Stop hand-syncing Claude Code and Codex configs

#claude #openai #cli #productivity

If you've spent any time in Claude Code or Codex, your config tree is probably substantial:

~/.claude/{settings.json, agents/, skills/, .mcp.json, CLAUDE.md}
~/.codex/{config.toml, AGENTS.md} plus ~/.agents/skills/

Two situations get painful fast.

Switching tools. You curated skills, agents, MCP servers, and permissions in one. Now you want to try the other. Migrating by hand is slow and easy to get wrong, and the two hosts don't share shapes. Permissions are allow/deny/ask lists in Claude, but sandbox_mode plus web_search plus prefix_rule in Codex. Agents are YAML frontmatter in Claude, TOML fields in Codex.

Using both. Same skills, same agents, same MCP. Two trees. They drift. Adding anything means two edits, and you forget which side is canonical.

ai-config-sync-manager bridges the two. It diffs both sides, plans changes, applies them with a backup.

The translation is host-aware:

Claude tool permissions to Codex sandbox modes (Bash stays out of read-only sandboxes by default)
Agent frontmatter (YAML) to Codex agent fields (TOML), including model alias mapping
MCP servers both ways, including remote MCP with bearer-token env vars
Skills as folders, not loose files
Vocabulary mismatches that can't auto-translate get paraphrase overrides instead of silent corruption

Quick start:

npx ai-config-sync connect       # register the plugin in both hosts
npx ai-config-sync status        # see what's out of sync
npx ai-config-sync sync          # dry-run plan
npx ai-config-sync sync --apply  # apply with backup