Digest / March 4, 2026
Codex Issues Digest — March 4, 2026
40 new issues filed ·
23 resolved in openai/codex (UTC day).
New issues (40)
- Fine grained context per task instead of per project.
- Add AGENT=codex environment variable to UNIFIED_EXEC_ENV
- Codex IDE has been stuck in thinking
- [VS Code extension][Windows 11] Codex is totally disregarding my constraints and doing its own thing
- Not loading env vars from config.toml
- Add C#/.NET support to codex-universal image
- Codex API latency is ~2x slower in OAuth harness vs Codex CLI for same task/model
- chat is always working or thinking
- Extension Thinking Processes Colapsed by Default
- Output history corruption when scrolling (Windows 10 / PowerShell)
- Need a "take over" feature when users need to edit commands or input password and then back to agent's workflow
- Multi account
- Auto-archive chats after configurable inactivity
- Codex task becomes stuck if VS Code is closed while the agent is running
- codex-tui.log
- App becomes unresponsive with no error UI when config.toml has duplicate [mcp_servers.*] blocks
- Composer placeholder: cannot disable + suggestions are not task-aware
- Windsurf extension: Codex stuck in Thinking in native mode
- Keep thread name in sync across CLI and app
- Cannot change model after plan is generated in Planning Mode (model selector overlaps “Implement plan” options)
- Initiating an Edit Triggers Start-to-End Chat/Session (Auto) Scroll-Down (Too Protracted)
- Windows regression in Codex CLI — version 0.105+ crashes with HW capability error (0x20000000)
- codex model really bad too read. so many code in one file. so many varaince. no consistent solution
- Session hangs silently after successful tool_call_output (no final_answer/task_complete)
- Excessive approval prompts after recent changes for Playwright MCP
- CLI feature request: add `codex openclaw install`
- Codex CLI: notification when plan mode is waiting for user answers
- Codex TUI stalled after successful file creation
- of type 'hidden_message' was provided without its required 'reasoning' item:
- Regression Bug Report: Plan Mode Rich Patch View Missing
- Codex won't start after upgrading to 0.107.0
- Hover over on slash commands shows "Slash command" instead of an explanation of the individual command
- 'Ctrl' is denoted as '^' in settings
- Codex app doesn't open in WSL with bash
- Sub-Agent Timeout
- Forked Worker Inherits Parent User Intent and Misinterprets It as Direct Instruction (Recursive Delegation Attempt)
- Windows execpolicy + PowerShell wrapping: safe delete rules are either bypassable or too noisy
- Attaching an image with large file size can nuke your thread and freeze the app.
- Proposal: experimental codex multiverse command for parallel solution evaluation
- Codex deleted my entire Drive
Resolved issues (23)
- Configurable App Server WebSocket port/endpoint to connect from VS Code to a Codex instance running inside a Dev Container (one container per worktree)
- probleme de connexion codex avec chatgpt plus
- Codex IDE has been stuck in thinking
- [VS Code extension][Windows 11] Codex is totally disregarding my constraints and doing its own thing
- chat is always working or thinking
- Multi account
- Output history corruption when scrolling (Windows 10 / PowerShell)
- Forked Subagents Using request_user_input
- Cannot change model after plan is generated in Planning Mode (model selector overlaps “Implement plan” options)
- CLI feature request: add `codex openclaw install`
- codex model really bad too read. so many code in one file. so many varaince. no consistent solution
- Windows regression in Codex CLI — version 0.105+ crashes with HW capability error (0x20000000)
- Codex fails to start on Windows with CPU capability error (requested: 0x20000000)
- Codex extension will not display reasoning chain unless you scroll up and back down in chat
- Session hangs silently after successful tool_call_output (no final_answer/task_complete)
- VScode插件Codex在wsl中Ubuntu系统超时:Reconnecting，stream disconnected before completion: error sending request for url (https://chatgpt.
- Codex API latency is ~2x slower in OAuth harness vs Codex CLI for same task/model
- Windows Codex App Support
- Extension Thinking Processes Colapsed by Default
- Possible Codex usage metering anomaly on Plus (very small tasks consuming large 5h + weekly quota)
- Drain Faster 5 hour usage limit Weekly usage limit
- Using subagents exponentially explodes "messages" usage
- Codex CLI: notification when plan mode is waiting for user answers