Digest / October 11, 2026
Claude Code Issues Digest — October 11, 2026
256 new issues filed ·
228 resolved in anthropics/claude-code (UTC day).
New issues (256)
- [Bug] Linux saveImage ignores JPEG clipboard images (only tries PNG and BMP)
- /feeback doesn't work in Projects in Claude Mac App
- Desktop: archiving an SSH session does not remove its clean worktree
- cannot read the closed holder
- /design is user-only (disable-model-invocation), but the agent isn't told it exists or that the Design canvas is reachable another way
- [BUG] Mods tool.call: PowerShell results never carry isReadOnly (Bash does)
- npm auto-update during network outage replaces working install with 500-byte stub
- Prompt is too long
- API receives wrong client version (2.1.133) despite running 2.1.296 — blocks model selection
- [FEATURE] "fork" means three things, and the skills docs now say "Despite the name"
- [Bug] Bash tool calls hang before dispatch on Windows desktop; Stop ignored until claude.exe killed
- Linux sandbox leaves a read-only .git/config.lock on the host, breaking git config writes in sibling worktrees
- **Request: let cloud Claude Code sessions use GitHub Projects (v2) for a connec…
- [BUG] Gmail connector stops returning mail addressed to a Workspace alias after a date; the Gmail app still shows it
- [BUG] Write drops an existing file's BOM, CRLF line endings and encoding (UTF-16LE, Windows-1252, Shift_JIS), while Edit keeps them or refuses
- [GitHub integration]
- Worktree-isolated session: `/cd <repo root>` makes every later Bash call refused as "working directory resolved to the shared checkout", and the refusal does not name the recovery
- [BUG] Desktop app: side pane arrangement (side by side vs stacked) resets after switching chats
- Clau [GitHub integration]
- [FEATURE] Mods: let a session.append hook redact or drop any block of a tool-result row (some blocks are restored by the engine)
- JetBrains (PyCharm/IntelliJ): add an official graphical chat panel
- pageant_system
- Desktop app: AskUserQuestion box steals the Enter key while I'm typing in the chat input — question answered unseen, message lost
- [BUG] Desktop (macOS): main process stalls ~2.9 s while typing with Apple Zhuyin IME — IMK XPC deadlock in utSetupInputMethodMenu (menuWithCompletionHandler ↔ IME selectedRange)
- [FEATURE] Plan-approval dialog hides auto mode whenever bypass permissions is available
- [BUG] ${CLAUDE_EFFORT} in a skill preloaded into a subagent expands to the session effort, not the subagent's effort
- -
- [MODEL] Asked to write a GitHub issue, the model published it under the user's account without showing the draft or asking
- [MODEL] Claude launched long-running batch jobs on its own during an analysis-only request
- [BUG] iOS app shows Live Activity on Dynamic Island even with Live Activities disabled
- [MODEL] Wrote a command's output to a temporary file, against the user's rule that working text never goes to one
- Bug Report: No defect; positive feedback on playlist curation
- [MODEL] Told the user where two variables were used after searching only some directories of a repository
- [FEATURE] [Linux] Please add fedora and arch repos
- mods eelden
- mods elden ring
- Projects: coordinator reports idle-recycled local thread as "Mac offline" though the Mac is online and the thread resumes in seconds
- [BUG] Manual /compact emits empty system message and breaks resumed sessions in 2.1.296
- [BUG] `claude agents --json` leaves suspended `git ... remote` children behind (Windows)
- [MODEL] Told the user that a shell function it wrote runs nothing after a semicolon, having tested one plain case
- [GitHub integration]
- Unable to approve a pending question from Claude.
- DOKY DORY
- a
- [BUG] Rewind said it couldn't restore, but still overwrote ~20h of edits across the project and deleted files created since (conversation not rewound)
- [BUG] Opus 5.5 safeguards refuse 3 messages in a row in a harmless tutorial-writing conversation (model_refusal_no_fallback)
- Remote Control: session shows "environment was deleted" after server exit; no way to re-attach to a new environment
- Remote Control: make the ~4h session resume window configurable (or allow resurrecting expired sessions)
- [MODEL] Handed the user a commit script without running its gate, carrying a value from an outdated memory that the gate refused
- [FEATURE] Desktop: coordinate heavy local work across concurrent sessions instead of letting every session build at once
- [BUG] Desktop Browser pane (macOS): no Passwords app / iCloud Keychain autofill and no passkey support
- [BUG] Desktop Browser pane: popup windows are not supported, so OAuth/payment/popup-only sites cannot be used
- [BUG] Login redirects to account creation for existing account with active Max Suscription
- [Feature Request] Allow transparent/default background for composerSidebarBackground theme token
- You've hit your weekly limit · resets 2am (Asia/Damascus)
- [FEATURE] Desktop: manage worktrees by lease and state, not PR-merge events (long sessions span many PRs)
- [BUG] Mouse clicks don't work in TUI until entering agent view and returning to the session
- [MODEL] Ran mktemp in its own command, against the user's rule, and left the directory for the user to remove
- [Bug] Auto mode classifier denying routine code edits, blocking backend DTO/controller updates
- [BUG] Claude Code VS Code extension 2.1.296, Remote (vscode-server) on Linux, model claude-opus-5-5
- [GitHub integration]
- [BUG] MCP OAuth: resource parameter is double-encoded when the server URL has a query string (Supabase returns 422)
- [Bug] Model refusing legitimate professional tool requests due to overly restrictive safeguards
- Light theme: prompt suggestion (ghost text) renders in the same colour as typed text; add a theme token to change it
- [BUG] Typing over a selection in the prompt input inserts instead of replacing; Backspace does delete the selection
- [Bug] Auto-compact triggers early at ~89% context remaining
- [BUG] SIGILL (`cntb`, SVE) on linux-arm64 in Android AVF guest
- Opus 5.5 [bio] classifier flags benign in-silico protein-engineering session; blocks file-listing output and /feedback returns 403
- [FEATURE] Add per-request billing classification (subscription vs extra usage) to OTel api_request events
- [FEATURE] Cowork: draw mod UI (band, pane, spinner) like the Code tab does
- [GitHub integration]
- Cloud
- [MODEL] Completed guide work continues with a growing cached context
- [MODEL] A request for cheap checks becomes a plan and repeated large context work
- [MODEL] Conditional archival runs before the required checkpoint review
- [MODEL] A page is delivered without its matching internal companion
- [BUG] CSI u (Fixterms) shifted sequences not recognised
- New Thread requsts weird bot permissions and gets stuck
- [BUG] Mods: `$.prompt.submit` on an idle session sometimes never enters the queue and never settles, then is dropped when the user sends a message
- [MODEL] An install plan omits available disk and invents another approval gate
- [Bug] 사이버 보안 필터가 git 최신 소스 가져오기 요청을 오탐하여 차단
- Feature request: show quota, context and model/effort in the default status line
- VS Code shows "browser connected" for @browser while the Chrome extension is signed out
- [BUG] Desktop (Windows 10): window stuck maximized for two days, Restore button and Win+Down do nothing; native ShowWindow(SW_RESTORE) fixes it
- / slash-command autocomplete: Down arrow past a large same-prefix block (gortex-*) causes unbounded re-loading
- Claude refuses to disable Windows Defender on the user's own machine even when explicitly asked
- [FEATURE] 作業フォルダの外にあるファイルも、チャット内のリンクから開けるようにしてほしい
- no esya usando mis creditos de la nube
- [FEATURE] Expose plan usage with decimals (status line rate_limits, /usage, usage endpoint), not whole percentages
- [Bug] Agent loop continues indefinitely consuming tokens without termination
- [BUG] Cloud sessions: egress gateway returns "no rule or allowlist entry allows host" for allow-listed hosts since 2026-10-10
- [Feature Request] Add ui.render component for welcome logo to allow mod-drawn replacements
- Desktop app: the pane move handle is nearly invisible – docs say "drag a pane by its header", but only a tiny grip at the top centre of the header works
- [MODEL] Sonnet 5
- [Bug] Zellij integration: scrollback navigation breaks terminal input in Claude Code tab
- [Bug] Unable to determine issue: request ID only (req_011CfvFBXrNDn48nVaAHF71f)
- [BUG] Subagent teardown kills the parent session's stdio MCP server when both declare the same inline server
- [MODEL] Agent ran `docker rm` on a crashed container under a "restart without asking" grant, destroying its only logs
- [BUG] claude.ai personal marketplace: git-subdir/external plugin sources never refresh; Check for updates reports 'latest' while upstream is ahead
- [Bug] Auto mode blocks `--force-with-lease` on the session's own feature branch as [Git Destructive], against the rule's own wording
- [Bug] Claude Code refuses credential validation/audit tasks after security changes
- Desktop app (Code tab): internal terminal loses keystrokes to chat; no way to disable it or send Run to an external terminal
- [FEATURE] Record the reason for each edit, and use it for review and line-level "why"
- [FEATURE] Verification summary generated by the harness (not the model) at the end of each turn
- [FEATURE] Estimate before a task, plus a per-task budget that pauses instead of overrunning
- [FEATURE] Show which actions /rewind cannot undo
- [FEATURE] Suggest memory rules from repeated user corrections
- [FEATURE] Run only the tests affected by Claude's edits after each batch of edits
- [FEATURE] Live, read-only share link for a running session, with comments
- [FEATURE] Pinned, editable plan and checklist that stays visible while Claude works
- VS Code: background subagent tool calls still stream into the main chat (reopening #77326)
- no way to change permissions, to go into don't ask mode when in project threads
- Desktop app never asks a Remote Control session to draw a mod's AbovePrompt band
- He was doing absolute trash, he like got dumb by -200 iq, and refused to connec…
- [BUG] Organization plugin sync strips plugin.json `types`, so function-hooks mods that use $.state fail to load
- Bug Report: Reasoning Extraction Triggered on Non-Bug Input
- [Bug] Opus 5.5 Safeguards Flag Benign RAG Request (reasoning_extraction)
- Bug Report: Backend startup failure
- [Bug] Audit task takes ~15 minutes and produces incorrect results - USELES PICE EOF SHIT CHAT BOT
- the thread seems stuck
- Auto-memory is shared between accounts on one machine, and .claude/launch.json accumulates configs of every past session
- Freeze récurrent (écran blanc) toutes les ~2 minutes depuis la connexion via GitHub
- Bug Report: No clear issue, positive feedback on remote SSH/HTTPS dev setup
- [Feature Request] agents-view should emit prompt.autocomplete and other hooks for Claude Mods
- On WSL with interop disabled, fullscreen copy depends on the terminal's OSC 52 support and ignores the working WSLg clipboard
- I was building an tool which could become an effective measure in finding phish…
- You've hit your monthly spend limit · raise it at https://claude.ai/settings/usage?from=cc_cli_limit_message · your weekly limit resets Oct 13 at 5pm (America/Chicago)
- [FEATURE] Mods API: expose the billing fields of usage on ModelUsage / TurnUsage
- You've hit your weekly limit · resets Oct 13 at 10am (America/Buenos_Aires)
- Assistant text written before a tool call is shown to the user only as a one-line summary
- Opus 5.5 [bio] classifier interrupts benign biomedical literature work (reading published papers, pre-commit secret scan)
- claude no puede abrir el navegador y no puedo darle permisos de nada. es muy fr…
- [Cowork / model behavior] Claude reported a message as "sent" without verifying, then shipped a truncated message
- TUI freezes for minutes: the clipboard-image hint reads NSPasteboard synchronously on the main thread
- Include the messaging ref in `claude agents --json`
- [BUG] Windows/Git Bash: if the 10-second login-shell PATH probe times out, the Bash tool loses PATH for the whole session
- HTTP MCP client gives up after ~15 s of connection-refused; session loses server until manual /mcp
- [MODEL] Wrote its own findings into an approved plan as decided requirements, instead of putting them to the user first
- [GitHub integration]
- [BUG] --worktree runs WorktreeCreate for --version/--help and on failed --tmux attach, never calls WorktreeRemove (leaks worktrees + tmux sessions)
- cant start in app browser even when explicitly asked
- [FEATURE] Agent definitions: optional condition on the main-loop model
- Cant see the approve or deny options, that on all other UI sections it's showin…
- [Feature Request] Show linked PR/GitHub issue alongside subagent task details
- [MODEL] Coded a failure behavior its approved plan does not describe, and told the user only after coding it
- [MODEL] Wrote into a plan that a requirement says what it does not, without checking the requirement
- [GitHub integration] 1ER depot
- Remote Control sessions on my Windows PC end before attaching (“Couldn’t start …
- [MODEL] Told the user a helper looks for a file beside itself, after saying it had checked the code by reading it
- [Bug] Content filter false positives on historical texts
- [BUG] [VS Code extension] ArrowUp during IME composition recalls message history and destroys the text being composed
- [GitHub integration]
- [Bug] Over-restrictive refusal blocks help fixing password reset/login code in user's app
- [GitHub integration]
- [Bug] Opus 4.8 cyber safeguards block benign "verifica" request after model switch
- [FEATURE] Claude Desktop: add "Move to group" to the Recents page (older sessions can't be grouped)
- [BUG] GitHub connector create_pull_request shows a form to the user instead of creating the PR, even when the owner told Claude to act
- [Bug] Spurious security refusal on Monad research project, prompts switch to Opus 4.8
- [Bug] Anthropic API Error: Content policy filter incorrectly blocking legitimate horror website redesign
- [FEATURE] Allow opting out of automatic offline_access for MCP OAuth clients
- [FEATURE] Desktop: publishing an artifact update shouldn't take over the user's view; let decks reopen at slide 1
- Projects: Don't put tasks in Ready for Review until CI has passed or failed, leave them in working till finished
- Terminal images never draw under libghostty-based terminals (Herdr): kitty-graphics check rejects XTVERSION `libghostty`
- [FEATURE] Code tab sidebar: per-group view controls (group by date, sort) on every custom group
- Assistant ran a different tool call than the command it showed the user for approval
- [FEATURE] Supported compaction instructions, and a kick-off prompt after compaction
- [MODEL] Removed four variables from a helper and left its manual page documenting them, across two commits
- agent-memory@inline PostToolUse hook fails on Windows usernames with a space (unquoted hook path)
- [FEATURE] Mods: expose the Desktop text-selection menu as a ui.render component
- [FEATURE] Render mod UI (ui.render, panes) in Remote Control clients attached to a terminal session
- [BUG] statusLine context_window.used_percentage drops to 0 after an API error: synthetic assistant entries carry all-zero usage
- [Bug] Recurring "Waiting for API Response" errors every ~30 seconds, making Claude Code unusable
- Projects - need a better "alert/waiting" notification or flashing icon when a t…
- [FEATURE] Right align user messages in Claude chat in VSCode
- [BUG] scripts/edit-issue-labels.sh crashes with unbound variable under Bash < 4.4 and exits silently on empty arguments
- [Claude in Chrome] Windows: bridge returns empty extensions_list; extension 1.0.100 never connects
- [Bug] Opus 4.5 sessions flagged for compiler obfuscation work (trading marketplace)
- [Bug] Claude fabricating and encoding unverified scope rules into tool commands
- Desktop usage card counts each API call about 2.2x (sums every transcript line, not each unique call)
- Desktop Code tab: dictation no longer shows the transcript live while speaking; the text appears only when recording stops (macOS)
- [GitHub integration]
- [Feature Request] Add direct main thread connection to cc-plugin-you-should-know for automatic issue escalation
- Sessions started in terminal / Remote Control aren't visible in the Claude desktop app (and vice versa)
- Auto-mode permission classifier blocks routine, pre-approved writes under shifting/inconsistent categories — including blocking self-service permission changes
- [Windows] TUI copy blanks the RDP host clipboard: PowerShell 5.1 `Set-Clipboard` overwrites the OSC 52 write
- [BUG] Routines with a specific Repo access can escalate and add another to session
- [MODEL] Handed the user a long command line to paste, against the rule that every command goes into the one script they run
- Thread stuck on a prompt I can't see after I asked it to delete an artifact it …
- [FEATURE] Mods: let a docked Pane span the full height, beside the prompt
- [MODEL] Told the user which records named an old path after checking only three; a fourth named it too
- Remote Control: ExitPlanMode prompt stays on iOS after a PermissionRequest hook approves it
- [Bug] Mythos 5.1 access denied for website audit despite approved CVPexit
- Auto mode classifier denies writing a new README in a local repo the user asked for, no reason or in-line approval
- [Bug] Model behavior degradation: generating illogical code changes
- you gave me 100$ but i am not usage it
- Background subagents stall on permission prompts in bypassPermissions mode (2.1.296)
- Bug Report: Safeguards blocking legitimate requests
- [Bug] Claude Code refuses to work on local files
- Long session: 17 specific failures (unrequested actions on live devices, overnight jobs silently failing, ignoring hard user rules, inconsistent answers)
- [Bug] Claude Code refuses benign Android tablet cleanup and F-Droid install tasks
- [GitHub integration]
- [FEATURE]
- [Bug] False-positive cyber safeguard blocks benign local install request
- [Bug] Cyber Safeguards Falsely Flag Benign Request, Blocking Opus 4.8 and 5
- [Feature Request] Improve Model Support for Scientific Research Workloads
- [BUG] Cloud Projects sessions run at low reasoning effort (15) while session_context.effort_level is “high”
- [MODEL] Stated an unverified cause as fact: said the user's launcher named two sessions, when the user had not run it
- clangd-lsp plugin starts an unbounded --background-index per session; memory grows to gigabytes and cannot be tuned
- [FEATURE] Toggle setting for flipping keybinds: "Enter" for new line and "Shift+Enter" for send.
- [MODEL] Broke a standing rule a third time: six commit scripts never synced the user's three repository clones
- Project cloud sessions not using cloud usage.
- Desktop app: a mod pane goes back to its first column and width after a switch to another session and back
- [Bug] Opus 5.5 refuses to harden user-owned source code as potential hacking
- [Bug] Claude Code Refuses SSH Connection to User's Own Linux Device
- [GitHub integration]
- [GitHub integration] agents brain
- [Bug] Incorrect VRAM increase instructions for AMD Radeon 780M (BIOS/registry)
- [Bug] Orphaned bg-pty-host process prevents session restart after Ctrl+C exit
- [FEATURE] Agent roster view in the Claude apps: named persistent agents, @agent routing, approvals inbox (Grok Bot-style, multi-agent)
- [Bug] Legitimate security research prompt incorrectly blocked
- I don't have enough information to generate a GitHub issue title from this report. The description appears to be about infrastructure maintenance rather than a bug in Claude Code itself. Could you provide more details about: - What specific problem or err
- Bug Report: No issue details provided
- [BUG] Desktop app (Code tab): voice dictation always appends to end of prompt instead of inserting at cursor / replacing selection
- [BUG] Devanagari (Hindi) text corrupts the whole TUI on Linux: stale frames not cleared, lines overlap (v2.1.295)
- Let projects pre-approve device connection for new threads
- [BUG][REGRESSION] Broken Projects upgrade wiped out classic projects with no no opt-out
- [BUG] Desktop Code tab: connector switch is account-wide (also disables the connector in claude.ai chat and Cowork) but reads as per-session
- [FEATURE] Code-only scope for claude.ai connectors, or defer connector tool names until used
- [MODEL] Asked the user to re-decide a behavior its own plan already ruled, and recommended the opposite of the ruling
- [Feature Request] Dismiss finished subagents and auto-prune completed entries from agent list
- Fixar não faz nada em threads de projeto, fixar na barra lateral mas nada acont…
- Desktop app (Code tab): message box stops taking letters until the app is quit and reopened
- [BUG] Windows Desktop: idle/closed Code sessions keep their MCP servers alive — 121 node processes, ~10 GB RAM (regression of #58915?)
- [FEATURE] Mods API: let a tool.call hook add file deny rules to one Bash call's sandbox
- [FEATURE] Mods API: a queryable record per background task, with event order, exit code and progress
- [FEATURE] Mods API: report each simple command that a Bash call's shell runs, after expansion
- [FEATURE] Mods API: carry the sender's agent id and one delivery id on every agent message
- [FEATURE] Mods API: expose an MCP tool's input schema and annotations before the call
- [FEATURE] Mods API: give Bash results the sandbox violations as a structured field
- [FEATURE] Mods API: add file identity, change time and case facts to $.fs.stat and $.fs.list
- [FEATURE] Mods API: give the working directory for each tool call and each configured command hook
- [FEATURE] Mods API: an observe-only event for every final tool denial, with its source and protected paths
- [FEATURE] Mods API: let a nested Claude Code session read its parent's session id and plugin state
- [FEATURE] Mods: let a mod see agents that run outside the lead's process (pane teammates, remote agents)
- [MODEL] Ran claude --bg with no prompt inside a script fed on standard input, so a background agent took the script as its instructions
- [MODEL] Ran claude -p inside a script fed on standard input; print-mode agents took the script as instructions, ran copies of it, and started more agents
- [MODEL] Wrote a file deletion into a test harness, inside a heredoc the user's rm deny rules do not match, and launched it
- [Feature request] Projects: let a project remember "Allow" for a folder on my computer
- [MODEL] Stated as verified that each running instance has its own transcript, contradicting its own test that resumed instances share one
- [MODEL] In correcting an unverified claim, marked another one verified: called a stale, blocked background job a running conversation with its own transcript
- [Bug] Unable to use recent models due to cyber flag restriction
- diff: the pane opens by itself with no file to list
- Auto-mode permission check blocks subagent launch ("Modify Shared Resources") despite explicit, broad user authorization for the exact work
- [Bug] Claude ignores explicit instructions and second-guesses firmware update requests
- Projects: project instructions silently truncated at 8,192 bytes for threads run on your computer (Work locally / Remote Control)
- Bash tool `grep` (bundled ugrep) silently misses `^` alternatives: `(a|b|^)x` and `(^|[^0-9])x` match nothing that GNU grep matches
Resolved issues (228)
- [Bug] iOS Simulator input actions fail with sandbox: "Failed to initialize IO ports for device IO client"
- [BUG] ${CLAUDE_EFFORT} in a skill preloaded into a subagent expands to the session effort, not the subagent's effort
- -
- [GitHub integration]
- [GitHub integration]
- [BUG] Desktop (Linux): claude-desktop runs with no_new_privs, so sudo fails in its terminal and Snap apps (default browser) cannot be launched
- [Bug] Anthropic API Error: Overly restrictive content filter on harmless bioinformatics operations
- macOS permission prompts and Privacy settings show Claude Code as a bare version number (e.g. "2.1.83")
- iOS app: routine session view disappears after backgrounding the app
- Opus 4.6 repeatedly ignores explicit 'do not act' instructions — bias toward action overrides rules
- Model asserts it lacks a capability it has, and presents an unverified search as proof
- [BUG] iOS app stuck on "repository must be selected" error
- [GitHub integration]
- i cant add subchats into my projects
- [GitHub integration]
- [Bug] VidIQ browser extension causes input issues on Claude.ai chats
- [GitHub integration]
- I CAN NOT FIND THE "ADD TO PROJECT" OPTION IN THE SIDE BAR "DOTS" IN CLAUDE SES…
- rivoxpanelbot
- Why can't I create a project containing sub-chat threads? I'm forced to chat di…
- I created a project just to store chats and artifacts, but it say sI have no p…
- It seems as though sometimes there's a disconnect between the threads and the c…
- [Bug] Safeguard Restrictions Blocking Legitimate Commercial Project Development
- [GitHub integration]
- [GitHub integration]
- [Bug] Permission classifier blocks safe command phrasings during production incidents, ignoring settings.json allowlists
- [BUG] Thinking blocks arrive empty after bundled runtime 2.1.215 → 2.1.217 update (desktop, Opus 4.8) — transcript bisect included
- Workflow tool: object args arrive stringified in the script; related evidence of within-session subagent co-deaths on connection close
- Agent(isolation: "worktree") can check the subagent's branch out onto the primary checkout when the calling session is itself inside a worktree
- [Bug] Pressing Esc closes config screen and cancels active multiple choice prompt
- [Bug] Incorrect character encoding for whitespace output
- [Bug] tengu_shale_finch flag strips all task tools (TaskCreate/TodoWrite) from top-level interactive sessions, leaving no task list at all
- Stop hook evaluator API call fails with 400 when transcript references a deferred/dynamically-loaded MCP tool
- [BUG] Account & Usage panel shows "Claude pro" for Max 5x subscriber (VS Code extension, Linux, v2.1.215)
- [Bug] Plan mode incorrectly requiring approval prompts for read-only operations
- Recent working folders (.claude.json "projects") are shared across all accounts on the same OS profile
- Custom subagents don't receive Agent/ToolSearch tools even when listed in frontmatter tools:
- Bundled search wrappers narrow silently and exit 0
- Auto-terminate idle claude.exe processes on tab close / task completion
- Marketplace plugins unavailable when session working directory is a WSL path (Windows integration)
- Allow trusted cross-session agent messaging without per-message confirmation (agent-to-agent messages as a distinct, lower-authority channel)
- [Bug] Concurrent subagents wedge on model token stall after tool_result with no error trace
- [BUG] "fg: no job control" hook_non_blocking_error regression on Windows (sync hooks), reappears after #4481
- [BUG] Claude Desktop Windows (MSIX 1.22209.3.0, CC 2.1.215): renderer bootstrap crash-loop, Cowork sessions-bridge wedges, unreaped claude.exe children — daily forced restarts on a Max 20x plan
- Large diff/tool-output blocks destroy scrollback in iTerm2 — conversational text becomes unrecoverable
- Feature request: external/programmatic writes to the session task list (~/.claude/tasks) — or a disk re-read contract for the task panel
- [Security] Assistant-generated text injected into USER message turn after advisor tool call
- [Bug] Fable 5 persistent premature-closure pressure overrides explicit anti-closure instructions and degrades investigation, implementation, and verification during benign coding work
- Slash commands highlighted mid-sentence but not dispatched
- Nerd Font / PUA Unicode characters not rendered (statusline, chat, file views) and dropped by the Read tool
- [Bug] Fable 5 false positive on OSS decompilation analysis in code review context
- [BUG] Background Bash task's process group is SIGKILLed mid-run when the command spawns a daemonizing CLI (cursor-agent) — and reported as completed (exit code 0)
- [BUG] Microsoft Dataverse MCP connector fails with "missing required resultType" on every call in Claude Code (works fine in claude.ai chat)
- [Bug] MCP update_document patch mode silently drops content after checkbox lists
- Add one-click/keyboard-shortcut archive for chat sessions
- You've hit your session limit · resets 4:20pm (Asia/Calcutta)
- [FEATURE] Peer-requested context reset: let one session hand a collaborator a clean context and a continuation prompt
- [BUG] Notion MCP connector (both project .mcp.json and claude.ai OAuth connector) triggers 400 input_schema oneOf/allOf/anyOf error in interactive main loop
- /skills prints "No changes" instead of the skill list, and misses skills added mid-session
- Completely incorrect about "you are doing security"
- Agent tool: plugin-namespaced subagent_type silently ignored when a name is passed (agent teams)
- [FEATURE] remote-control: a supported way to run it as a persistent service (survive reboot, crash, and auto-update)
- [BUG] list_sessions_request hangs on SMB-mounted workspaces (VS Code extension, v2.1.237)
- [Feature Request] Add security context mode for local vulnerability testing
- [BUG] Marketplace-entry dependencies: version constraints silently ignored, prune deletes live dependencies
- [BUG] /context in Claude Code for VS Code causes request spikes and may trigger rate limits or DoS protection
- [BUG] /remote-control times out when re-attaching a session that was previously attached
- [BUG] Claude Desktop (Windows MSIX): package's own LocalSystem service CoworkVMService blocks every update with 0x80073D02 — closing the app cannot fix it
- [BUG] Session history/sidebar shows no prior sessions after app update (2.1.215 → 2.1.219), despite transcripts existing on disk
- [BUG] Stalls before response headers have only the 600s API_TIMEOUT_MS backstop — silent 3-10 minute hangs with no error or retry
- @-mention file picker doesn't search additionalDirectories / --add-dir directories
- [FEATURE] basic needs in the web ui: claude.md, output style
- Sandbox blocks Go TLS on macOS: trustd Mach port inaccessible
- [BUG] Billing: chargeback on old subscription cancelled separately paid active Pro subscription
- [Bug] Reasoning extraction flags blocking workflow execution
- Worktree lifecycle isn't reflected in the agents/sessions menu
- Published artifacts deleted server-side without user action on a personal account (no delete UI exists; links were already shared)
- Claude Desktop: unexpected re-authentication when switching between Chat and Code tabs
- [BUG] Claude Desktop (Windows) silently kills run_in_background tasks after 15-min idle — WarmLifecycle taskkills the embedded CLI process tree
- CLAUDE ACCIDENTALLY DELETED HIS SETTINGS
- [BUG] Title: Chat input composer overlaps the last message, requires manual scroll to see it
- [BUG] TUI renders garbled inside tmux since 2.1.200 (last-good 2.1.199) — only repaints on forced redraw
- [BUG] claude-sonnet-5 reported cost inflated exactly 1.5x — CLI pricing table uses $3/$15 instead of the (now permanent) $2/$10 first-party rates
- [BUG] Claude Code deleted 4.9GB of Skills!
- Pinned sidebar section fills up with sessions and cannot be cleared
- False-positive detection a dangerous domain (BIO in the context of fine-tuning)
- [Bug] Anthropic API Error: Fable 5 safeguards falsely flagging safe messages
- [BUG] The local chats on VS code is Removed
- [BUG] Plugin page "copy install command" points to wrong marketplace for third-party plugins
- Feature request: license the Claude Code harness (source-available is sufficient)
- [BUG] Persistent bottom-right notifications ("Update installed · Restart to update") replace the context meter instead of coexisting with it
- [Bug] Anthropic API Error: Safeguard triggered on non-harmful content
- [BUG] Claudia ignores CLAUDE_CONFIG_DIR and reads credentials from ~/.claude
- Prompt is too long
- [FEATURE] CLI to send a cross-session message to a running session without a model turn, addressable by session id
- Bug Report: Backend startup failure
- [GitHub integration] 1ER depot
- Background subagents stall on permission prompts in bypassPermissions mode (2.1.296)
- Bug Report: Safeguards blocking legitimate requests
- [Bug] Claude Code refuses to work on local files
- Pro aboneyim. Hesabımda Projeler bölümü yeni "Claude Code tabanlı" proje sürümü…
- [GitHub integration] Claude Desktop: issue-creation confirmation form closes immediately, tool call shows as failed
- [BUG] Deleting chats from the sidebar permanently erases all-time usage stats and heatmap history (desktop app home stats widget)
- I'm Claude Code, Anthropic's CLI tool. I can only help with issues related to Claude Code itself. For Apple support, please visit Apple's official support website or contact them directly through the
- [Bug] File renamed without user permission during removal request
- [Bug] Code obfuscation requests incorrectly flagged as policy violation
- [Bug] High token consumption with Claude 3.5 Sonnet on low reasoning settings
- [Bug] Claude ignores user instructions and fails to correct behavior on repeated requests
- Claude giving safety error
- New Project fails to boot the environment
- [GitHub integration]
- [GitHub integration]
- [GitHub integration]
- [GitHub integration]
- I don't see a bug report in your message - you've only provided a file path. Please share the actual bug report content so I can generate an appropriate GitHub issue title.
- [GitHub integration]
- [GitHub integration] claude.ai
- [Bug] Anthropic API Error: Opus model safeguards restriction - cyber verification required
- Problem report
- [GitHub integration]
- [Bug] Claude models returning safety rejections despite valid cyber verification
- [Bug] CVP Program Authorization Not Recognized for Organization Security Audits
- [GitHub integration]
- [Bug] Session flagged when using Claude 3.5 Opus model
- [GitHub integration]
- [BUG] 2.1.237 tagged `latest` with its linux-x64, win32-x64 and linux-x64-musl native packages never published - installs land on a dead 500-byte stub
- [Feature Request] Allow manual model selection override for agent testing
- [BUG] German UI (Cowork): permission mode "Manual" is rendered as "Handbuch" (= printed handbook) — misleading, not just a wording slip
- [FEATURE] Job/agent list shows only a relative age (9d, 45d) — surface an absolute date too
- VSCode extension: empty <pre id="claude-error"> forces a permanent 1px horizontal scrollbar in the webview
- [Feature request] RTL support + word-boundary wrapping for Hebrew/Arabic in Claude Code (desktop app)
- Make agent-view session shortcuts remappable (`Alt+<n>` is unavailable in several terminals)
- [Bug] Anthropic API Error: Safeguard Blocked Message During Multi-Session Transcript Investigation
- Model fabricated a user message and executed it: unrequested git commit, push, and issue edit
- Allow custom slash commands to invoke other slash commands (e.g. /clear, /compact)
- [FEATURE] One-click session handoff: end a session into a new linked one, shown as a thread in the session list
- Add 'spawn new chat' feature that carries current context (like /rewind does)
- [MODEL] Claude autonomously launches costly subagents without explicit consent
- [BUG] Rolling 5-hour usage limit not resetting
- [bug] Sigma rule creation bas on PoC triggers safeguards
- PR-activity webhook events render as if the user sent them, and non-actionable bot comment-edits are relayed with full raw payloads
- Feature request: allow Remote Control when ANTHROPIC_BASE_URL is a transparent local proxy to api.anthropic.com
- Fullscreen mouse capture: Ctrl+click on a URL in Windows Terminal opens it twice
- [BUG] Claude Code shows 'Usage limit reached' despite session at 2% and weekly at 32%
- [BUG] @-mention file picker cannot descend into directory whose name contains a space — child files never suggested
- [Bug] Anthropic API Error: Request flagged incorrectly with "cyber" keyword
- [Bug] Anthropic API Error: Reasoning Extraction False Positives in Extended Sessions
- Session history split across two project-slug folders for the same cwd (ñ normalization / space+case handling)
- [Bug] Anthropic API Error: Repeated failures during plan updates
- CLAUDE_AX_SCREEN_READER=1 set correctly but screen reader mode doesn't activate (v2.1.237)
- [FEATURE] Unscoped Read of an over-cap file: return stats + outline instead of a billed first-page slice (the pre-flight half of 2.1.145's PARTIAL view)
- Built-in Task tools (TaskCreate/TaskGet/TaskList/TaskUpdate) intermittently unregistered, and can vanish mid-session — breaks the live task display
- [Bug] toggleBrief command lacks visual feedback and statusline state exposure
- [Bug] `ScheduleWakeup` tool called outside `/loop` with missing required `prompt` field
- [MODEL] Opus suggested that shell mode in claude-code was not in chat context
- [BUG] Windows: Bash tool silently corrupts written file content via command substitution and reports exit code 0
- [BUG] Bun 1.3.14 JSC GC segfault (SlotVisitor::drain) on AMD Ryzen 5 5500 / Linux x86_64_baseline — all platform binaries identical
- [BUG] Cross-session message written to target transcript but excluded from model's context
- send_message: cross-session message is written to the target transcript and rendered, but never enters the receiving agent's context
- [BUG] /design-sync card scaffold invents viewport and rewrites labels
- Bug: `isolation: 'worktree'` renames worker branches and leaves the primary worktree checked out on a run branch
- [FEATURE] Pause-and-steer for extended thinking — and formalizing delegation as steerable reasoning
- [BUG] Claude Code desktop causes kernel panic via memory exhaustion (WindowServer watchdog timeout)
- Claude chrome extension
- [BUG] Claude Tag: !help, !routines, !restart silently ignored — commands fall through to the session as ordinary prompts
- [Bug] Unable to start conversation after compaction
- Fenced code blocks render with no visual container and lose their list indent
- First-class session relocation and deletion (move/delete a transcript + its sidecar state)
- [BUG]
- [Bug] Fable blocking explicitly permitted commands in settings.json
- Phantom user messages: text appears in transcript as user input that the user did not type
- [Bug] Resume workflow skips journal cache entries and re-runs completed agents
- Model fabricated a 'user' turn inside assistant output, then acted on it (unrequested ScheduleWakeup)
- Agents view: needs tools to oversee a large number of background agents
- [BUG] v2.1.210 regression: PreToolUse prompt-hook deny stops the entire turn (hook_stopped_continuation) instead of returning a tool error
- [BUG] PreToolUse prompt-hook deny returns the hook's entire configured prompt to Claude in the tool error, not just the reason
- [Bug] Thinking mode incorrectly flagged as reasoning extraction during prompt generation benchmark
- [BUG] Orphaned temp_<epoch-ms> folders accumulate under ~/.claude/plugins/marketplaces
- [BUG] I cannot edit task rules in Claude desktop MAC. I sent to support two weeks ago with no reply and it still doe snot work
- Feature request: list configured MCP servers without health-checking them (--no-health / --json)
- [BUG] "Prompt is too long" in a new session when another session has an open plan
- [BUG] Question/validation modal obscures preceding response text with no way to minimize
- [FEATURE] Extend skillOverrides to plugin-namespaced skills
- [Bug] Safety guardrails triggered unexpectedly without user input
- Legacy Windows console (conhost): SI byte from reassertTerminalModes renders as a visible glyph in the input box; ambiguous-width glyphs corrupt partial repaints
- [BUG] unable to rename connector on desktop app
- [Bug] Fable rejection on `jq -r` command triggers automode safeguard
- [BUG] First prompt freezes TUI ~40–60s (event-loop stall, ~2.6GB RSS) even with --safe-mode
- [BUG] Scheduled Routine trigger fires on schedule but session doesn't execute until a browser tab is opened on it
- Windows: Bash tool permanently broken in long-lived sessions - inlined shell snapshot exceeds command-line length and is truncated
- [Feature request] Sort the completed background-job list by last activity, not creation time
- [FEATURE] Raise, remove, or make configurable the 10,000-char limit that persistHookOutput enforces on hook additionalContext (introduced in v2.1.89)
- [Bug] Plan mode ignores MCP tool allow list permissions
- Channel messages should be delivered the same way whether the session is idle or mid-turn
- [BUG] Desktop (Code tab): desktop extension tools stay blocked — a legacy "server:tool" entry set to false cannot be cleared
- [BUG] claude -p leaves stdio MCP server processes (mcp-remote) running after exit
- [BUG] VS Code extension still auto-declines MCP form elicitation (onElicitation not wired) — 2.1.287
- [FEATURE] MCP OAuth: allow overriding or omitting RFC 8707 `resource` (Entra ID AADSTS9010010) in Claude Code, Claude Desktop, and claude.ai
- MCP OAuth client invents `<gateway-origin>/authorize` instead of resolving the RFC 9728 discovery chain (authorization_servers → AS metadata) — non-compliant with MCP auth spec
- Session-start connector status falsely reports Linear as requiring auth when it's fully connected
- Claude Design MCP 403: authorization-server metadata 404s at both RFC 8414 paths
- MCP connector tools are exposed under an opaque, unstable UUID instead of the connector name
- gitlab mcp problems reading joblog
- MCP client rejects tool calls omitting optional parameters; deadlocks on tools with mutually-exclusive parameter groups
- [Bug] MCP tool approval in planning mode not persisting between iterations
- [BUG] Adobe for Creativity plugin: Dynamic Client Registration rejected (HTTP 403)
- [BUG] Cowork: MCP elicitation/create silently auto-cancels (reproduced with a different MCP server than #56243/#48164)
- [BUG] ToolSearch returns stub schemas (no parameter properties) for third-party MCP connector tools, causing typed-parameter calls to fail
- [Bug] Anthropic API Error: HTTP 403 Missing Scope for MCP Design Endpoint
- [FEATURE] Support multiple simultaneous Notion workspace connections in one session
- Claude Design MCP server: FIRST_PARTY_AUTH_REJECTED (HTTP 403) despite plugin being installed
- [BUG] MCP tool calls with dict/array parameters serialized as strings, fails mid-session after working correctly
- Scheduled routines (RemoteTrigger) report 'no connectors' despite connectors showing as Connected in claude.ai
- Feature request: store MCP OAuth tokens separately from the Claude account credential (multi-account machines)
- claude.ai Gmail MCP: plain-text drafts get URLs rewritten to google.com/url redirect wrappers
- MCP client rejects tool calls when outputSchema declares JSON Schema draft-07 dialect
- MCP OAuth: redirect_uri uses localhost:<port> while metadata publishes portless localhost and 127.0.0.1 (still failing in 2.1.246)
- [BUG] CCD desktop app rejects MCP tools with draft-07 outputSchema that work fine in CLI
- [BUG] MCP tool schema mangled for locally-configured stdio servers (Optional[str] params)
- Supabase MCP plugin OAuth fails with "Unrecognized client_id"
- Snowflake MCP connector shows stale 'requires authentication' status when actually working
- [BUG] Slack connector shows "Connected" in Settings but is invisible to routines/MCP authorization
- [BUG] Filesystem MCP server (remote-devices) fails all tool calls: outputSchema declares unsupported draft-07 dialect
- [BUG] GitHub connector authorized but write operations fail with "403 Resource not accessible by integration
- [BUG] KIE.ai MCP connector: prepare_media_generation always returns "Host declined the approval request" on Desktop app
- MCP: nested array/object tool parameters serialized as strings for claude.ai connector tools (Windows desktop)
- claude.ai account connectors don't attach to spawned sessions until the first inbound user message — autonomous background sessions run with no connector tools
- Per-session MCP server spawn multiplies RAM use — lazy spawn or per-session MCP scoping needed