Digest / September 24, 2025
Codex Issues Digest — September 24, 2025
38 new issues filed ·
8 resolved in openai/codex (UTC day).
New issues (38)
- Unexpectedly hitting rate limits in IDE
- [status] Account plan displays "Free" despite active Plus subscription
- “stream disconnected before completion” + “input exceeds context window” loop
- gpt-5-codex unable to use MCP servers
- Subject: Missing “Analysis / Plan” Button Without Code Generation in Codex
- Add a `codex upgrade` command for easier updates, or implement automatic version updates
- auto compact error
- Codex Sometimes Breaks Fish Key Bindings
- PowerShell shows and writes Turkish characters incorrectly (encoding issue)
- Codex VS Code extension injects empty image_url blocks into some chats, causing permanent 400 errors.
- Weekly usage on pro
- gpt-5-codex via litellm attempts to use v1/chat/completions endpoint
- config `max_output_tokens` does not work
- Codex CLI doesn't know about the MCP in conversation
- Search Tag for Codex Exec
- Ctrl-C should not exit cli (linux)
- How to define multiple models for the same model_providers in config.toml
- feature request: make ctrl + shift + c default keybind for opening codex in vscode
- The "Ask" and "Code" buttons have disappeared, replaced by a single button now.
- codex debug seatbelt ignores sandbox_mode in ~/.codex/config.toml
- Codex CLI ignores "read-only" mode and approval policy by executing MCP edit tools
- Copy-Paste in turkish
- 60s default timeout for MCP is a considerable regression for many types of servers
- Migration from API key to ChatGPT login breaks codex with config error (os error 2)
- Stream retry does not honor “Try again in x seconds” delays; falls back to exponential backoff
- Named sessions for --resume
- MCP tools silently excluded due to errors
- Headless supervise-and-repair for Codex CLI: live log streaming + Playwright MCP (auto-fix, resume, zero-error stability)
- MCP servers fail to start on Windows due to missing inherited environment variables
- BUG: --output-schema limited to "gpt-5" as guard is too narrow on model_family.family
- azure responses gpt-5-codex 400 The encrypted content for item rs_68dxxx could not be verified
- Increase non interactive Mode Timeouts
- Search does not function, flat-out.
- VS code ide websearch tool non functioning.
- Address annoying NPM warning about http-proxy
- bad system call termux
- SIGTERM when calling from from pid 1
- Usage limits halved
Resolved issues (8)
- Scroll bar for Codex VS Code extension
- gpt-5-codex unable to use MCP servers
- Fall back to read 'reasoning_content' when 'reasoning' to support VLLM-served gpt-oss
- Search Tag for Codex Exec
- Codex Rust: now way to provide api version for azure provider
- Install failed
- TUI: "The application panicked (crashed)" + "byte index 64 is not a char boundary; it is inside '\u{202f}' (bytes 63..66)" while composing prompt
- BUG: --output-schema limited to "gpt-5" as guard is too narrow on model_family.family