You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I've been using a "hacky" way to use this framework with codex.
Essentially, it just involves following the general setup instructions using claude. After the files have been created in your repo, copy the files from .claude/commands to your codex prompts directory (probably ~/.codex/prompts).
This works well enough, but has some sharp edges. The agent will not update the context file this way, and some prompts won't work as expected (When running /specify, I always need to give it the description twice).
Official support will be nice, but the above should get you going if you only utilize codex.
Reacted by Joshua Fuller, Ruslan Dzherikhov, Bob Bergman, Atsuki Nohira, Nolan Blew, Loknath Behera, isvicy, Bad3r, Vyacheslav, Seongju Mun and 4 more
Can't wait for the support. Meanwhile I was thinking to use Claude Code, but let it use Codex in the background using codex --dangerously-bypass-approvals-and-sandbox -m gpt-5 -c model_reasoning_effort="high" exec <query>
I guess I need to first improve claude.md to let it properly offload spec-kit tasks to codex. Any suggestions?
It would be nice if this also supported codex, especially given how GitHub and Open ai basically belong to the same company