codex - 💡(How to fix) Fix Codex Desktop crashes on startup with exit code 3221225501 (0xC000001D) on Windows 11 25H2
ON THIS PAGE
Recommended Tools
×6Utilities matched from this issue’s tags and category — try them while you read without losing context.
GitHub issue graph ai analysis
Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.
The report is written in English Markdown for sharing and archival.
Helpful · Quick feedback
Error Message
Codex Desktop app crashes immediately on startup with the following error: Most recent error: Codex app-server websocket closed (code=3221225501) 3. App immediately shows the error screen with code 3221225501
- Installed Git for Windows (resolved an initial "git not found" error, but this crash persists)
RAW_BUFFERClick to expand / collapse
What version of the Codex App are you using (From “About Codex” dialog)?
Unable to check — app crashes on startup
What subscription do you have?
Free
What platform is your computer?
Windows 11 25H2 (Build 26200.8457) x64
What issue are you seeing?
Codex Desktop app crashes immediately on startup with the following error:
(code=3221225501, signal=null). Most recent error: Codex app-server websocket closed (code=3221225501)
Exit code 3221225501 corresponds to 0xC000001D (STATUS_ILLEGAL_INSTRUCTION), indicating the bundled binary is executing an unsupported CPU instruction.
What steps can reproduce the bug?
- Download and install Codex Desktop from the official website (installed as a Windows app via MSIX/Store package)
- Launch Codex
- App immediately shows the error screen with code 3221225501
No further interaction is possible — the app never reaches the main UI.
What is the expected behavior?
Codex Desktop should launch normally and display the main interface without crashing.
Additional information
- CPU: AMD Ryzen AI MAX+ 395 w/ Radeon 8060S (Zen 5, 16 cores / 32 threads)
- RAM: 124 GB
- GPU: AMD Radeon 8060S
- Git: Git for Windows 2.54.0 (freshly installed)
What I've tried:
- Installed Git for Windows (resolved an initial "git not found" error, but this crash persists)
- Tried with VPN on/off (Astrill VPN, tested OpenWeb and StealthVPN protocols)
- Clicked "Update Codex" and "Reload" buttons — no change
- Rebooted the system — no change
This appears to be the same class of issue as #57698 in anthropics/claude-code, where the bundled Bun binary crashes with 0xC000001D on certain systems.
Vote matrix · Quick signals
Still need to ship something?
×6Another batch ranked right after the header list — different links, same matching logic.
TRENDING
- Feature Request: Configurable per-minute rate limiting (RPM) for models to prevent 429 errors
- Android: Hermes App + Termux install share ~/.hermes and cause silent permission loops
- hermes update emits unicode-animations ANSI demo in non-interactive logs
- hermes update downgrades aiohttp from 3.13.4 to 3.13.3
- npm install warns about deprecated @babel/plugin-proposal-private-methods
- DingTalk inbound media URLs are skipped as unreadable native image paths
- fix(dashboard): ChatPage clears header action buttons on ALL pages, not just Sessions
- [Bug]: check_web_api_key() hardcodes built-in backends — third-party web search plugins silently disabled
- Hermes Web UI 修复经验:GatewayManager 补丁、进程 D 状态、数据库升级问题
- Telegram gateway can silently drop turn after /stop with response=0 chars while internal work continues
- Feat: Load global agent instructions from ~/.agents/AGENTS.md
- Feature: China legal holiday / workday detection skill (with 调休 support)
- Multi-peer agent dispatch via proxy + Tailnet IPs as a complement to #9295
- Direct Gemini cost tracking is always $0.00 — resolve_billing_route() lacks gemini/google-gemini-cli branches + missing gemini-3.5-flash pricing entry
- [Feature]: feat(gateway): Support multiple delivery targets per webhook route
- Proposal: consolidate ~12 auxiliary fallback PRs into per-task fallback_providers
- Wecom always can't send files,
- [Gateway/Telegram] user.id is captured but not included in shared multi-user message prefix
- [Feature]: Publish effective ACP mode after mode changes
- [Bug]: HTTP 402 (billing/spend limit) misclassified and misleading advice given
- [Bug]: Content-filter triggered stream stall (output new_sensitive) does not trigger fallback_providers
- 예약 작업 실행 카드에서 원문 문구를 숨기고 예약 작업 상세 화면으로 이동 지원
- Context window changes to 256K after interrupted compaction and resume
- [Bug]: kanban.db corruption when multiple profile gateways share the same board DB
- [Bug] 移动端 /chat 存在不可隐藏的右侧空白占位栏
- cronjob tool fails to include required schedule parameter when using Grok models
- AI开发交流群
- 예약 작업 탭 카드 클릭을 풀 화면 상세로 전환
- 예약 작업 실행 카드의 원문 안내 문구 제거
- Add command and auto-recovery for disabled dispatcher
- [Feature]: 每次模型调用给出实时反馈
- [Bug]: terminal tool hangs 50-65s when reusing existing session in CLI/TUI
- TUI: 채팅창 하단에서 YOLO 모드 설정 가능하게 하기
- [Feature]: 目前没有兼容飞书富文本交互式卡片
- Proposal: Add kanban-artifacts to official dashboard plugin registry
- Feature: /queue cancel command to clear queued prompts
- [Bug]: /tasks and /agents do nothing
- Telegram busy-session interrupt buttons can disappear during long turns
- [Bug]: Default Docker image missing `markdown` dep — cron delivery silently falls back to plain text despite PR #5271
- Memory provider (holographic) silent failure: corrupted DB causes dead tools with no user-facing error
- [Bug]: Hermes unexpectedly modifies its own skills/system prompts during normal task execution, even when the user request is unrelated to agent configuration.
- [Bug]: MEDIA file not found sends message to main room instead of thread
- [Bug]: "_budget_grace_call" and "_budget_exhausted_injected" are redundant.
- Skill proposal: auditable decision trail for long autonomous coding runs
- Skill proposal: evidence-cited rationale archaeology for code history
- [Feature]: Add config option to suppress job_id and management hint from cron delivery output
- [Bug]: TUI+tmux looses keyboard/mouse input linux
- Gateway silently uses cloud API keys from environment with no warning, no cost visibility, no spending cap
- QQ Bot private chat (C2C) button approvals always rejected as unauthorized due to chat_type mismatch (dm vs c2c)
- feat(kanban): add busy_timeout PRAGMA to prevent WAL corruption under concurrent writers
- fix(dashboard): systemd service crash-loop when web_dist exists but npm unavailable
- TUI: No context explosion detection — sessions grow without warning
- [Feature] Allow autonomous coding delegation skills to consume central .providers config
- [Bug]: WhatsApp Integration Failure
- [Bug]: Kanban DB intermittent corruption after worker crash — missing WAL checkpoint
- [Bug]: Docker permission failure on Unraid (UID 99) — s6-overlay breaks HERMES_UID remap
- [Bug]: Web Dashboard: Session resume doesn't show full conversation history
- WhatsApp group chats: Mia responds to non-agent messages and restates instructions unnecessarily
- Gateway needs periodic platform liveness watchdog to catch zombie connections across all adapters
- bug(copy): /copy N counts from session start instead of latest response
- Support LM Link as a Provider Backend
- Implement Resource-Aware Model Fallback Chain (Local → Remote via LM Link)
- Docker backend: image arg clobbered with shell PATH export string, container fails with exit 125
- Kanban dispatcher writes a new .corrupt.bak on every tick after corruption (7,861 backups / 1.7 GB over 37h)
- Bug: web_crawl is implemented but never registered as a model-callable tool
- Bug: web-search-grounding skill routes models to browser/curl SERP scraping instead of the registered web_search tool
- Feature request: Oxylabs AI Studio as a web provider (search + extract + crawl)
- fix(weixin): MEDIA tag leaks for .md/.json/.yaml/.toml/.log files + no retry on session expiry
- Feishu messages: markdown tables not rendered, making them unreadable
- Discord thread action latency: 338s–405s responses with provider 300s timeout + retry (gpt-5.3-codex)
- [Bug]: Cron ticker dies silently — no error log, no watchdog, misleading status
- Dashboard WebSocket connections fail through Cloudflare Tunnel after recent security patch
- xAI OAuth fallback after provider switch due to stale encrypted_content replay
- bug: _extract_pricing crashes on list-typed pricing values, causing context_length fallback to 256K
- [Bug]: tool execute_code no longer inherits from the machine's environment
- [Bug/UX] Three config semantic traps for users debugging memory / multi-profile: silent failure of memory_enabled, profile config override, and per-profile log paths
- Desktop app: SSH Tunnel connection settings not persisted across restarts
- [Feature]: propagate DAYTONA_API_URL env to the daytona SDK
- Telegram MEDIA audio attachment can report success while delivering no attachment
- Add post-update coagency acceptance gate for critical Hermes capabilities
- fallback_providers not activated when 429 follows prior timeout recovery
- [Bug]: Tools array missing from API calls to custom Ollama endpoint
- acquire_scoped_lock: zombie processes (state Z) not detected as stale, causing false lock conflicts
- OpenViking memory provider sends slash-skill scaffolding as search text
- feat: per-agent toolset restriction for orchestrator pattern (parent-scoped no_mcp)
- `_load_config()` in delegate_tool.py never reads persistent config — breaks all delegation.* settings in v0.14.0
- [Bug] GCP Vertex AI connection fails with 404 using 'gcp' and 'rest' drivers, while direct curl works
- Path guard blocks write_file in temp dirs on macOS (/private/var/folders)
- Skill triggering is purely prompt-driven with no code-level enforcement
- [Bug]: browser_cdp opens stateless CDP connections, breaking Browserless/BaaS targets between tool calls
- bug(desktop): PROVIDER_BASE_URLS.anthropic hardcoded with /v1 suffix overwrites config.yaml base_url on every restart
- SessionDB: no crash recovery path when state.db is corrupted after unclean shutdown
- Add timestamp field to session JSON messages for proper ordering
- Backend health-aware fallback for Anthropic-compatible facades and auxiliary title routing
- Add opt-in proactive assist classifier for group chats
- [Bug]: search_files times out on WSL when ripgrep is missing
- [Bug]: web_extract gives dead-end error with only SearXNG configured
- [Feature]: Capability-Based Multi-Model Routing — unified config for task-type, domain-tier, and cost-aware provider selection
- Bug: `vision_analyze` tool fails to send images when using Docker terminal backend
- delegate_task() lacks model and provider parameters — prevents per-call subagent routing
- Startup noise: repeated Bitwarden Secrets Manager warning when BWS_ACCESS_TOKEN is unset
- [Bug]: Copilot Responses API 401 'input item ID does not belong to this connection' on assistant codex_message_items replay across credential/connection changes
- [Feature]: Tiered Memory Storage Architecture — Replace flat memory with 4-layer storage
- [Bug]: auxiliary title generation config is hardcoded to 30s and ignores config
- hermes auth add openai-codex writes only to credential_pool; runtime resolver reads providers.tokens — fails with 'missing access_token'
- [Bug]: Tirith shell scanner blocks pipe-to-interpreter even when LHS is a local user-owned executable
- hermes gateway install silently resets HERMES_HOME on macOS LaunchAgent
- Pre-send secret redaction hook + piiapi-filter skill
- Kanban dispatcher accepts kanban_complete with zero tool calls (no tool-evidence gate)
- Protocol violations (rc=0 without complete/block) don't increment failure_limit counter
- [Bug] kanban.db corrupted on WSL2 with multiple gateway processes
- [Bug]: Data privacy
- gateway: Telegram clarify-button resolver loses state on SIGTERM restart and 600s timeout
- Tirith approval buttons misleading: "permanently" is actually session-scoped
- computer_use (cua-driver backend) is too fragile and breaks auxiliary vision routing
- [Workflow Pilot] Validate infra Hermes contracted-goal workflow docs/template only
- Feature: built-in inter-agent mailbox tool
- Profile skills directory fully shadows global skills — no cascade fallback
- TUI: fallback_providers not passed to AIAgent (CLI works correctly)
- [Bug]: `/restart` silently fails on Windows when no Scheduled Task is installed
- [Feature]: Model name fields should be dropdowns sourced from /v1/models per provider
- Hermes mislabels upstream 429 quota exhaustion as missing Codex credentials
- Multi-bot free-response channels enter infinite ack-loops; bot-allow guard bypassed; human STOP ignored
- feat: route built-in session_search and memory tools through OpenViking when configured as memory provider
- [Feature]: Per-auxiliary reasoning effort configuration
- __write_scope_test__
- [Setup]: Connection problem
- [EPIC] Premium Product SDD v0 — Hermes Assistant Platform
- Spec v0: product vision, ICP, value prop, and metrics
- Architecture v0: control plane, brain, memory, voice, client, workers
- MVP v0: smallest shippable premium assistant
- Operating model: SDD, ADRs, DoR/DoD, and quality gates
- Roadmap v0: phased execution plan and validation gates
- [bug] same_tool_failure_warning does not auto-escalate to block; allows unbounded retry-loop cost amplification
- [Bug] Mobile thumbnail creator shows transcription controls and lacks thumbnail/export options
- [Bug] Mobile content cards open at too-low resolution / blurry when tapped
- [Bug] Transcript section renders an empty panel despite transcript header/copy controls
- [Bug] Detailed Analysis displays raw Markdown instead of rendered formatting on mobile
- [Bug] History item opens in awkward viewer with missing export options
- [Feature] Gallery modal should support horizontal navigation, caption editing, and high-res previews
- [Bug]: SOUL.md and AGENTS.md not loaded for Telegram gateway sessions in v0.14.0
- Bug: `--list_distributions` CLI flag crashes with TypeError due to name shadowing
- Bug: Race condition in OpenAI client replacement can close active connections
- Bug: `_generate_summary` returns None when max_retries=0, injected as conversation content
- Collection of minor bugs, typos, and inconsistencies found during code audit
- Credential pool: token_invalidated / token_revoked should be terminal failures, not 1-hour cooldowns
- delegate_task interrupt semantics: subagent killed while work is already complete
- Bug: Background Curation Prompts Leak into User Memory & Honcho Representations
- [Feature]: Allow agents to opt out of replying with `[SILENT]` marker
- Honcho issue sweep: duplicate PR clusters and cleanup plan
- approval.py only wired to terminal_tool; MCP-wrapped commands bypass dangerous-command + Smart-mode gate
- TUI remote gateway attach (HERMES_TUI_GATEWAY_URL) fails with 404 — /api/ws missing from APIServerAdapter
- Fix Codex stream None output recovery
- gateway: `gateway_state.json` heartbeat tick missing — WebUI cross-container liveness check fails for idle gateways
- [Bug]: Error: 'NoneType' object is not iterable
- [Bug]: Codex Responses stream aborts on terminal null output
- fix: cron ticker thread stops silently, jobs never fire
- fix: hermes cron list crashes when job has null deliver field
- Per-subscription toolset override for hermes webhook subscribe
- openai-codex provider crashes: SDK parse_response fails on null output from Codex backend
- Proposal: record which SOUL version each kanban run executed under (provenance)
- Proposal: per-board approval policy override (scope approvals.mode per kanban board)
- [Bug]: Codex Responses stream ending without `response.completed` crashes with TypeError ('NoneType' object is not iterable) — bypasses existing backfill at codex_runtime.py:265-280
- Feature Request: Integrate Microsoft SkillOpt for Self-Evolving Agent Skills
- [Bug] Gateway systemd service fails on Fedora with SELinux enforcing (status 203/EXEC)
- [Feature]: Multi-profile Telegram group communication — bots don't see each other's messages
- Avoid systemd gateway restart/takeover loops from Restart=always + --replace
- Detect duplicate gateway supervisors in doctor/status diagnostics
- [Bug]: Codex Responses streaming crashes with TypeError: 'NoneType' object is not iterable (openai-codex / chatgpt.com backend)
- The SSE connection remains active indefinitely and fails to close normally when an annotation reply is triggered.
- On the MCP list page, the search results include non-MCP tools.
- dep inject model validate
- provider llm do not add
- --frozen flag in uv sync causes plugin installation failure in China / limited-network environments
- [SECURITY] Missing SECURITY.md — no vuln reporting channel for 142K-star project
- [SECURITY] CORS allows wildcard * in production default (CWE-942)
- [SECURITY] CRITICAL: Empty SECRET_KEY + default credentials across .env.example (CWE-798)
- Chatflow with file and Human Input is stuck
- Add CLIENT SETNAME to Valkey vector store for connection identification
- The ChatFlow App is unable to reply properly when I upload file using a file url
- SEEKDB can not use in V1.14.2
- Add Amazon URL parsing and metadata extraction for product comparisons
- Restore Ctrl+B to 'cursor.left' (backward-char) by default in Unix environments
- Support customizing execution shell (e.g. to zsh) in Unix-like environments
- YOLO mode ignoring user requests
- Bug when pasting json String
- Feedback: Code Output Formatting in CLI Environments
- Fix model config name in token counter
- ℹ High memory usage detected (8.64 GB).
- prno
- AGY 1.0.2: -p/--print mode outputs nothing to stdout on Windows
- n8n Cloud executions page causes instance to go offline with 502 error
- Bug with scripts JavaScript & Python
- workflow not saving
- Couldn’t connect with these settings No testing function found for this credential. While editing SSH and SFTP credential
- Broken Node Report
- AI Agent v3 does not set tool_choice on iteration 1, causing small models on OpenAI-compatible providers to skip tool calls
- Getting activation key
- ## Bug: Python Task Runner fails on native npm install — `src/` folder missing from published package
- [BUG] Trust prompt and session reload triggered repeatedly when invoking 'claude' with uppercase casing (CLAUDE) on macOS
- [BUG] Claude Desktop (MSIX) bundled bash leaks TEMP=/tmp into child JVMs, breaking Gradle and any Java Selector.open() on Windows
- [Bug] Ultrareview synthesis step crashes after backend completes findings analysis
- [FEATURE] Support audio file attachments (WAV, MP3) in Claude Code
- [BUG] Text is randomly turning blue while using Claude Agents
- [BUG] Keychain credential partition_id silently resets to "apple-tool:" alone on ~5min cycle (macOS Tahoe 26.5, Claude Code 2.1.145)
- AskUserQuestion: no way to view truncated preview content ("X lines hidden")
- [Bug] Mac Desktop App Crashes Repeatedly During Session
- Team in-process mode: process residue on disband + agents don't consume task_assignment
- [FEATURE] Support Multiple GitHub Accounts
- ✨📲 {{ 🎀^【【%#@&*^#🎀^✚卍 卍Flynas infolinia polska 24h %#@&*^#🎀^】✚卍 卍】✨📲 Czy Flynas oferuje całodobową obsługę klienta?%#@&*^#🎀^】】✚卍 卍⁂}}{{ 🎀^
- 🍉 ⋆ 🍓 🎀 ✚卍 卍 卍 +48 58 8810 086 卍 卍 卍✚ 🎀 ❀ 💗 🍓 ⋆ 🍉卍【2➃𝕙ㄖ尺𝔸Ŝ {{{Flydubai infolinia polska 24h}}】Numer kontaktowy Flydubai 【2➃𝕙ㄖ尺𝔸Ŝ{{Jak skontaktować się z obsługą klienta Flydubai? ?⁂}}{{ 🎀^ }}}】卍✚2➃𝕙ㄖ尺𝔸Ŝ{{{❀ 💗 🍓 ⋆ 🍉
- [BUG] ALT+TAB back to VS Code extension causes certain key presses (e.g., PrtSc, Henkan) to trigger menu bar focus
- 🍉 ⋆ 🍓 🎀 ✚卍 卍 卍 +48 58 8810 086 卍 卍 卍✚ 🎀 ❀ 💗 🍓 ⋆ 🍉卍【2➃𝕙ㄖ尺𝔸Ŝ {{{Qatar Airways infolinia polska 24h }}】2➃𝕙ㄖ尺𝔸Ŝ{{ Numer kontaktowy Qatar Airways【 Jak zadać pytanie na temat Qatar Airways? 】}}卍✚2➃𝕙ㄖ尺𝔸Ŝ{{【❀ 💗 🍓 ⋆ 🍉
- Feature Request: Support global ~/.agents/AGENTS.md along with project-level AGENTS.md
- `bgIsolation` guard recovery path is broken: error tells agents to call `EnterWorktree`, which is a deferred tool
- [BUG] False positive safety filter block for Polish-language security audit question in own PHP CRM project (VSCode extension)
- [Bug] Environment variable expansion not working for $HOME in paths
- [BUG] Payment declined for Claude Pro subscription from Nicaragua with multiple cards
- Claude Code agent set Meta Ads budget 100x wrong — caused NT$11,168 (USD ~$350) financial loss
- The worst experience with a AI network, I'm not willing to pay $100 for it anymore.
- [BUG] PTY leak: Claude desktop app accumulates ~490 /dev/ptmx file descriptors
- [Bug] Anthropic API Error: text content blocks must be non-empty
- Shell snapshot uses unguarded $ZSH_VERSION, breaks set -u (nounset)
- [DOCS] Document server-side system prompt experiments and their opt-out controls
- [DOCS] Document feature-flag and bootstrap cache behavior when opt-out variables are enabled
- [DOCS] Network docs should warn that untrusted LLM proxies can modify Claude Code request context
- Feature request: allow customizing terminal tab title (project name vs conversation summary)
- [BUG] Regression in v2.1.150: cannot select text outside current screen
- [Bug] Claude agents interface: incomplete skill loading, copy functionality UX issues, and terminal responsiveness degradation
- [FEATURE] Animate the project name chip while waiting for user input
- [BUG] [VSCODE] Algebra formatting broken on VS Code extension
- Drag-drop and screenshot clipboard paste fail on Linux; works only after re-copy from image viewer
- [VSCode Extension] AskUserQuestion panel hides conversation context with no collapse shortcut
- [Cloud Mode] Allow overriding the diff view base branch (currently locked to repo default branch)
- [FEATURE] Configurable Spinner
- [BUG] /remote-control fails with "failed to connect: /login" on Pro account, desktop app v1.8555.2
- Thumbs up/down
- [FEATURE] Show pending diffs in IDE before edit execution
- API Error: 400 messages: text content blocks must be non-empty on session resume when assistant turn contained only tool_use blocks
- Durable stats ledger independent of session jsonl files
- [BUG] JetBrains plugin 0.1.14-beta: NoClassDefFoundError io.ktor.server.cio.CIO on startup (IDEA build 253.31033.145)
- [BUG] --dangerously-skip-permissions ignored when SSH_CONNECTION is set or TERM is empty
- [BUG] Plugin runtime loader and `claude plugin validate` accept divergent schemas
- [Bug] Microsoft 365 MCP connection fails in Claude Code but works in Claude Desktop
- Sonnet 4.6 - Destructive rm -rf executed on explicitly excluded directory during cleanup session
- Bun segfaulting in claude-code, thought you should know
- Sandbox: hardcoded deny on .idea/.vscode/.claude/.husky writes has no user override
- [BUG] Gateway model discovery `capabilities.image_input.supported=false` is ignored in session image behavior
- Agent crashes with 'exit 1 before init' when cwd no longer exists (e.g. after project move)
- Tool-call opening tag gets stuck on a malformed form and is unrecoverable within the session
- [Bug] Corrupted ASCII characters in CLI output on Antigravity IDE
- [BUG] Slash command name collision: plugin skill shadows built-in /release-notes
- [BUG] Missing agents at start of cli, due to using Glob on .Claude/Agents folder starting with a dot
- [FEATURE] Show resulting context-window % when using esc+esc to rewind the conversation
- [Generative UI Widget] 默认字号过小,建议提高默认值或支持用户自定义字号
- [Bug] Non-verification error causing repeated failures in Claude Code actions
- Preview blocks cross-origin redirects (Stripe Checkout, OAuth, etc.) — silent failure with no fallback
- [Bug] Claude Code silently creates paid GitHub Actions workflows without user consent or cost warning
- [BUG] When opening a terminal in the company's internal CodeLab container and entering Claude Code, the output often shows garbled characters, especially for Chinese characters. English characters are mostly fine.
- [BUG] 双击 ESC 回滚会话后,Task 列表未清理
- Bash tool returns "internal error" on git commit under fish + multi-step pre-commit hooks (cmd actually succeeds)
- [BUG] Auto-update orphans MCP child processes on Windows, breaking next launch of npx-based MCP servers (EBUSY)
- [FEATURE] CWD visible in claude app
- [BUG/FEATURE] LLM agents fabricate non-existent `disabledSkills` settings field — silent no-op needs validation or the feature itself
- WorktreeCreate hook 'Hook cancelled' for worktree Agent when dispatched alongside a sibling Agent call in one message
- [BUG] Claude Code uses legacy XTMODKEYS even when terminal advertises the kitty keyboard protocol
- [Bug] Opening Claude pane resets VSCode Neovim to Insert mode
- ultraplan: 'Claude GitHub app must be installed' despite app being installed
- [BUG] Rate limits blocking multi-agent Claude Code workflows even at highest paid tier
- CLI-created sessions invisible in Desktop app (reverse direction broken)
- No UI to restore/unarchive archived sessions in Desktop app
- [BUG] Cowork Workspace Fails on Windows with EBADF on rootfs.vhdx rename #47343
- [BUG] Cowork sandbox VM bundle not found in Roaming path on Windows 10 Pro (Store/AppX install) — hardlink workaround
- [BUG] /exit prompts to remove worktree even when other Claude Code sessions are still active in it — accidental remove breaks live sessions
- Sandbox cannot create subdirectories under project tmp path
- Plan mode should enforce read-only at the tool layer, not via LLM instruction
- [Feature Request] Optimize token usage for git commit and push operations
- [BUG] 鼠标滚轮在 Android Studio 集成终端中无法滚动(功能回退)
- [BUG] Bash tool fails with exit 127 when shell snapshot replays extglob syntax in non-interactive bash 5
- [BUG] PreToolUse hook is not invoked after a static ask rule for the same command pattern receives session-level approval
- [BUG] How to disable ctrl + c deleting the hole import ??????
- [Bug] Misleading usage credits error message displayed unnecessarily
- [Bug] Claude ignores context and provides incorrect feedback on skill implementation
- [BUG] Plugin sensitive userConfig not persisted to keychain/credentials.json/settings.json — values lost on Claude Code restart
- [Feature Request] Improve file completion suggestions for `@` prefix with smarter ranking
- [FEATURE] Sound Notification for claude agents
- [FEATURE] Handle non-gitignored files in `.worktreeinclude`
- Support version pinning in plugin install command
- Claude Code desktop app steals focus on every notification (regression — was: silent banner only)
- [BUG] `claude respawn --all` advertised in --help but rejected as "unknown option" (v2.1.150)
- Detached/pop-out windows ignore zoom level and Cmd+/- doesn't work on them
- [Bug] Session reload triggered unexpectedly during rename operation consuming tokens
- [BUG] Plan mode state inconsistency when ExitPlanMode is interrupted by user
- [BUG] [Cowork UI does not show plugins installed from "claude-for-legal" marketplace — claude- prefix conflict
- [BUG] Inputs mouse movements in chat area when switching from /background or agents' view
- [FEATURE] Always-visible / always-expanded tool call output in VS Code extension
- [FEATURE] Tab autocomplete inside bash mode (!) — files, paths, and command history
- [Feature Request] Auto mode should respect allow list settings from non-auto mode
- Ctrl+Shift+C no longer works in GNOME Terminal (regression)
- Feature request: actionable Approve/Deny on mobile push for permission prompts
- [Bug] Plan-mode-exit dialog ignores `permissions.defaultMode` setting
- [Feature Request] Add confirmation prompt before executing destructive database operations
- [BILLING FRAUD] Unauthorized gift subscription charged to my account - No support response after 10+ days
- Cannot copy text from Claude responses in interactive TUI (GNOME Terminal, Ubuntu 24.04)
- [BUG] SSH marketplace add error message hardcodes "GitHub" regardless of host
- [FEATURE] Emit OSC 7 on agent-focus-change so new terminal tabs inherit the focused agent's cwd
- [Feature Request] Improve custom API key configuration for cost attribution tracking
- [BUG] Repeated "Image couldn't be processed" API errors consuming usage limit in Claude Code
- [Bug] Opus 4.7 with extended thinking intermittently emits unparseable tool_use blocks
- [FEATURE] Predictive agentic loop
- WebSearch tool returns 400 invalid request parameters on all queries
- [BUG] /tui mode switching causes cursor to disappear and breaks mouse wheel
- /usage and /context commands stuck on 'Loading usage data…'
- [FEATURE] Cowork: per-task-title retention policy for scheduled-task session residue
- [BUG] Plugin cache not rebuilt after git pull on marketplace repo — LocalPluginsReader finds 0 local plugins
- [BUG] Azure VIrtual Desktop Cowork Deployment
- claude plugin sync — auto-install plugins from settings.json (like npm install)
- [BUG] Claude Code silently deletes conversation transcripts after 30 days by default
- [FEATURE] Allow configuring WebFetch cache TTL
- Duplicate session names when multiple sessions share the same git branch
- claude.ai connectors (Gmail/Calendar/Drive) expose only auth stubs in a session despite being Connected + authorized
- [FEATURE] Don't re-prompt for approval on WebFetch cache hits
- [Feature] Public deep link to open 'Add marketplace' dialog in Claude Desktop with prefilled URL
- [BUG] Remote Control: worker's custom skills (~/.claude/skills/) missing from Claude app slash-command autocomplete
- [Bug] Inconsistent adherence to coding conventions in generated code
- [BUG] tui: fullscreen + /memory + select memory = unusable terminal, hard stuck even closing/reopening doesn't work
- Plugin skills from custom/private marketplaces appear in slash menu but fail with "Unknown skill" on invocation
- settings.json partial rewrite strips statusLine, enabledPlugins, hooks mid-session
- [BUG] Switching to mimo-v2.5-pro fails after reading images in mimo-v2.5
- Vim NORMAL-mode motion (h / ←) should return from an agent to the agent list in agent view
- [Feature Request] Add disableManagedConnectors setting to disable auto-loaded connectors
- [FEATURE] Switch between plans in VS Code extension
- [FEATURE] Read-Only Auto Mode
- settings.json: add user-approval-required mode for dangerouslyDisableSandbox
- AskUserQuestion overlay obscures the assistant's previous message
- [BUG] Claude Agent in PhpStorm no longer has the feedback actions
- [BUG] Device lock stuck after account switch on Windows - automatic release mechanism fails
- [BUG] VS Code extension goes silent and deaf
- [Feature Request] Add ability to remove/uninstall individual plugins via /plugin command
- Hardware Buddy BLE cannot surface or answer AskUserQuestion options
- auto-mode classifier mis-reads conversation state (subagent verdicts + disambiguating instructions)
- Plugin command/skill named "doctor" is invoked instead of built-in /doctor
- False positive: legitimate UE5.7 game project flagged as potential malware, blocks all code editing
- [FEATURE] Ability to reference files using the "@" symbol in the agent view
- [BUG] Appeal Form Redirect Loop After Account Restriction
- bug: Task subagents receive empty MCP tool surface (main-thread works, subagent context drops all mcp__* tools) — Windows + Opus 4.7
- /ultrareview returns empty findings + consumes free credit on large PR (Issue #50029 repro)
- Security: settings.local.json permissions file has no integrity protection — silent injection enables pre-approved execution
- [BUG] Account & Usage modal stuck on "Loading usage data..." — never completes
- [FEATURE] Voice dictation: recognize spoken punctuation commands as punctuation marks
- EPERM cascade on Read/Edit/FileHistory for session-created files in long-running autonomous sessions
- Thai text input: Backspace deletes entire grapheme cluster instead of one code point at a time
- Remote Control syncs only desktop->iPhone (iPhone->desktop messages not delivered)
- Remote Control: input silently dropped on all clients — messages appear sent but never arrive at local process (regression ~May 25)
- Desktop Recents list doesn't surface sessions previously created via the CLI
- Feature: Expose background task state to SessionEnd hooks
- Developer → Open Hardware Buddy… menu missing after enabling Developer Mode
- No way to suppress OAuth warnings for unconnected claude.ai integrations at startup
- Feature request: per-repo hook-approval prompt for ./.claude/settings.json (parallel to tool permissions)
- [BUG] Auto-generated permission rule with a literal : inside a quoted argument is rejected on load ("The :* pattern must be at the end")
- [BUG] Active runtime config (model / context window / effort) not observable from inside agent turn — Opus session self-reported as Sonnet, blocking workflow
- [BUG] The model ignored all the rules and ignored all the constraints in memory as well as in global files
- [BUG] Cowork unavailable on standard user accounts - VM Platform not enabled error misleads to reinstall
- Agent tool: add context isolation mode to prevent evaluator bias in multi-agent setups
- [FEATURE] Auto Mode in the Cloud Repo - Only available on local at the moment
- [BUG] [Desktop App] Deleting session working directory makes app unusable with no recovery path in GUI
- [BUG] API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited
- [BUG] /usage heatmap is off by one weekday
- [BUG] Claude Desktop - Dispatch shows "Desktop appears offline" persistently
- [MODEL] Claude Code Web multi-agent review synthesis fabricates verification and expands issue scope
- [BUG] Write/Edit tools truncate to previous file size when overwriting existing files (Windows desktop / virtiofs)
- [BUG] Running claude in a folder with a .git folder without rwx permissions for the current user prevents running bash commands
- [BUG] [Billing] Pro Annual → Max upgrade blocked by "billing address has changed" error, no human support available
- Long streamed responses occasionally triplicate sections of output
- [Refund Request] claude-opus-4-7 reported unverified inference as 'root cause' in declarative tone, violating user's explicit global rules
- [BUG]Claude Code desktop app, AWS Bedrock, us-east-2, "invalid model identifier" error, started today, two machines affected, one machine that is the root user on AWS Bedrock is working fine through Claude Code -working fine with identical credentials.
- [BUG] Windows: PowerShell tool absent from schema when Git Bash is on PATH, despite pwsh.exe being present and detectable
- Add config to disable idle/awaiting-input reminder notification separately from completion notifications
- Feedback: Limitations found during intensive 8h MCP development session
- Plan proposal card: action buttons layout breaks when the Plan side panel is wide (narrow middle pane)
- [BUG] Plugin version not respected
- Account-linked MCP machine delegation across devices
- Plan side panel: content stops expanding at a fixed width — large wasted empty margins on wider panel
- [BUG] [session limit]
- Cowork sandbox (virtiofs bind-mount) silently drops git unlink() with EPERM, leaving stale tmp_obj_* files and .lock files on host
- Desktop App broken on Windows with Bedrock after v2.1.149
- isolation: worktree subagents silently write outside the worktree via absolute file_path (Edit/Write target primary clone, not the isolated worktree)
- [BUG] Cowork macOS: Gmail and Google Calendar MCP connectors return "The caller does not have permission" on every call for one account; other users on the same Workspace tenant unaffected
- [Bug] CLI performance degradation with large codebases
- [BUG] ASCII characters corrupted content
- [Bug] Linked worktree sandbox auto-allow missing main .git directory and denyWithinAllow paths off-by-one
- [BUG] claude mcp add my_mcp_server ... settings not updated
- Remote-control: phone-app approval dismisses popup but never reaches local process
- [BUG] Crap in the console!? WHY
- [BUG] Claude Code Crash
- [BUG] Cowork macOS: all claude.ai-hosted MCP connectors fail — desktop OAuth token requested without `user:mcp_servers` scope
- VSCode extension does not render `permissionDecisionReason` in permission prompt for "ask" decisions
- Claude responds "No response requested" after tool rejection, ignoring user messages
- [BUG] Switching focus between local Code sessions kills the unfocused session's CLI process (SIGTERM, exit 143) instead of leaving it running in the background
- [Bug] Anthropic API Error: Prompt is too long in stop hook evaluator
- [BUG] mcp__github__subscribe_pr_activity silently drops check_run completion events
- Opus 4.7 tool-use compliance cliff between 24th-26th May — instrumented evidence
- [BUG] Commands in code blocks cannot be copy-pasted directly into terminal
- [BUG] Claude extension in VS Code fails to honor autoMemoryDirectory setting in settings.local.json
- Monitoring cycle 18:03 UTC — 20k pending orders accumulation in ft_orders
- [DOCS] /btw fork hotkey not documented
- Feature Request: Allow Claude Desktop to be configured to only read MCP servers from claude_desktop_config.json (not ~/.claude/.claude.json)
- [BUG] It's not easy to understand fast mode turned ON or OFF
- [Bug] Unexpected High Token Consumption Across Sessions
- [Bug] Credentials exposed in git history via .env.dev file commit
- Feature request: explicit model-version pinning in settings.json
- Plan approval popup: pressing Space while typing auto-confirms and dismisses without undo
- [Feature Request] Improve discoverability of session history in Agent View: peek vs attach unclear, Ctrl+O not hinted
- [BUG] MSIX-packaged Claude Desktop (Windows) strips PATHEXT to .CPL and empties WINDIR for spawned MCP subprocess shells
- [DOCS] /login *is* available when CLAUDE_CODE_USE_BEDROCK is set
- Memory rules saved but never applied — Claude Code operates in wrong environment across sessions
- [Bug] Anthropic API Error: Tool result block missing corresponding tool use block
- [Bug] Tool calls hang with no response on macOS - Edit/Bash/Write become unresponsive
- [FEATURE] tab-completion mode similar to bash
- [BUG] Edit tool fails on files containing embedded Unicode PUA characters (Nerd Font glyphs) — falls back to PowerShell workaround unnecessarily
- [FEATURE] adding comments to chat dialog
- [Feature Request] macOS sandbox: auto-manage allowMachLookup for TLS/system services
- [BUG] Auto Archiving Sessions in Use....
- [BUG] `normalizeToolInput` strips `cd <cwd> &&` from Bash commands before PreToolUse hooks can inspect them
- Remote Control "not yet enabled" on Max plan — v2.1.126 (Arch Linux)
- Preview MCP `preview_start` ignores `runtimeExecutable` and crashes on corepack-wrapped pnpm under Node 20
- Directory connector visible to one Owner but invisible to another Owner in same Team organization
- Auto-retry sandboxed Bash commands with sandbox disabled instead of instructing the model to do it
- [BUG] @-file references resolve to parent git repo root instead of cwd when launched inside a jj workspace (.workspaces/<name>/)
- [BUG] TUI emits strikethrough on arbitrary text - reproduces on 2.1.150 in both diff view and normal output
- Claude regenerates entire files instead of surgical edits for metadata-only changes
- MCP server config: project-level should merge with user-level (env, etc.), not replace the whole entry
- mental issue of opus
- [BUG] Claude Desktop macOS — Helper crash loop in c-ares (`ares_dns_rr_get_ttl`) — false "Endpoint Security blocked Q6L2SF6YDW" dialog (1.8555.2 + 1.9255.0)
- [BUG] Scheduled task ran 100s of times unprompted as desktop app crashed
- [BUG] iOS Dispatch Streaming Input Mode (orange bubbles) replaced by banner UI after app update 1.260521.0
- Windows: stale lock directories block credential persistence after abnormal exit
- Claude does not check if tool/software is already installed before attempting reinstall
- Claude forgets actions performed earlier in the same session (intra-session amnesia)
- Sub-agents declare work as PASS without performing actual verification against reference
- Sub-agents ignore explicit rules embedded in their own prompt
- Claude defers fixing known problems instead of acting immediately
- [Bug] Missing or invalid command input
- Feature request: exclusive-allowlist mode for --allowedTools (currently additive on top of built-in default safe-list)
- [Bug] Diacritical marks disappear when typing uppercase letters
- Claude Code makes confident wrong claims, deflects blame, and expands scope without verification
- VSCode extension voice input forces English transcription, ignoring pt-BR system locale
- [BUG] Unable to connect to API : TCP shows ESTAB but app reports ConnectionRefused on prompt input
- [BUG] Cannot scroll up or down in Calude CLI chatbox
- [Bug] Unable to determine issue - please provide bug report details
- [BUG] Windows 11 Claude Desktop App at launch asks for Security code from Email to Log in: Email has a log in button that launches Edge and logs me in.
- Plugin manifest should support `permissions.allow` for bundled console scripts
- [Bug] pdf-viewer MCP built-in server - UtilityProcess spawn timeout (Claude Desktop, Windows 10)
- [REDACTED]
- [BUG] False positive "Usage Policy violation" errors on normal coding requests
- Desktop CI monitoring panel: state cached at session start, no refresh, misleading error sends users down wrong debug path
- [FEATURE] Expose transcript prompt-jump ({ / }) as a bindable action for non-US keyboard layouts
- [BUG] Slash command does not work with modified keybindings
- [Feature] Allow customizing or simplifying thinking status verbs
- [BUG] Login magic link routing to wrong email — locked out of account, AI support loop for 2+ hours. UNRESOLVED, Receive emails, codes, etc. to correct email. (Issue #61892)
- Serious accessibility issue with "clever" synonyms for "Working..."
- [FEATURE] Allow switching agent configurations mid-session via the /agent menu
- [BUG] ctrl+<letter> bindings silently fail on non-Latin keyboard layouts (Cyrillic / Greek / Arabic / ...)
- [FEATURE] Allow clearing or compacting the Dispatch thread
- [BUG] isolation: "worktree" — only 1 of 7 concurrent worktrees branched from current HEAD; other 6 used origin/HEAD
- Inter-session coordination: wake / notify a parent session on child (sub-session) turn-end or idle
- [BUG] Bash tool creates 0-byte phantom files with code-fragment names during complex python -c execution (Windows + Git Bash)
- Agent tool: automatic async escalation when run_in_background is unset is undocumented
- [BUG] `/remote-control` fails in desktop-app-embedded Claude Code with "no org UUID", and there is no user-facing remediation
- [BUG] Button hidden behind X button in Cowork connector tool calls sidebar (macOS)
- [FEATURE] Let me change the location of .mcp.json and or define its config in setting.json or with environment variables
- [Feature Request] Add user confirmation prompt for unauthorized shell commands in auto mode
- MCP reconnect causes AI to hang waiting for ToolSearch instead of proceeding
- [BUG] PreToolUse hook does not fire when Read tool reads image/binary files
- Claude fabricates visual analysis of images to appear confident
- Feature request: Add setting to disable automatic file reveal in editor during Read/Edit operations
- [BUG] Bash tool: injected grep/find shell functions terminate the calling shell under bash < 4.0 (macOS system /bin/bash)
- Gmail (and other claude.ai) connectors not available in Claude Code CLI
- [BUG] "Buy credits" button permanently disabled - Free tier account incorrectly shows $500 limit, HTTP 429 errors on billing page
- [BUG] Cowork impossible on MSI MPG X570 / Ryzen 9 3900X — vmms service missing, Hyper-V causes boot failure
- [BUG] Persistent MCP error: tools.77.FrontendRemoteMcpToolDefinition.name - Account-level corruption
- [BUG] no indication of completion, "Resumed session >" messages continue to appear in a loop
- Monitor task list — stale entries persist across app restart + UI/CLI count divergence
- Claude in Chrome: extension never connects to Claude Code CLI (macOS)
- [FEATURE] vscode plugin option to open claude as new tab in an existing pane rather than split
- [BUG] Bash (dir/b "C: \ FilePackage") error Path contains null bytes
- * wildcard in hook `if` pattern doesn't match strings containing spaces
- [BUG] mcp__Claude_Preview__preview_eval hangs the entire preview session without returning any error
- [Feature Request] Add color sorting option to Claude Agents view
- Codex Desktop high CPU from unbounded active thread metadata and local history/list processing
- Codex CLI 0.129.0+ fails in WSL with "stream disconnected before completion"
- 已撤回
- Display file inline diffs
- IDE context could not be enabled
- Claude settings migration prompt clobbers user-level config.toml when no project-level config exists
- Model reads image files with the same name, UI always shows the first loaded image
- Add team-level Skill Hub for sharing, syncing, and governing Codex skills
- TUI turn interrupt failure can lose session context
- Feature request: allow moving conversations between projects
- The status display shows that the ID and credit limit are not displaying correctly.