Codex can import supported setup from other coding agents without changing or deleting the original. That makes import a low-risk way to test continuity rather than a migration commitment.
Bring selected Kiro or Claude conventions into Codex, then review the translation before relying on it.
| Existing item | Codex destination |
|---|---|
| Instruction files | AGENTS.md |
| Settings | config.toml |
| Skills and slash commands | Codex skills |
| MCP, hooks, subagents | Equivalent Codex configuration |
| Recent sessions | Codex threads |
Hooks, command interpolation, permissions, and connection authentication can behave differently. Imported means translated, not verified.
Import items in an order that makes failures easy to diagnose. Static guidance is easier to inspect than executable hooks or authenticated connections.
AGENTS.md with the source guidance.| Question | Evidence |
|---|---|
| Did guidance land at the correct scope? | Global versus repository file location |
| Can any imported tool mutate external state? | Tool list and approval policy |
| Are credentials still available safely? | Environment-variable names, never secret values |
| Do command placeholders still expand correctly? | Harmless dry run with sample arguments |
The original agent setup remains untouched. If an imported item behaves badly, disable or remove that Codex copy instead of changing the working Kiro or Claude setup.
Open the import preview and record which three items would provide immediate value. Rank them by execution risk, identify each Codex destination, and write one harmless verification task for each item before completing the import.
Import to Codex documents supported items, destinations, and post-import checks.