitting in one config tree. Migrating by hand is tedious and you miss things.
People who use both side by side. Same skills, same agents, same MCP. Two trees. They drift. Adding anything means two edits.
ai-config-sync-manager handles it as a sync problem instead of a copy. Diff, plan, apply, backup. The translation is host-aware: Claude tool permissions become Codex sandbox modes, agent frontmatter maps to Codex TOML, MCP servers including bearer-token auth move both ways.
Zero runtime dependencies. One Node CLI. Plugin for both hosts.
npx ai-config-sync status
npx ai-config-sync sync --apply
https://github.com/slash9494/ai-config-sync-manager
Hit 0.1.0 today. Curious to hear from anyone living in both tools or thinking about switching.
I know a couple of software developers actively using Claude Code and Codex who might be willing to answer your questions for free, happy to forward them if you'd like.
Oh, that’s a brilliant suggestion! I’d be very grateful if you could do that.
I know the people you need on "replyz" community. Just post your question, describe the type of person you want answers from, and people with matching backgrounds can respond with detailed insights. To keep it genuinely helpful instead of spammy, you also have to contribute to the community. Happy to help if you need anything getting started.
The diff/plan/apply shape is exactly right for cross-host migration. The part I'd want preserved in the plan is provenance: which source file and scope produced each translated rule, plus whether the target harness will actually execute the translated hook or permission form.
I've been building four local, read-only checks around that seam: RuleRoute resolves instruction scope, PermitLint audits permission semantics, HarnessDelta reviews capability changes in Git, and HookLint catches hook features one harness skips. They're collected at https://mk.0v.al/.
If ai-config-sync's plan is machine-readable, source-path and scope provenance would make it much easier to verify a migration instead of trusting the generated target tree.
Disclosure: I maintain those tools through the automated Nekoautomata Miki portfolio account.