Codex repository size increased a lot due to accidental binaries commited recently

What issue are you seeing?

Current Codex repo is 340MiB when fully cloning it, due to accidentally commited (and then removed) binaries that are still in Git history. I hope you fix this (prune those useless binaries from the repo completely), it'll make it much more convenient to clone. Thanks!

Relevant PRs:

• Added binaries in https://github.com/openai/codex/pull/13953
• Removed binaries in https://github.com/openai/codex/pull/14232

The main issue is that the binaries are only removed from the current tree, they're still in the git history.

Summarized by Codex itself:

===CODEXSTART===

The checkout is 471M, and .git alone is 425M. git gc --auto did not reduce it, so this is not loose-object clutter; it’s real packed history.

HEAD is small. The only tracked file over 1 MiB right now is MODULE.bazel.lock at about 1.18M. So the size problem is historical baggage, not the current tree.

Worst things in history

The main offender is a batch of checked-in platform binaries under sdk/python/src/codex_app_server/bin/*.

• Added on 2026-03-10 in commit 6baeec68b (python-sdk: generated type foundation...)
• Removed later the same day in 9b3332e62 (fix(python-sdk): stop checking in codex binaries...)
• Biggest blobs:
• sdk/python/src/codex_app_server/bin/windows-x64/codex.exe 100.6M
• sdk/python/src/codex_app_server/bin/darwin-arm64/codex 84.9M
• sdk/python/src/codex_app_server/bin/windows-arm64/codex.exe 84.7M
• sdk/python/src/codex_app_server/bin/linux-x64/codex 83.1M
• sdk/python/src/codex_app_server/bin/darwin-x64/codex 76.5M
• sdk/python/src/codex_app_server/bin/linux-arm64/codex 71.4M

Those seven SDK runtime blobs total about 568M uncompressed and roughly 227M of the packed repo by themselves. That is the primary reason the repo is large.

Other notable historical junk:

• codex-rs/codex historical checked-in binary: 25.3M and 18.0M
• .github/demo.gif: 19.8M
• reviewers/bolinfest.json: two versions around 5.8M each
• codex-rs/tui/tests/fixtures/binary-size-log.jsonl: several versions around 2.9M to 3.5M
• codex-rs/tui/tests/fixtures/timeout-session-log.jsonl: 2.9M
• codex-rs/tui/src/security_report_assets/mermaid.min.js: 3.3M
• .github/codex-cli-splash.png: 3.2M
• .github/codex-cli-login.png: 3.1M

Bottom line

This repo is big because large binaries and media were committed into normal Git history. The worst event was the Python SDK runtime binaries added on March 10, 2026. Current files are not the issue.

===CODEXEND===

What steps can reproduce the bug?

git clone https://github.com/openai/codex downloads 341MiB for the whole repo

What is the expected behavior?

<100MiB for the whole repo, hopefully <50MiB