An opinionated Codex CLI. This is strictly a personal hobby project, forked from openai/codex.

Build from source:

cd codex-rs cargo run --bin codex

We will never merge code from the upstream repo; instead, we re-implement our changes on top of the latest upstream code.

Iteration flow (aligned with .agents/skills/codex-upstream-reapply):

flowchart TD A[Freeze old branch: commit changes + intent docs] --> B[Fetch upstream tags] B --> C[Pick tag + create new branch from tag] C --> D[Generate reimplementation bundle] D --> E[Read old branch + bundle for intent] E --> F[Re-implement changes on new branch] F --> G[Sanity check diffs vs tag] G --> H[Force-push to fork main] 

• DX Focused: Focus strictly on optimizing developer experience, without adding new features.
• Upstream Sync: We sync with the upstream repository regularly.

• Added a TUI status header with model/effort, cwd, git summary, and rate-limit status.
• Collaboration mode presets now accept per-mode overrides and default to the active /model settings.
• TUI watches auth.json for external login changes and reloads auth automatically (with a warning on account switch).
• AGENTS.md/project-doc instructions are refreshed on each new user turn, and Codex shows an explicit warning when a refresh is applied.
• Full change log: see CHANGED.md.

1. DO NOT update any test codes
2. After making code changes, verify the CLI still launches:

cd ./codex-rs cargo run --bin codex

# config.toml [collaboration_modes.plan] model = "gpt-5.4" reasoning_effort = "xhigh" [collaboration_modes.code] model = "gpt-5.4"

When syncing to the latest upstream codex version, use .agents/skills/codex-upstream-reapply to re-implement our custom requirements on top of the newest code, avoiding merge conflicts from the old branch history.

Example:

$codex-upstream-reapply old_branch feat/rust-v0.94.0, new origin tag: rust-v0.98.0 

Status bar design reference: https://linux.do/t/topic/1481797