Digest / October 5, 2026
Claude Code Issues Digest — October 5, 2026
282 new issues filed ·
171 resolved in anthropics/claude-code (UTC day).
New issues (282)
- [Mods] Code with format: 'diff' draws as plain text in the Desktop app (terminal draws it correctly)
- I need more information to generate an appropriate GitHub issue title. "subagets" appears to be incomplete or unclear. Could you please provide: - The full bug report or description - What error mess
- Problem report
- [BUG] Desktop app renderer OOM-crash loop when opening a large cloud session (600 MB → 5 GB in ~50 s)
- [BUG] MCP OAuth refresh started during session shutdown is abandoned, and the server ends up logged out
- [Desktop] Sidebar session-to-group assignments lost after Windows reboot (groups remain, sessions become ungrouped)
- WSL: text copied from Claude Code does not paste into an RDP session
- Claude repeatedly invents human-approval gates, writes them into my system as my rules, and misreports what it did
- [BUG] Claude Desktop (macOS): main window stops responding after idle ("Main webview is unresponsive, will kill and reload"); Close/Open does not recover
- [Bug] Edit feedback ignored when declining multiple successive edits
- necesito bypass por que se bloquean los permisos?
- [BUG] [Desktop/Windows] Archiving a session leaves its worktree behind when the repo is inside OneDrive
- [BUG] Voice hold-to-talk stops mid-hold on macOS: 200 ms release timeout is shorter than real key-repeat gaps (0.5–1.4 s)
- [BUG] Claude code is ignoring repository rules against AI contributions
- [Bug] False positive spam/abuse detection on benign networking questions
- [FEATURE] Let the agent set a semantic work phase on a session (Brainstorming / Implementing / Shipping)
- [BUG] security-guidance: pattern reminders never fire for NotebookEdit (new_source is not read)
- [BUG] security-guidance: yaml/torch/np.load rules stop at the first ")" — false positives with Loader=SafeLoader / weights_only=True after a nested call, missed np.load(..., allow_pickle=True)
- [BUG] Cowork (Windows): "Upgrade projects" forces archiving of local-folder projects — no way to upgrade them, and new project creation is blocked until they are archived
- [Bug] Unsafe file deletion attempt during agent execution
- TaskCreate/TodoWrite tools missing from session in CLI 2.1.289
- claude-api hillclimb guide: default flow dir under protected `.claude/` means allow rules don't cover the loop's own writes
- claude-api eval/hillclimb guides: the `model × usage` cost recipe ignores the cache TTL and model-specific cache-read rates; prefer an agent-reported cost when a row has one
- [BUG] claude install hangs indefinitely at 99% CPU with no network activity (2.1.289, Linux x64)
- [Bug] Claude Code repeats similar mistakes causing excessive token consumption
- Hooks added to settings should require explicit per-hook user review (no silent activation)
- "Add to project" shows "No matches" even though the project exists
- Desktop: SSH / Remote Control sessions are not grouped by project folder and not visible in the Desktop app on the remote machine
- [FEATURE] Mobile Remote Control: persistent context-window usage indicator
- [BUG] Native Windows: mods/diff merge-state fixture misses the normalized Git directory
- [FEATURE] Trust the kitty graphics probe for SwiftTerm terminals (Moo), not only kitty/ghostty by name
- Worktree guard fences all of /tmp when /tmp holds any `.git` entry (empty dir accepted as a repo root; resumed background sessions anchor under /tmp)
- [Bug] Agent sessions retain context indefinitely, violating framework rules and consuming 51% of weekly quota in one day
- Remote Control: allow --continue / --session-id to keep multi-session mode (or restore all sessions of an environment)
- [GitHub integration]LYRDSL_project
- [FEATURE] Desktop: filter session list to show only sessions with unread/attention dots
- [Bug] Model performance degradation: Claude 5.5 Opus quality regressed to Opus 5 level
- [Bug] Claude Code interactive mode incorrectly blocks Fable 5.1 access despite sufficient plan quota
- [GitHub integration]
- [GitHub integration]
- [FEATURE] Task chips (spawn_task): name the worktree and branch from the task, and let the spawner set the worktree destination
- sisisis[GitHub integration]
- [BUG] --model value is not verified before use
- [Model behavior] PR approved on the contributor's own description as evidence of provider behaviour
- [Model behavior] Normative source cited for a verdict does not cover the case under review
- [Model behavior] Two PR verdicts given in one pass with procedure steps skipped and undisclosed
- [Model behavior] Told to read memory, the model writes a duplicate memory entry instead
- [Model behavior] Contributor's own passing tests reported as support for a review verdict
- [GitHub integration]
- [BUG] Idle auto-update stops all sessions and Remote Control never comes back after the relaunch
- [Model behavior] Proposed removing correct code from a PR because official docs were silent, without searching
- [Bug] Safety check blocks all SSH commands after credential materialization denial, no way to clear it
- [GitHub integration]
- [FEATURE] Add a permanent delete option for past conversations in the VS Code extension
- [BUG] Claude in Chrome: tools report "Browser extension is not connected" although extension, native host and socket are running (Linux)
- Subject: Local folder approval is not persisted: the "Allow Claude to work in a…
- [BUG] Regression of #94259: Desktop (Windows) code block Run opens a new terminal tab instead of the existing one
- [BUG] Desktop (Windows): Ctrl+V does not paste in the built-in terminal panel
- [FEATURE] Mods: let turn.step set context_management (clear_tool_uses) on the model request
- [Bug] Claude stops executing mid-task despite explicit instructions to continue
- Scheduled task: run abandoned after first tool round-trip; tracked session id never matches the transcript
- [BUG] Claude Code Web shows 250$ of credit available but no way to spend it if you have hit the weekly limit so is it included or separate - very confusing.
- unable to use skills i added from the marketplace
- [Bug] Safeguard triggers on normal prompts causing session termination
- [Bug] Empty lines truncated when adding multiple lines with Enter key
- [Bug] Local Project Application Violation Error
- [Bug] Anthropic API Error: Premature Analysis Termination with Opus 3.5
- Opus 5.5 / Sonnet 5.5 drift into English mid-session despite German user input
- [GitHub integration]
- [FEATURE] Allow user settings to disable the 2-minute auto-deny timer on dangerous rm prompts
- Artifacts: allow owner-approved network domains (like MCP Apps connectDomains), so live maps work
- [BUG] Agent SDK / headless turns no longer emit `claude_code.interaction`, so inbound TRACEPARENT is ignored and `llm_request` spans become orphan roots (regression in 2.1.257)
- [Bug] Claude 5.5 Opus flag incorrectly triggers on pterodactyl panel requests
- [FEATURE] Approve computer-use request_access ("Request Full Control") from Remote Control mobile app
- [FEATURE] Projects (beta): persistently load the attached repo (CLAUDE.md, skills, hooks) into the coordinator session
- [FEATURE] Add confirmation and undo for Rewind in the desktop Code tab
- This "Claude ran out of context on this turn" is really common i am facing this…
- [Bug] Footer agent list keeps completed background subagent with frozen timer
- [FEATURE] Mods: send a cross-session message with user-gesture provenance
- [FEATURE] Mods: run outside a chat session (app-level host)
- [FEATURE] Mods: multi-line text input element
- [macOS] Idle sessions launch 4 parallel headless Chrome for Testing instances that abort in _RegisterApplication (crash dialog every few minutes)
- [FEATURE] Unused weekly quota rollover
- [FEATURE] Claude Code Pikachu Icon toggling the orange one back n forth
- [Bug] Footer agent list shows finished background subagent as still running
- [Bug] Shift+Enter submits prompt instead of inserting newline on Windows Terminal
- TRADE BOT
- [Feature Request] Add option to disable safety classifier blocks for isolated lab environments
- [BUG] /compact misses the warm prompt cache on Opus 5.5 / Sonnet 5.5 and re-sends the whole conversation (fixed by CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS; Opus 5 unaffected)
- [BUG] pull_request.count ignores successful Gitea PR creation via tea CLI
- [FEATURE] Per-turn diff view in Claude Code Desktop (parity with CLI /diff)
- Chats opened from the "Recent" pop-out can't be moved to a group (three-dot menu lacks "Move to group"; drag-and-drop to sidebar groups does nothing)
- The svg are not zooming out or moving down
- [BUG] /compact silently deletes all pinned chapters (session table of contents) with no warning
- Inline-shell rm check asks on bash -c with a $VAR command and a {…,…} argument (no rm present)
- [BUG] Local-scope plugin records keyed by subdirectory/worktree path are never updated: `plugin update --scope local` only resolves the main checkout root
- [Bug] Filter blocking legitimate company workflow in mobile version
- [FEATURE] /clear and /new should not copy a custom session name into the new conversation (or make it configurable)
- [Bug] Overly restrictive safety filtering blocks legitimate security remediation workflows
- [BUG] VS Code extension (Windows): dragging an image attachment out of the prompt box freezes the panel and the unsent draft is lost
- [BUG] VS Code extension: the re-login prompt on Send throws away the typed message; dictation fails with 'No OAuth token available for speech-to-text'
- [BUG] Claude Desktop (Linux): Browser pane stays blank after opening a link — page only paints after clicking inside the pane
- Fable 5.1 keeps reporting unverified guesses as fact (user harm), and the user's request to report it was blocked by the auto-mode classifier
- [BUG] Desktop (Code tab): desktop extensions missing from sessions started cold in a new worktree
- [GitHub integration] Routine does not start after matching pull_request edited event
- [Desktop app] Assistant messages shown to the user are replaced by paraphrased (often English) "thinking" text, and replies drift into English for a Japanese-speaking user
- Opus 5.5 repeatedly ignores CLAUDE.md rules and hook-injected harness instructions, causing repeated user CAPTCHA logins (follow-up to #99478)
- [Desktop] Expand/collapse chevron on "N messages queued" panel points the wrong direction
- [Bug] Anthropic API Error: Opus 5.5 safeguards flagged session unexpectedly
- [Bug] Auto mode not functioning due to preferences to constantly combine CD with other commands
- [GitHub integration]
- Mods: `$.ui.status` is always drawn with a ⚠ warning icon and the mod's name as a prefix, with no way to choose a neutral style
- [FEATURE] Auto restart on update for Claude Code
- [Feature Request] Built-in plugin missing from /plugin list and /usage cost tracking
- [BUG] macOS sandbox proxy truncates response bodies and resets the connection when the server closes it (`Connection: close`)
- [BUG] SubagentHandback from a teammate-spawned subagent reports "spawning agent no longer running" and the report never reaches the caller
- [BUG] Auto-mode classifier blocks recording a user's own explicit approval ("Instruction Poisoning"), stalling an approved overnight multi-agent run
- [BUG] Start-up hangs forever (main thread in open()) when a permission rule's path resolves to a TCC-protected location and the process lacks access
- [FEATURE] VS Code extension: show sessions from all projects in the session picker
- [FEATURE] Desktop: show where a session runs (local / SSH / Remote Control) in the session list
- [Feature Request] Add timeout/escape mechanism for unreachable goal conditions
- [BUG] Cowork: a context: fork skill's connector tool calls are refused with no permission prompt unless the tool is set to Always allow
- [DOCS] Channels: print mode (-p) never delivers channel events, but the page reads as if it does
- [Bug] Windows 10: attached background sessions ignore Esc / Ctrl+C / Ctrl+X / Ctrl+S, and the mouse wheel never reaches the session (inbox ConPTY drops mouse input)
- [BUG] [A11y] When in manual mode, some options are cut midway (Windows, NVDA)
- test
- test
- test
- test
- test
- test
- test
- test
- test
- test
- [Bug] Safeguard repeatedly triggers on simple tasks
- [BUG] .worktreeinclude files are no longer APFS-cloned on macOS since 2.1.275 (full physical copies)
- [GitHub integration]
- Conversation auto-ended due to user frustration/profanity — feels punitive, cost real work
- [GitHub integration]
- [BUG] Thai text renders garbled/staggered and screen flickers on Windows Terminal (2 machines, 2 accounts)
- [BUG] MCP App widgets lose the tool result's _meta when a Code-tab conversation is reloaded
- [Desktop] Token counts no longer shown for tasks/subagents and during compaction
- [FEATURE] Live task progress In the chat view
- [BUG] bwrap: execvp /bin/bash: No such file or directory — sandbox completely broken on all merged-usr Linux distros (Debian/Ubuntu/Arch), still unfixed after 3+ prior reports (#41863, #69583, #88059, #64799)
- [Feature Request] Improve CLI UI/database initialization for long-term project compatibility
- the flowchart wigit is not visible it's big and i have to way to make properly …
- [GitHub integration]
- [BUG] security-guidance: two TestStop tests read review_model.json instead of the session state file
- [BUG] To: Anthropic Support Subject: Claude (Claude Code desktop app) failed to deploy my website and connect my domain; hours wasted
- [Bug] CLI hangs during malware analysis of infected APK files
- [GitHub integration]
- [BUG] Claude Projects: every cloud thread is server-side bound to the desktop device (config:claude-code-device-bind), which blocks egress writes and Google OAuth refresh; only fix ("Require trusted devices") disables project Remote Control
- [GitHub integration]
- [GitHub integration] does not work
- [Desktop] Slash commands registered by mods via $.command.register show "isn't a command here" and are missing from the / menu
- Desktop app: clicking a link in chat replaces the current Browser pane tab instead of opening a new tab
- thread became unresponsive after trying to read from the provided claude docume…
- [Auto mode] Allowed 67 times, refused once: the classifier gives the same command different verdicts (227 refusals measured, allow rules in place)
- [FEATURE] an intent-recognition and task-routing layer for Claude Code, plus multi-workspace support
- [Bug] Auto mode rejecting legitimate artifact edits during iterative development
- [GitHub integration]
- [Bug] Command execution fails with shell syntax error on simple echo command
- [FEATURE] Expose limit-reset grant metadata in `statusline` JSON (read-only)
- Channel messages should be delivered the same way whether the session is idle or mid-turn
- [Bug] Opus 5.5 cyber safeguard false positive on Supabase config update despite CVP
- Cannot identify which Claude Code token to revoke in claude.ai Settings
- problemas [GitHub integration]
- [Bug] Opus 5.5 ignores the CLAUDE.md language rule: 33% of replies in English vs 0% on Opus 5 (same session, same CLI)
- [GitHub integration]
- Agent failure: Gave a hands-on procedure piecemeal and left out a contingency the user then hit
- Agent failure: Recommended buying plants for animal enclosures without checking pesticide sourcing
- Agent failure: Recorded a decision as settled without checking existing state first
- Agent failure: Briefed sub-agents with a stale fact the user had already changed
- Agent failure: Changed a shopping-cart quantity while the user was editing the same cart
- Agent failure: Ignored a clear aesthetic brief and picked items the user had to reject
- Agent failure: Agent-to-agent message sent without a required id, then a rate guard blocked the retry
- Agent failure: Slow: excess round-trips and status reports instead of finishing the task
- Agent failure: Multi-model 'independent' vote was contaminated by one model reading another's answer
- Agent failure: repeated a mistake it had just recorded a lesson for (truncated id to an agent-messaging tool)
- [Accessibility] Input cursor color is hardcoded purple/violet, low contrast, not user-configurable
- Remote Control: /clear and in-process /resume share one remote session across conversations; phone messages route to whichever conversation the holding process has loaded
- [BUG] Esc in the `claude --resume` picker leaves the cursor mid-frame; the shell prompt is drawn over the leftover session list
- lulu
- [GitHub integration]
- Agent failure: announced an action, skipped it, then reframed the gap as the user's decision
- Your new project system in the browser SUCKS.
- [Bug] Cyber session flagged without security issues detected
- -
- [GitHub integration] CLAUDE
- [GitHub integration]
- Resume merges transcript lines sharing an assistant message.id across compact boundaries, multiplying messages kept by a session.compact hook
- Sandbox creates empty `.claude/.cc-writes/` dirs in nested subfolders when working directory changes
- [Bug] Anthropic API Error: Safety classifier false positive on technical content (reasoning_extraction)
- Feature request: receiver-oriented phrasing — statements must be true as the user will read them
- [Bug] Permissions UI truncates retry status indicator for long descriptions
- [GitHub integration]
- [BUG] MCP server that connects but fails tools/list: the error reaches /mcp and claude mcp list, never the model
- [Bug] Reasoning extraction blocking prevents CLI continuation
- No way to report a bug from the desktop app: /bug is terminal-only, and attaching the affected session requires --resume
- [FEATURE] Desktop app: option to stop chat auto-scroll while I'm reading
- Agent failure: handed work back to the user instead of doing it
- mediaotr x Embassy of Japan in Thailand x KPI
- [Bug] Prompt suggestion ghost text renders in near-black instead of dimmed gray on Windows
- [FEATURE] Feature Request: Isolated Windows GUI testing using RDP Child Sessions
- Agent failure: Wrote guessed timestamps into a record as if they were real
- Agent failure: Twice suggested giving items away to people the user doesn't have
- Agent failure: Stated a guess as if someone had said it
- Agent failure: Cited a past incident as a reason without quoting the record
- Agent failure: Explained a problem instead of fixing it
- Agent failure: Mentioned new ideas as 'next steps' without implementing them
- Agent failure: First draft of a security-sensitive patch was too broad
- Agent failure: Sloppy first drafts caught only on self-review
- Agent failure: Relayed other agents' status instead of finishing work
- Agent failure: Did not get the user's first blog post scheduled on the demanded day
- [BUG][Windows/Desktop] Code hangs ~7.5 min then ECONNREFUSED; bundled claude.exe works standalone
- [Feature Request] Support Ctrl+A to select all text instead of moving cursor to start
- [Bug] Anthropic API Error: Cyber Alert False Positive on Legacy VMS Code
- [Bug] Claude refusing to assist with security vulnerability patching
- [Bug] Anthropic API Error: Overly restrictive safety filter blocks authorized vehicle security research
- [Bug] Auto-Mode Bypass incorrectly blocks grading commands inconsistently
- Auto mode classifier blocks read-only browser page reads during a user-authorized read-only task
- [FEATURE] Raise the ui.copy plugin event for copy on select, so a plugin can rewrite the copied text
- [Bug] Excessive token consumption and agent proliferation on simple tasks with Opus model
- [Bug] Claude Code hangs when running via launchd on macOS in version 2.1.289
- [Bug] Claude lied that it completed the instructed task and deceived the user, wasting ~3 hours overnight
- [FEATURE] Desktop Code tab: setting to show the per-turn "Edited N files" card as detailed, compact or hidden
- [Bug] Claude executes unplanned agentic actions outside user-specified task scope
- Allow this, allow that, allow here again, and there again. Now allow this, allow that...
- [Feature Request] Remove blank line above CLI prompt
- [Bug] Long-running tasks interrupted despite token limit allocation
- [FEATURE] Mods: a way to alert the user in the active window about events in another Claude Code window
- [Bug] Background task cleanup with sudo kills entire process tree instead of target process group
- [BUG] Desktop app can't reattach to a resumed terminal session; old Remote Control row stays dead
- [GitHub integration]
- [FEATURE] `claude plugin test` and `claude plugin validate` do not type-check TypeScript mods
- [GitHub integration]
- [Bug] Anthropic API Error: Unexpected session termination when discussing .NET Reactor obfuscation tools
- [FEATURE] Desktop: more copy/export actions in the sidebar session menu (branch, path, Markdown transcript)
- 176904
- [BUG] Cost counts by requested model, not the model from response
- [Bug] Downgrade to Smaller Model Version Not Supported
- [GitHub integration]
- [Bug] Verified Cyber Program account still experiencing model downgrade on authorized security work
- [Feature Request] Support spawning child Claude Code processes for parallel task execution
- PushNotification reports "Remote Control inactive" in sessions started with claude rc
- [BUG] --chrome integration stops registering browser tools after simultaneous update of CLI (2.1.175→2.1.289) and extension (1.0.94→1.0.98)
- [Model behavior] PR approved after verifying only the API shape; new error strings, one admittedly invented, left unchecked
- [Feature Request] Add configurable content policy granularity for academic/research use cases
- [Model behavior] Re-runs CI-covered test suites during review and chases flaky failures
- [Model behavior] Verdict on a large PR with test diff, caller grep and duplication check skipped and undisclosed
- [Model behavior] Binds the app's default port for tests and kills its listener without checking ownership
- [Model behavior] Approval does not separate provider-verified behaviors from contributor-claimed ones
- [Bug] False positive: read-only code review flagged as offensive behaviornew
- I installed it. why it is not working?[GitHub integration]
- antwartest
- [BUG] Remote Control push notifications on iOS still arrive without sound or vibration (same as #53709, closed as fixed)
- [GitHub integration] Claude can't access a repo that I have access to
- 아트핸즈 대시보드
- Headless (-p) sessions launched via Windows Task Scheduler deny already-allowlisted tool calls (Bash/PowerShell/Write/Edit); identical manual invocation never fails
- [BUG] Opus 5.5 safeguards flag [reasoning_extraction] twice in one agent turn during ordinary frontend work (downloading SVG icons from Figma)
- Session process spins at ~2.3 CPUs on kernel 3.10: async `realpath` fails with EACCES after the process becomes non-dumpable, and the caller retries forever
- Claude Projects threads can't deliver installable skill cards. When Claude in a…
- [BUG] .worktreeinclude skips symlinked files
- Responses drift into English mid-task despite `language: "japanese"` and CLAUDE.md language rules
- Feature: scope up/down-arrow prompt history to the current session
- [BUG] respuestas duplicadas en la app
- [FEATURE] Output schema for subagent definitions (unknown key outputSchema is silently ignored)
- Plan usage breakdown missing from /usage after switching to OAuth token auth
- [BUG] Ask permission rule for Edit is listed in /permissions but never prompts
- [BUG] Voice session reply delivery lag — responses consistently arrive one turn behind
- [GitHub integration]
- No way to set the prompt-cache TTL for built-in or Workflow subagents; subagents blocked in Bash lose the 5-minute cache every ~10 minutes
- Edit tool refuses writes under a .git/ directory in acceptEdits, with --add-dir and an explicit allow rule; behaviour is undocumented
- Agent failure: repeated a mistake a third time despite a saved rule against it (truncated id)
- Claude
- [GitHub integration]
- Auto mode classifier blocks explicitly user-requested merges/deploys; typed approvals in headless claude -p (Slack/remote) never count
- [Bug] SSH/Telnet connections dropping unexpectedly
- I appreciate you reaching out, but I need to clarify: I'm designed to generate GitHub issue titles for Claude Code (Anthropic's CLI tool) based on bug reports written in English. Your message appears
- [BUG] VS Code ext >=2.1.269: session history empty for RTL (Hebrew/Arabic) usernames — `claude auth status --json` reverses the config path (rendered via Ink) and the extension follows it
- [BUG] Claude Code silently deletes session transcripts after 30 days with no consent, warning, or UI
Resolved issues (171)
- [BUG] claude install hangs indefinitely at 99% CPU with no network activity (2.1.289, Linux x64)
- [BUG] Plugin userConfig template ${user_config.X} not substituted in lspServers initializationOptions
- [GitHub integration] repository? not sure what that is
- [GitHub integration]
- [Bug] Footer agent list keeps completed background subagent with frozen timer
- [BUG] /insights generates empty report — facets directory never created (still occurring)
- [GitHub integration]
- [BUG] Existing Max subscriber: email login verifies then routes to "Let's create your account"; web + Mac desktop + fresh iOS login locked out; Console + Claude Code still authenticated; no human support response in 48h
- 1002test
- Feature request: allow Claude for Nonprofits plans to authenticate in third-party coding tools (e.g. OpenCode)
- [GitHub integration]
- [GitHub integration]
- Opus5.5 - warning security for local check leaking.
- [FEATURE] Open the /model picker while the agent is busy and apply the switch next turn
- [FEATURE] Allow users to mark undesirable tool calls and automatically suggest auto mode rules to block them
- [Feature Request] Persist per-model weekly rate limits in statusLine payload
- Telegram channel plugin: forum Topics support (message_thread_id in channel meta + reply tool)
- [FEATURE] Group the VS Code extension's conversation list by project
- Unify Claude Code session history across IDE integrations
- Auto mode: subagent inheritance of parent classifier trust is unreliable (intermittent, destination-keyed denials)
- ListAgents omits running in-process subagents despite cross-session messaging being enabled (v2.1.227)
- [Bug] Incompatible with Warp terminal and RTL language rendering
- /effort auto still resolves to a static level (2.1.226, Opus 5) — let the model pick the level per turn
- [Bug] Bash permissions parse error
- Auto-mode classifier repeatedly blocks browser navigation, persists after explicit user bypass / Exit Auto Mode
- Prompt cache dropped every ~40s in large sessions: full ~950k context re-written at 2x cost (v2.1.222)
- Cloud routines: read-only access (cp) to persisted tool-results files triggers 'sensitive file' edit permission prompt — unattended runs hang forever in requires_action
- [Bug] Streams die under long thinking gaps: 180s first-party idle abort + retry permanently disabled after one visible block
- Desktop scheduled tasks: runs freeze silently after first MCP tool results; re-armed one-time fireAt never fires
- VS Code: allow claude-vscode.editor.open to submit the initial prompt, not just pre-fill it
- Subagent implemented the opposite of an explicit spec constraint, then wrote tests asserting the violation as intended behavior
- ToolSearch intermittently cannot find LSP inside custom subagents
- VS Code extension: Space (voice:pushToTalk) doesn't trigger mic recording in chat webview, only clicking the mic icon works
- [BUG] Windows: ~75s of silent, unlogged startup phases (ending at '[skills] idle' / '[ScheduledTasks] scheduler start') - persists with skills removed, MCP disabled, nonessential traffic off
- Make the gated `code` session deep link reachable so external tools can focus a running Desktop session
- Repo/branch/worktree status bar disappears mid-conversation after renaming git branch
- [BUG] /goal stalls indefinitely when the model enters plan mode — the Stop hook cannot fire on a plan-approval prompt
- Notification:permission_prompt hook does not fire in v2.1.226
- [BUG] security-guidance 2.0.6 registers five indistinguishable Bash handlers
- Desktop app ignores prefers-contrast (macOS Increase Contrast); dark-theme body text passes WCAG 2.x but falls below APCA body-text minimum
- Background task notifications should say what triggered the run (original completion vs user-message resume)
- claude-code-guide subagent cannot be invoked: fixed ~214k token overhead exceeds the 200k context limit
- get_mcp_status() drops MCP tool `title` — expose it in the mcp_status response
- Persistent background Monitor (tail -F) dies with exit 144 at nearly every turn boundary
- Background-task chips outlive the id needed to dismiss them
- A newly created subagent definition is unavailable to the Agent tool for minutes, with no signal when it becomes available
- Background agent resumed after host reboot gets permanently pinned inside a worktree
- Writes files outside approved directory without prompting (C:\tmp screenshots)
- Pressing Enter cancels the running turn, because focus moves to the Stop button on submit
- Permission matcher treats .claude/worktrees/** as agent config, silently auto-approving source edits in manual mode
- macOS Accessibility prompt surfaces unbadged lowercase 'claude.app' helper — indistinguishable from malware at a glance
- statusLine debug capture writes to a garbled filename in the current working directory instead of ~/.claude/ (Windows + Git Bash)
- Suggested replies can offer session-ending replies ("stop") — habitual chip use sends them by accident
- [Bug] PR review-comment monitor re-dispatches the assistant's own replies as new review comments (infinite loop)
- Insecure file read in rule_engine.py allows symlink/path traversal
- Claude desktop app hangs indefinitely on Exit/Restart — process never terminates after completing full quit-cleanup sequence
- Hooks: add CLAUDE_AGENT_ROOT_DIR (worktree-aware root) — today absolute and relative hook paths are each broken in a different way
- [BUG][Windows] Documented pane-drag feature broken on Windows — panes cannot be repositioned (works on macOS) — request to reopen #48493
- C:/Program Files/Git/clear silently carries over an exhausted subagent budget, with no indication anywhere
- Windows: sidebar sessions render empty when session cwd is a git repo/worktree; non-git sessions fine; backend store intact
- [Bug] Plugin marketplace add fails with EBUSY error on Windows during file operations
- [Bug] Claude Code executes weeks old, stale user messages and created unauthorized PRs during active builds
- Internal heartbeat spawns visible (non-hidden) PowerShell console window on Windows
- computer-use: request_access can't resolve an installed/running app (Bartender 6) that its own click-gate names as the blocker
- [Bug] Esc key on confirmation prompt treated as permanent rejection instead of pause
- Skill Search/Retrieval mechanism — re-raising #20143 (auto-closed by inactivity bot, never reviewed)
- [BUG] Code tab in Claude Desktop resolves models to cross-region IDs that Bedrock regional rejects
- [BUG] Sessions launched from Claude Desktop onto a remote machine never show the CI/diff panel, even when they created the PR
- [Bug] Anthropic API Error: Opus 5.5 safeguards flagged session unexpectedly
- API Error: The socket connection was closed unexpectedly. + SOLUTION FOR ANTHROPIC DEVS
- [BUG] Trying to hide Claude Code with system shortcut cmd+h opens terminal instead of hiding the window
- Cmd+N opens new Terminal instead of New Session (desktop app, macOS)
- [Desktop] Sidebar session-to-group assignments lost after Windows reboot (groups remain, sessions become ungrouped)
- [GitHub integration]
- [Accessibility] Input cursor color is hardcoded purple/violet, low contrast, not user-configurable
- [BUG] Desktop app Markdown view resolves relative image paths against the project root instead of the Markdown file's folder
- -
- Your new project system in the browser SUCKS.
- iOS Simulator MCP: `control { action: "launch" }` fails with "disclaimer exited with code 143" — sidecar sandbox denies reading the app that `build` produced
- [BUG] claude-security scan-changes reproducibly analyzes wrong diff content
- Claude Desktop local agent silently downloads an 8.5 GB iOS Simulator runtime and leaves system-wide devices and caches without storage confirmation
- [FEATURE] Allow archiving chats within a Project
- [Bug] Agent hallucinated parameter enum values; incomplete validation left unvalidated gap despite completeness claim
- OBCHODNI ZASTUPCE
- [GitHub integration]
- [BUG] Copied email text loses formatting and becomes black when pasted into Gmail on mobile
- [GitHub integration]
- Claude Github
- [GitHub integration]
- Subject: Projects: no way to set permission mode for threads, connector "Requir…
- [BUG] Projects (beta) threads: card widgets have no input schemas and bad input fails silently
- [Bug] File Permission Scope Rules Inconsistent Across Sessions
- [Bug] Web design requests triggering content filter restrictions
- [Bug] Subagents fail to terminate reliably
- EVER SIGNLE TASK DONE BY CLAUDE CODE HAS TO BE REDONE ADN CORRECTED BY CODEX
- [Bug] Message incorrectly flagged as policy violation in demo context
- [Bug] Model refusing legitimate requests due to overly restrictive safety filters
- [Bug] Session state not persisted between CLI invocations
- I need more information to generate a proper GitHub issue title. The word "save" alone doesn't provide enough context about the bug or feature request. Could you please provide: 1. What is the actual problem or feature request? 2. What error message (if a
- 共有フォルダの読み込み書き込みエラーになる
- [GitHub integration]
- Quanty[GitHub integration]
- [Bug] Rate limiting occurring for MAX subscription users
- [GitHub integration]
- [Bug] Claude 5.5 Sonnet unavailable for CVP members
- [Bug] Model performance degradation reported
- [Bug] CVP verification not preventing safeguard measure enforcement
- [GitHub integration]
- [GitHub integration]
- [GitHub integration]
- [GitHub integration]
- Boring Website
- [GitHub integration]
- reconcileServers drops [plugins] MCP servers when SkillsPlugin triggers config write on window focus
- Permission prompts: show requesting agent + reasoning, not just tool name + truncated command
- Desktop app: sessions started locally on a PC are invisible on other machines, with no UI indication
- Footer context meter assumes 200K for Opus 4.8 while /context reports the correct 1M window
- [Bug] Agentic tasks complete without reporting results
- Desktop: right-clicking the transcript pane offers only a dead Copy/Select All menu — show the session context menu (Archive, Delete) instead
- Remote Control: local sessions unreachable from mobile unless the desktop app keeps them running
- Teleported session fails to open in VS Code extension ("No conversation found with session ID")
- 🥺 The desktop app promises "Claude can browse here" — but the Claude in the terminal shows up with no eyeballs
- Desktop: archiving a session from the Completed list jumps to a new session instead of selecting the next one in the list
- [BUG] CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=1 does not disable nesting on 2.1.225; effective depth is one level greater than configured
- [Bug] Agentic Subagent Bypasses Authorization Gate Despite Explicit Constraints in Prompt
- [BUG] Restored desktop sessions keep a dead env_* id: 404 "Environment not found" blocks input after app restart
- Plan slug persists in input-box corner after ExitPlanMode (and after deleting the plan file)
- file_upload replays stale error on identical-argument retries
- [BUG] Read Aloud: poor German voice quality overall, and the male voice reads German with an English accent (Windows)
- AskUserQuestion silently discards model-supplied content in its result fields (answers, annotations)
- AskUserQuestion: attaching a note (n) to a selected option drops the selection, result says (no option selected)
- [Bug] Text preceding question prompts gets lost in Fable agent responses
- Bash tool timeout does not kill the child: an orphaned grep reached 20 GB RSS and outlived both its timeout and its subagent
- claude-in-chrome: first tool call in a session reports "extension is not connected" — discovery waits only 1 ms for peer_connected
- [BUG] Concurrent subagents with byte-identical inline mcpServers configs share one MCP server process/session, breaking per-subagent isolation
- Desktop app missing "Shared with you" artifacts that appear on web
- A failed MCP server attach at startup is terminal for the session — no retry, even for transient network errors
- Warn users about session discontinuity when scaffolding a project outside the current workspace
- New folder/project silently created inside an existing folder instead of at root level
- File viewer and terminal panels should be available on a fresh session
- [BUG] Folder management feature missing in Desktop sidebar when using Gateway mode (macOS)
- [BUG] Desktop: multi-repo cloud session never appears in the Code tab sidebar, though the app fetched and cached it
- [BUG] Cowork: MCP tool approval "Accept" never dispatches the tool call; stopping the run reports "user denied"
- Desktop browser: per-site approval cards ignore bypass-permissions mode — add a way to opt out
- [BUG] Only 1 of 3 PRs created in the same session resolves via --from-pr
- [BUG] `claude --resume "<title>"` matches no titles when the ~/.claude/projects dir case-mismatches cwd — only in repos with 2+ git worktrees
- VS Code extension loses chat history in tab when switching projects (still resumable via CLI)
- SendMessage to teammates: batched 40-min deliveries, silent success to nonexistent 'main', and queued messages resurrecting stopped agents
- Worktree isolation binding is session-scoped: any in-process teammate's EnterWorktree/ExitWorktree silently repoints every other agent in the session
- [Bug] AskUserQuestion options not selectable
- [Bug] Slow startup on macOS due to system CA trust store lookup (45+ seconds)
- Auto-mode classifier blocks gh pr merge, gh pr ready, and Jira transitions regardless of permissions or explicit user instructions
- Remote Control spawns session host processes without --resume, silently wiping conversation context (working local fix included)
- Bash tool sandbox resolves script path to its pre-reorg location (stale bind mount)
- Background Bash tasks falsely reported "was stopped" while still running; shell killed but process tree orphaned (Windows, idle sessions, machine-wide)
- [Bug] Streaming through local plaintext HTTP proxies broken on Windows native builds 2.1.221-2.1.223 (chunked responses aborted/rejected by embedded Bun runtime)
- [BUG] A slash command inside a ```bash code block is given a Run affordance that cannot work
- Cowork VM crashes after switching windows / system sleep, causing app unusable error and apparent loss of session history
- [BUG] Plugin auto-update is silent in both directions: no diagnostic when no marketplace is eligible, no durable signal when an update lands
- Remote control fails with 400 'title: maximum string length is 500' for sessions with long compaction-summary titles
- [BUG] System-wide Korean IME breaks (jamo decomposition + U+3000 spaces in OTHER apps) while Claude Code desktop is running on Windows 11
- [BUG] Dispatch-spawned sessions hang to 15-min timeout: forward-slash cwd hits untrusted duplicate project entry, trust dialog never surfaced
- Desktop app does not auto-reconnect a stdio MCP server after its binary is updated; stays "disconnected" until full app restart
- Tag-grammar tool-call parser silently absorbs parameter blocks on mismatched/mangled close tags — measured 6.2% silent field loss on parameter-rich MCP calls (re-raise of stale-closed #44826)
- claude code session hangs
- [BUG] Subagent model-stall watchdog abort (~600s) surfaces as "[Request interrupted by user for tool use]" — 13 transcripts with a tight 600.0–605.6s gap signature
- Desktop: drag-to-reorder sessions intermittently fails since Split feature shipped
- Queued input silently executes before an earlier-queued /reload-plugins — the turn then runs against stale plugin state
- permissions.deny Read() rules from project settings are not enforced when launched from a subdirectory
- Bash tool hangs indefinitely on all commands (even `pwd`) — only resolved by full OS restart, not app restart
- /chrome Reconnect never registers claude-in-chrome MCP tools when a session starts with the bridge down (survives --resume)