Commit attribution: optionally supply the model name instead of the generic Codex name

Open 💬 2 comments Opened Mar 17, 2026 by dgunay

What variant of Codex are you using?

CLI

What feature would you like to see?

The git attribution/Co-authored-by trailer feature only gives Codex as the name of the co-author.

I like to know which model produced the code as there are huge differences between generations and sizes of LLMs - this is something Claude Code does as well. Example:

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

I imagine Codex could either template the model name for use in the custom attribution message like this:

commit_attribution: "{{MODEL}} <email@example.com>"

Or could add another config parameter like commit_attribution_use_model: true/false.

Additional information

_No response_

View original on GitHub ↗

2 Comments

inickt · 4 months ago

Also interested in this, our org is trying to get some better observability on which agents/models are being used for observability, and having the model set would be super useful.

franciscojavierarceo · 2 months ago

Hey folks, I maintain a few open source projects and I really value git attribution because it's transparent when I receive changes from unknown authors. Would really love to see this supported.

I have a working implementation that restores the Co-authored-by trailer feature as opt-in, addressing this issue along with #18095, #14051, #23007, #938, and #19799.

Branch: franciscojavierarceo/codex@feat/git-commit-attribution

What it does:

  • Opt-in via [features] codex_git_commit = true in config.toml
  • Default trailer dynamically includes the model name: Co-authored-by: Codex o4-mini <noreply@openai.com>
  • Customizable via the existing commit_attribution config key (which was available before the feature was removed in d18a7c982): commit_attribution = "MyOrg Codex <codex@myorg.com>"
  • Set commit_attribution = "" to disable while keeping the feature flag on
  • Uses noreply@openai.com by default (avoids the codex@example.com misattribution issue from #18095)
  • 10 unit + integration tests covering all config combinations

Happy to open a PR if the team is interested.