Digest / August 7, 2025
Codex Issues Digest — August 7, 2025
17 new issues filed ·
262 resolved in openai/codex (UTC day).
New issues (17)
- Closing and LOCKING stale PRs via bot, which is bad UX
- finish with InternalAgentDied
- Support for ChatGPT account
- Logging in with ChatGPT failing
- codex exec failing with & without update_plan
- Insufficient quota if you have an organisation created because it always wants to connect to your organisation, even if you want to use the ChatGPT plus/pro subscription instead
- Codex CLI consumes OpenAI platform credits despite authenticating with ChatGPT Plus account
- Codex just not working at all.
- Unable to Authenticate with ChatGPT account
- stream disconnected before completion: stream closed before response.completed
- When running models from openrouter providers the model messages are not displayed
- codex-ts fails at markdown / codex-rs doesn't support ChatGPT login?
- GPT-5 HIGH Reasoning via ChatGPT
- MCP server config error
- proposal: Standardized Location & Format for Agentic Toolkit Documentation (agents/*.md)
- Enable Signing in with ChatGPT account for browserless machines
- Can't available Codex CLI with Team Plan subscription
Resolved issues (262)
- codex-cli is crashing in software development prompt
- Codex CLI Overwrites Existing AGENTS.md Due to Missing Enforcement Layer
- Closing and LOCKING stale PRs via bot, which is bad UX
- [Bug] Bad interaction with GPG `pinentry`.
- Bug: Quota exhausted reading in symbols from DerivedData
- Certain Large Models Fail
- Rate limit causes process to abort
- bug: resizing terminal leads to bad artifacts
- bug: agent sometimes forgets he has tools
- unify file‐globbing under CODEX_SEARCH_CMD
- [Bug] Leading `/` stripped from absolute path, followed by `ENOENT` crash due to attempt to access non-existent file.
- on arch host, codex thinks it is in debian sandbox
- `/clear – clear screen & context` does not clear the screen
- Codex not able to write in files in full auto mode
- Codex hang using ~99% CPU after thinking for a few seconds on long conversation
- Model assumes Windows PowerShell is running on WSL resulting in seeing an "empty directory"
- applyPatch function doesn't handle creating directories properly
- Non o4-mini models all fail
- bug: reproducible ENOENT for missing image
- bug: invalid _request error when exiting the tool via user interruption (2x ESC).
- bug: Output truncated: too many lines or bytes
- bug: add readline to support history, tab completion, and key bindings
- Schema error in 0.1.2504220136
- config.yaml not respected (approvalMode ignored), error spawn ENOTDIR and ENAMETOOLONG
- bug: non-openai model provider stops responding after a few turns
- bug: non-openai model is not streaming output
- Codex hangs
- bug: codex "cannot parse json" and "cannot iterate over a consumed stream" errors
- bug: server error after a long session.
- Invalid Context error from patching
- Codex cannot find things when running on WSL2
- File Write Operations Not Reflected in Codex CLI Environment
- Feature request: automatic back-off + prompt/token optimiser to avoid rate-limit exits
- o4-mini: unusually high token count (71 k) for moderate-sized prompt – guidance?
- Models not using apply patch or other cli tools correctly then crashing (Gemini)
- bug: codex doesn't answer anymore after asking to fix/improve a code (Gemini)
- Codex not see settings file
- Codex breaks down in a python SDK to typescript conversion
- apply_patch bug is occur when the patch has long context
- Error occurred while processing the request
- Process, spend money and nothing is outputed
- run_in_container.sh forces --full-auto and discards user flags for approval-mode
- esc esc to "cancel" completely wipes chat history
- Unexpected error when writing a .md
- Failed to call a function. Please adjust your prompt. See 'failed_generation' for more details.
- `Failed to parse JSON result` endless loop
- Status: 400 No tool output found (breaks conversation)
- No file saves using DeepSeek
- Multiple thinking processes within terminal
- Message box isn't displayed in the terminal
- Error: ENOENT: no such file or directory, open "Error: ENOENT: no such file or directory, open 'C:\Users\willm\AppData\Local\Temp\oai-codex\codex-cli-2025-05-09T15:47:02.log'
- OpenAI rejected the request. Error details: Status: 400, Code: unknown, Type: unknown, Message: 400 status code (no body)
- ERROR crypto is not defined
- More elaborate error description
- gpt4.1, gpt-4o not actually applying patches to code. Only o3 and o4 models actually appear to make or apply any changes
- command completions broken
- Show/hide menu erases/shifts output
- Bug or Question clear text in
- Receive error anytime Codex tries to suggest a shell tool call: `Cannot read properties of undefined (reading 'length')`
- sandbox required to us eth ecli
- Openrouter Openai/codex-mini not support
- Cannot use codex-mini
- Bug application crashes
- TimeoutOverflowWarning
- The log file name should not use the colon character
- The `--image` seems not working
- Asked to use `openai` CLI commands instead of `codex`
- ⚠️ OpenAI rejected the request (request ID: req_852561adbc27e3cb768be3dce1b477d9). Error details: Status: 400, Code: unknown, Type: invalid_request_error, Message: 400 No tool call found for function call output with call_id call_xUWMD9ZjBX4CyOntda7j5geU..
- Allow setting default model per provider
- Context does not seem to update effectively using codex-mini-latest
- Codex CLI can not edit file while VS Code is open (WSL)
- "explain" feature non-functional when using `codex-mini-latest` model
- Cannot execute commands in Windows 10 Pro with legacy Powershell
- Using codex CLI on Linux desktop to connect a LiteLLM/Ollama server, both on a the same LAN, can it work?
- Azure OpenAI: TypeError: Cannot use 'in' operator to search for 'tool_calls' in undefined
- 404 on `--image` with codex-mini-latest
- OpenAI API Key Authentication Fails when OpenRouter as provider
- codex --history display incorrect tool calls for codex-mini
- Network Error
- "Rate limit reached" crashes the codex instance instead of sleeping/waiting
- Paths with parentheses often not escaped
- Codex model stuck in thinking after success
- Codex with V0 unauthorized
- "Please pass patch text through stdin" Error When Using -q Parameter
- codex --model accepts any model name and codex will report it is using it
- Quiet mode doesn't seem to work for me. At least not with provider set to azure.
- Deprecated node-domexception@1.0.0: Use your platform's native DOMException instead
- blank output with workaround forced providers
- Keeps getting 401 issue when using Azure instance even everything is being passed correctly
- Terminal output isn't echoed properly
- Codex re-prints the thinking block multiple times if the terminal window is resized
- I used Codex for coding tasks and it displayed normally, but I couldn't find the code in my local directory
- Server error - falls flat
- ChatGPT plus user can use Codex but not CLI?
- Invalid URL (POST /v1/engines/codex-mini-latest/completions)
- OpenAI rejected the request
- Improve Markdown rendering
- Support the `/model` command
- Support the `/bug` command
- Provide suggested prompts
- In environments where username not in /etc/passwd codex errors out
- Commands not executed
- Asks for ChatGPT key when `-p azure`
- OpenAI rejected the request
- sandboxing? no files seem to be written
- Can't explain empty command on default settings
- Bug on codex-cli/tests/apply-patch.test.ts L328
- 401 Unauthorized from v0.dev API when using Codex CLI (`--provider vercel`)
- Quiet mode (-q) not truly non-interactive due to git warning prompt
- "│Unable to generate explanation: 404 This model is only supported in v1/responses and not in v1/chat/completions."
- escape does not work to exit help menu
- 400 Errors when attempting to use xAI
- Azure OpenAI integration: 401 and 429 errors with o4-mini deployment (valid key/endpoint)
- Error status code 400 when trying to use MistralAI Devstral
- Warning: model "codex --version 0.1.2505172129" is not in the list of available models for provider "azure"
- Configuration per documentation does not work. Connecting to Azure AI Foundry OpenAI models
- codex-rs: codex cannot obtain error information when receiving non-zero exit codes
- Cannot use Azure as a Provider post #1324 fix
- No API key configured for provider
- wrong files created and broken syntax error
- Crashed on first prompt to run unit tests
- BUG: Codex does not look at project type, assumes python project
- codex cli crashes on rate_limit_exceeded
- Request for understanding, updating and search failed.
- codex will silently ignore malformed config.json instead of alerting that its malformed
- init_firewall.sh fail at container start
- Support for local / other LLMs
- API Key Security Feature Requests
- Openrouter api support
- npm warns about deprecated packages
- CLI crashes with “server_error: The server had an error while processing your request” when context window is nearly exhausted
- apply_patch aborted
- Add ability to hide sensitive files from agent
- handle node_modules dir correctly
- vim-like input mode
- "Previous response cannot be used for this organization due to Zero Data Retention"
- bug: unable to continue after hitting esc esc to interrupt assistant
- feat: don't close program when out of credits or when API key is missing
- Implement /cost command and track session usage
- Feat: Support streaming all outputs and also support passing history with -q
- bug: No such file or directory
- Adhere to the macOS File System Programming Guide on macOS
- feat: session persistence
- Can you add MacOS alerts when it's the user's turn?
- Sandbox was mandated, but no sandbox is available!
- can't see user prompt or much of response from LLM
- relax node 22 or newer requirement
- Support Windows
- [Feature request] Rules
- [Feature request] More structure for diffs
- Azure OpenAI Python SDK works, but codex CLI fails with 404
- Keep thinking
- git fails to access tmp directory
- feat: make it work with large codebases
- Error when Codex has to install libraries
- Setup information missing from documentation?
- no o3 model in list
- [Feature Request] .codexignore file
- How to set web proxy for the GPT api inside codex?
- option - backspace (to delete the previous word) actually deletes all text in the prompt box
- OpenAI rejected the request
- WIN/PS - Insists on using sh - assumes linux/mac - assumes chat wants to tool
- feat: Add the ability to run `/compact`
- Home/End keys not working in Codex on Windows Terminal with WSL
- (feat) create log file with all actions executed
- Instead of editing my files, it deleted them
- error: can't create new files
- Unclear Running Instruction - Was not able to use.
- Add ability to interrupt a request without exiting the process completely
- feat: grounding in documentation
- feat: lessons
- Request for --semi-quiet option to allow user interaction while showing normal output
- Modernize Monorepo Structure with pnpm Workspaces and Turborepo
- Bug: Read-only sed and grep commands should not ask for permission to run
- feat: enable api/llm model to converse in between each step
- feat: multimodal inputs in REPL mode
- Create releases for an easier diff between versions
- Can touch files and dirs, and run npm commands but will not write to files - macOS
- Unexpected token import
- MaxListenersExceededWarning
- Error: ENOENT: no such file or directory, lstat '/Users/{$profile}/.pyenv'
- feat: verbose output mode
- feat: allow sandbox to read/write some files/directories anywhere on the machine
- feat: add codex.md in this repository
- Need CLA‑assistant unlock for PR #348 (Windows support)
- feat: Add a voice input mode
- allow config per repo
- GPT-4.1 for quick answers
- feat: /flex-mode command
- network connection error
- feat: Print /bug info to terminal if opening a browser is not available.
- Build failure due to missing package-lock.json using provided nix flake
- Request - Add Support For Github Copilot
- feat: model registry
- feat: support for dedicated configuration file for API keys
- bug: quiet mode still printing json formatted text to terminal
- [Error]: 400 No tool output found for function
- PATH is overwritten inside of the sandbox
- bug: hangs for groq provider
- bug: codex doesn't recover from loss of connectivity
- Environment variables for custom providers are not getting picked up
- Azure Open AI
- feat: word-based text navigation and deletion
- Codex CLI exits abruptly on `rate_limit_exceeded`; should back-off & retry instead of closing
- Shell cmds not working
- Feature Request: Add /cost and /doctor commands for usage stats and health checks
- OpenAI rejected the request
- docs: duplicate -q --json call in example cli reference
- Support for ChatGPT account
- Supporting Azure OpenAI
- feat: add a dev changelog
- Cut and Paste from Terminal
- Codex Incredibly Slow - How to Debug?
- finish with InternalAgentDied
- readme points to wrong file for personal global guidance
- [Feature request] Add (codex) to author in commit message and as "codex ${model}" as Co-author # Aider feature parity / convention
- Hang / Freeze / No thinking
- `history` doesn't work in headless mode
- Feature request: C-x C-e to open the current prompt inside the editor [TUI parity with Unix tools and Aider TUI]
- Codex ignores its system prompt after the first message
- please warn users that the values on config.json are meant to be the enviroment variable names themeselves
- Azure OpenAI Integration is not working
- Codex faills with sandbox error trying to execute python file
- Add logging input and output of LLM requests
- Feature Request: Add /save command to store sessions for later review
- Non-reasoning OAI model API calls have an extraneous reasoning_effort parameter
- codex fails on sandbox error
- Support TODO/task management ability
- Show token counts similar to `claude-code`
- rate_limit_exceeded Version 0.1.2504301751
- codex ends without doing something
- failed to run codex on a remote server
- feat(run_in_container): optionally mount `~/.codex` so Linux users keep configs & logs
- Codex CLI v0.1.x fails with "Sandbox was mandated, but no sandbox is available!" on Fedora with Docker installed
- Add /init slash command to bootstrap repository context for sessions
- Minor Enhancement Request: Would be nice to add a model_name or alias attribute to model config
- [TypeScript] Unable to skip sandboxing on platforms Codex considers supported
- Azure Open AI not supported(?)
- Cannot redeem credits when using log with ChatGPT (PRO)
- Can't use other provider model
- Allow sign in via link instead of trying to use XDG to open a system browser
- Codex mini stuck thinking
- Problem with logic when it comes to change providers
- An error occurred while processing your request
- When using another provider, setting the OPENAI_API_KEY should not be necessary.
- Ubuntu 24
- ⚠️ Connection closed prematurely while waiting for the model. Please try again.
- (node:62928) TimeoutOverflowWarning: 1200000000000 does not fit into a 32-bit signed integer. Timeout duration was set to 1.
- Sign in with ChatGPT does not work (windows wsl)
- Feature Request: Allow switching model mid-session
- codex-mini-latest thinks for a long time then eventually times out on WSL
- WSL: No solution to "Error: The combination of seccomp/landlock that Codex uses for sandboxing is not supported in this environment."?
- When using a Deepseek provider, the tool sends a request to the OpenAI API and results in a 401 authentication error
- Pasting secret values pastes it to key instead
- When reading a file using sed or tail, it takes output from the terminal as content of the file
- Display the current session's cumulative token usage (input/output)
- Fails to connect to llm proxy in secured network.
- bug: Unable to continue due to rejections of calls with tools
- Feature: Entra authentication for Azure OpenAI provider
- Codex does not respect API keys specified through environment variables for other vendors
- Logging in with ChatGPT failing
- `config.toml` missing/not recognized