This action needs serious investment — it's falling far behind claude-code-action

[ChrisWiles]

The numbers speak for themselves

I'm a paying OpenAI customer actively using openai/codex-action in production workflows. I want to keep using Codex. But the state of this action compared to Anthropic's equivalent makes it increasingly hard to justify.

Side-by-side comparison

Metric	openai/codex-action	anthropics/claude-code-action
Stars	786	5,502
Commits	38	448
Contributors	5	71
Releases	0 (no tagged releases)	30
Last commit	Nov 19, 2025 (~3 months ago)	Today (Feb 5, 2026)
Open issues	7	318 (active community)

What this means in practice

• No releases. We're pinning @v1 and hoping for the best. Claude Code Action has 30 tagged releases with changelogs.
• 38 commits total. Claude Code Action has 448. That's nearly 12x the development velocity.
• 5 contributors. Claude Code Action has 71. The community is building with them, not here.
• Last commit was November 2025. That's almost 3 months of silence. Claude Code Action ships updates daily.
• GPT-5.3-Codex just launched today and this action's hardcoded default model is still gpt-5.2-codex. We have to manually specify model: gpt-5.3-codex because the action hasn't been updated.

What we need

1. Active maintenance — Regular commits, releases, changelogs
2. Feature parity — Progress tracking, structured outputs, better integration with GitHub's PR/review APIs
3. Community engagement — Respond to issues, accept PRs, build in the open
4. Same-day model support — When a new model drops, the action should support it out of the box

The ask

Sam, please put a full team on this. The Codex CLI and cloud product are incredible — but the GitHub Action is the bridge between Codex and every CI/CD pipeline in the world. It deserves the same love.

Please don't force me to switch to Anthropic. I want to stay, but you have to meet us halfway.

[umm0n]

It's actually remakable how little attention is being given to this action. Not a good look OpenAI.

[adamhenson]

Building production tooling on top of both actions for the past few months. The two gaps I've hit hardest are token outputs and progress visibility.

On tokens: anthropics/claude-code-action exposes input_tokens, output_tokens, cache_read_tokens, cache_write_tokens, and turns as step outputs, so downstream steps can gate on cost or log it without any extra parsing. Codex's equivalent data exists - the rollout JSONL in $CODEX_HOME/sessions/ has it - it just isn't surfaced. PR #83 is an attempt to fix this the same way claude-code-action does it.

On progress: there's no native mechanism to post interim comments or update a PR while the run is in progress. claude-code-action's use_sticky_comment handles this out of the box. For long-running Codex runs in PR workflows it's a noticeable UX gap.

Both are solvable without a rewrite - the data is already there.