Digest / April 3, 2026
Codex Issues Digest — April 3, 2026
65 new issues filed ·
90 resolved in openai/codex (UTC day).
New issues (65)
- The 5-hour daily usage limit is being consumed extremely fast (in about 20–30 minutes of normal usage).
- Panic: plugin cache root resolution fails when running in a git worktree
- Output repeated
- feature/alphavantage-provider-rate-limit-handling
- Windows sandbox instability: apply_patch intermittently fails with "windows sandbox: setup refresh failed with status exit code: 1"
- Bug: Thread history data loss on restart — async persistence not flushed on shutdown (RolloutRecorder, SQLite WAL, session_index)
- Bug: UpdatedAt pagination skips valid sessions — sort key (mtime) and cursor comparison key diverge in should_skip()
- 502 Bad Gateway
- Codex CLI prompt runs indefinitely without completing
- Codex CLI update bug
- Opening Codex app causes Chrome / VS Code / Fork to hang on macOS
- Inference routing failure
- Feature request: reload-skills / reload-plugins for the current session
- Local-agent runs can resolve home/workspace paths as /home/[]/... instead of a real user path
- codex login --api-key shows misleading missing-value error instead of deprecation guidance
- multi_agent_v2 causes subagents to loop on MCP resource listing instead of completing simple tasks
- app-server websocket disconnect leaves orphaned MCP-backed threads loaded
- Closed per reporter request
- Local Plugin UI Containment Bug in Codex App
- Codex CLI: oversized tool outputs are ingested into context and can drain 5h quota disproportionately
- Native GUI / Computer Use Tools for Desktop Automation
- Allow adjustable chat content width in Codex App
- New version mcp bug
- Add global search across thread titles and conversation content
- Preserve custom thread titles in archived threads
- Startup latency regressed in 0.117.0: codex blocks on app-server bootstrap before the TUI is interactive
- Codex cannot use bundled rg.exe on Windows but tries every time
- TUI shows expired promotional tip after April 2nd
- [UI] Ctrl+F search bar overlaps with window control buttons (min/max/close), making search UI elements unclickable
- Tokens unreliable consumption
- Improve TUI to include an 'overview' panel in order to see completed and remaining tasks
- Allow editing/removing pending steer messages before they are applied
- Fix plugin installation
- Add comments in extension
- MCP tools are switching theirs positions when enabling/disabling
- MCP tool calls always cancelled in exec mode with 'user cancelled MCP tool call'
- Empty [mcp_servers] in agent role files does not clear inherited MCP servers
- Cursor jumps and flashes when deleting text in input field
- TUI freeze during agent fan-out
- Codex-side MCP discovery/display bug or limitation
- WSL resume can lower-case session cwd (/mnt/...) and make Codex switch path spelling mid-session
- stream disconnected
- VS Code extension host Code Helper (Plugin) high CPU with repeated thread-stream-state-changed warnings on macOS
- Interactive CLI hangs on multi-command / branching turns
- Feature request: per-automation sandbox/profile overrides (isolate permissions per automation)
- Codex desktop ignores configured Playwright MCP server and falls back to Chrome-launching Playwright path
- Preserve and display custom chat titles in archive
- Codex desktop can show MCP enabled while thread tool surface omits llm-tldr/serena
- **Bulk Codex Workflow Injector**
- 5h and weekly usage fluctuate wildly and diverge across sessions
- GPT 5.4 Performance degradation
- Completely unrelated and lengthy response
- Backspace handled on both key press and key release under Sway/Wayland
- feat: configurable Windows agent shell (powershell/git-bash)
- Figma plugin reads time out after 120s
- openai_base_url config and OPENAI_BASE_URL env var are ignored — always calls api.openai.com
- Codex Web Screenshots Are Not anymore available
- codex is using chrome devtools to read files
- Proposal: Use the patched crossterm and completely remove the paste logic that depends on burst.
- Codex CLI reliance on keychain for OAuth MCPs makes using
- Bring back prompts
- Archiving conversation sometimes does not remove it from sidebar
- Bug: TypeError when processing empty input array
- ApplyPatchHandler doesn't emit PreToolUse/PostToolUse hook event. Hooks only fire for Bash tool.
- Feature request: support for custom output formats
Resolved issues (90)
- Feature request: auto-plan mode with automatic recommended choices
- Plan mode: Add explaination command
- Select text in agent response to give targeted feedback (macOS app)
- Implement backoff policy when 429 would have happened
- Better notification for multi-agent
- Allow opting out of forced GIT_PAGER=cat and NO_COLOR=1 in Codex tool commands
- tui: add --no-keyboard-enhancement for terminal compatibility
- Make CLI “usage bar” less ambiguous
- tui: add codex resume --history-lines to cap replayed scrollback
- Allow list for commands instead of danger-full-access + deny list
- Keep agents busy, treat them as a pool rather than run them in batches
- Add a "Notes" tab next to Diff/Logs for task-specific note taking or writing down reminders
- Don't show command failure when ripgrep returns no matches
- Persist sort preference in resume picker
- Have Plan mode make/edit .md and .txt files
- "Read" tool should contain line
- Built-in sound notification for interactive turn completion
- I would like to see a feature added that clearly indicates whether GPT-5.3-Codex-Spark is available or not.
- Allow disabling skills for memory agents
- Add a setting to configure persistent instructions
- Clarify intent of "Search recent tasks" feature in IDE Extension
- Enhance/Add "Search entire recent tasks" feature to IDE Extension
- Separate approval policy for file edits vs. shell commands
- As a user, I would like the ability to edit or drop previously queued messages without stopping the current turn.
- Make /review respond with numbered list
- feat(explorer-agents): built-in `input_context` tool — for iterative context pack building
- Feature: Hierarchical Agent Delegation & Per-Agent Tool Access Control in TOML Configuration
- Todo list above the input line
- Add AGENT=codex environment variable to UNIFIED_EXEC_ENV
- Allow codex to see images the code its working on generates
- Sub-Agent Timeout
- [FEATURE] Make context window usage more prominent in macOS app
- quick switch between edit/think modes
- Feature request – automatic orchestration from a plan file (Boss+ Workers)
- Support for pasting text with<Ctrl-v>
- Automatically run automation after rate limit window allows continued usage
- Retain queued prompts when temporarily switching to subagent thread
- Feature request: Limit Codex auto code review to default branch (or configurable branch allowlist)
- The 5-hour daily usage limit is being consumed extremely fast (in about 20–30 minutes of normal usage).
- Preview screenshot
- feature/alphavantage-provider-rate-limit-handling
- Output repeated
- Bug: Thread history data loss on restart — async persistence not flushed on shutdown (RolloutRecorder, SQLite WAL, session_index)
- Bug: UpdatedAt pagination skips valid sessions — sort key (mtime) and cursor comparison key diverge in should_skip()
- 502 Bad Gateway
- [Bug] --json and --output-schema are silently ignored when tools/MCP servers are active, resulting in malformed outputs
- Text clipped when scrolling in TUI
- Code change in CLI is hard to read
- Difficulty entering chat text in Codex CLI.
- "Introducing gpt-5.3-codex" message shown frequently / many times
- YOLO/full-access still runs non-interactive login shell; zsh-only env (rbenv) not loaded
- Conversation interruption/steering broken with 0.98/5.3
- TUI /agents: sub-agent shutdown exits app instead of restoring parent thread
- Codex CLI update bug
- Inference routing failure
- Feature request: reload-skills / reload-plugins for the current session
- multi_agent_v2 causes subagents to loop on MCP resource listing instead of completing simple tasks
- codex login --api-key shows misleading missing-value error instead of deprecation guidance
- Closed per reporter request
- app-server websocket disconnect leaves orphaned MCP-backed threads loaded
- Delete / DEL key immediately ends session
- Codex CLI: oversized tool outputs are ingested into context and can drain 5h quota disproportionately
- Codex' responses in CLI TUI are cut off
- Startup latency regressed in 0.117.0: codex blocks on app-server bootstrap before the TUI is interactive
- Windows desktop bundles rg.exe in WindowsApps, but executing it returns 'Access is denied'
- Codex cannot use bundled rg.exe on Windows but tries every time
- Message remains in composer after thread switch
- Desktop app sidebar only shows loaded recent subset, not full workspace history
- Preserve custom thread titles in archived threads
- Steering Message After Compaction Won't Work
- Codex-side MCP discovery/display bug or limitation
- VS Code extension host Code Helper (Plugin) high CPU with repeated thread-stream-state-changed warnings on macOS
- Tokens unreliable consumption
- Status sidebar similar to Opencode
- stream disconnected
- ■ {"detail":"The 'GPT-5.3-Codex-Spark' model is not supported when using Codex with a ChatGPT account."}
- New version mcp bug
- `/mcp` stops showing tools for MCP servers with hyphens in their names in `0.115.0`
- Suppress the warning `⚠ Codex could not find system bubblewrap at /usr/bin/bwrap` when sandbox is bypassed.
- codex exec panics on macOS when SCDynamicStoreCreate returns NULL (sandboxed environments)
- TUI composer sometimes auto-fills with `codex(pid) MallocSt...` text
- Support remote `--cd` for `codex --remote` command
- Codex client output truncated when scrolling in Zellij
- GPT 5.4 Performance degradation
- Completely unrelated and lengthy response
- openai_base_url config and OPENAI_BASE_URL env var are ignored — always calls api.openai.com
- codex is using chrome devtools to read files
- Bring back prompts
- Bug: TypeError when processing empty input array
- Feature request: support for custom output formats