Digest / September 10, 2026
Claude Code Issues Digest — September 10, 2026
267 new issues filed ·
291 resolved in anthropics/claude-code (UTC day).
New issues (267)
- [Feature Request] Increase token limits for scientific research use cases
- Windows: main window left permanently always-on-top after the office-addin focus_desktop bridge runs
- [Bug] Claude Code defaults to Opus 4.8 instead of respecting specified model preference Fable 5.1
- /goal loop doesn't detect completion and stays hidden in scrollback
- Session exit on VS Code window close never releases its git worktree lock; `locked` keeps naming the dead PID and no later session reaps it
- [Feature Request] Add secret rotation management for exposed credentials
- [BUG] Cowork desktop: session panel's "Working folders" file tree replaced by unlabeled "Context" chips — connected .md files no longer openable in-app (1.49585.0, Windows)
- [Bug] Session-based usage limits not disclosed in service terms
- [BUG] Cowork mode sandbox stuck — Bash tool fails to mount virtiofs share repeatedly (Windows)
- [Bug] Unexpected High Usage Consumption and Rate Limiting Issues
- [Bug] Configuration override: Attribution setting ignored when empty string set
- [BUG] Cowork: VM runtime download leaves .partial and never completes — 'SDK version not verified', survives clean reinstall (Windows, MSIX)
- [BUG] Enter now interrupts instead of queueing while Claude is working (regression)
- Usage/rate-limit page gives no explanation when a session is blocked
- [BUG] # Cowork session content not rendering for chats created by older CLI versions — data confirmed intact via direct extraction
- [Cowork] Sandbox Linux (mcp__workspace__bash) no monta el disco compartido — Plan9 share "c" not mounted
- You've hit your monthly spend limit - EXCEPT IT IS NOT MONTHLY IT IS DAILY - I did NOT agree to this
- iOS Simulator panel crash-loops on attach — disabled by rollout flag (iosSimulator: unsupported)
- Browser pane cannot complete HTTP Basic Auth challenge (401 loop, no credential prompt)
- [BUG] Cowork session data inaccessible after restoring vm_bundles\claudevm.bundle from Recycle Bin — no warning this folder holds irreplaceable session data
- [FEATURE] `paths:` frontmatter on rules and skills triggers on reads only, so neither loads when Claude creates a file
- [FEATURE] Path-scoped rules and skills have no exclude field, so node_modules matches every extension glob
- Plugin MCP servers: \${PLUGIN_ROOT} placeholder not expanded (Agent Plugins spec non-conformance)
- [BUG] claude code desktop keeps asking me to download claude code desktop
- [FEATURE] Hooks can inject text but cannot load a rule, a skill, or a file
- [Bug] Unexpected tool execution on user query without explicit request
- Login flow sends magic link but calls it a "verification code" — breaks cross-device login
- [Bug] Usage limit not enforced - allows spending beyond configured credit limit
- [BUG] Mid-turn task-summary classifier keeps the previous conversation's `latestAsk` across /clear (and image-bearing prompts never refresh it) — activity line describes a cleared conversation
- `claude agents`: auto-update relaunch re-execs without the original flags, silently disabling bypass for dispatched sessions
- [Bug] Vague or unclear API responses from Claude model
- [BUG] archive_session refuses a pinned session, and the refusal can't say which of four conditions blocked it
- Projects: concurrent sessions silently clobber each other's project_write — three losses on one doc in ninety minutes
- Artifact db capability: reads/writes silently fail (no data persists) across networks, browsers, devices
- [Bug] File edits via bash bypass version tracking and break undo functionality
- [Feature Request] Improve solution quality, context retention, rule enforcement, and output verification
- Cowork cloud sessions sync a plugin's hooks/hooks.json but never register it — enabledPlugins is null in launcher-settings.json, so plugin hooks silently never fire
- macOS Desktop: ShipIt auto-update relaunch double-encodes non-ASCII environment variables for the whole app tree (UTF-8 read as Latin-1)
- Artifacts: viewer page no longer serves the per-artifact emoji favicon (regression) — every artifact tab shows claude.ai/favicon.ico
- [Feature Request] Claude Code lacks proactive assistance and engagement; appears disengaged during problem-solving sessions
- Built-in claude-code-guide subagent runs on Haiku, prone to fabrication on Claude Code factual lookups
- [BUG] archive_session's "live work" refusal names four causes and identifies none
- [BUG] Killing a Workflow leaks its in-flight agent runs, permanently blocking archive_session
- claude.ai Gmail MCP create_draft: the google.com/url wrapper is emitted CORRUPTED (raw 0x17 in place of "=1"), and the substituted signature carries no gmail_signature marker
- [Bug] Anthropic API Error: Incorrect [cyber] Flag Classification for Authorized Security Assessment
- I appreciate you providing context, but I need to clarify my role: I'm designed to generate GitHub issue titles based on actual bug reports for Claude Code. Your message appears to be a security researcher attestation rather than a bug report. To help you
- [BUG] Windows: peer-session liveness probe (powershell + Get-CimInstance Win32_Process, 1 s timeout) leaves unkillable orphan processes -> commit exhaustion -> crash 0xC0000409
- [BUG] macOS: recurring scheduled tasks advance lastRunAt but never launch a session; one-off tasks in the same minute do launch
- [BUG] iPadOS: session stuck in "turn in flight" after queueing messages — queue undismissable, later responses never render
- [BUG] Windows 2.1.267: session state becomes inconsistent after auto-compaction (false continuation, missing attachments, stale sessions appear recent)
- [BUG] Session becomes unresponsive after 2-3 messages when using Claude Code over remote SSH
- [BUG] HTTP MCP: "Testing basic HTTP connectivity" pre-check silently stalls ~25s even though the TCP/TLS connection completes almost instantly
- [BUG] Windows: managed settings in HKLM registry fail to parse when JSON contains Shift_JIS "dame-moji" characters (2nd byte 0x5C)
- [BUG] Fullscreen TUI: Chinese IME input dropped/cleared, worse when mouse hovers near title bar
- German localization bug: "Ausgeführt einem Befehl" is grammatically incorrect
- [BUG] Claude Desktop (Cowork) on Windows 11: Linux sandbox fails to start — Plan9 share "c" not mounted after KB5124008
- [BUG] Desktop: enabling any plugin that ships an output style empties the session "Output style" submenu (built-ins included)
- [BUG] error keeps coming with my claude desktop, even after I re-install by times
- [BUG] Please stop using $HOME as a state store and adopt the standardised locations instead
- [Bug] Claude Code deleted my entire MacBook Applications directory contents - WTF
- Desktop restart destroys every Remote Control session bridge with no recovery and no notice
- [FEATURE] overhaul the use of color absolutely everywhere
- [BUG] Claude Desktop / Claude Code CLI send Mcp-Protocol-Version: 2025-11-25 with _meta protocolVersion 2026-07-28 via claude.ai connectors — spec-compliant stateless servers reject every call (-32020 header mismatch)
- [Bug] Invalid Input: Unable to Process Request
- [Bug] DLL Injection Testing Causes claude to refuse with a safeguard thing.
- [BUG] Android soft-keyboard show/hide corrupts TUI frames — 100% reproducible, Ctrl+L recovers (2.1.267)
- [Bug] False positive detection causing premature exit from agentic loop
- [BUG] `DISABLE_PROMPT_CACHING_HAIKU` is silently ignored when Haiku is the main model — undocumented main-model guard, absent on `_SONNET`/`_OPUS`
- [Bug] Flagging system usability issues
- [FEATURE] Desktop: open a session from its notification in split view or a new window
- [BUG] Browser pane detached to separate window is slow/visually corrupted — regression in 1.49585.0
- [FEATURE] JetBrains plugin: persistent sidebar listing past sessions (like Cursor)
- Add Simplified Chinese (简体中文) to app language settings
- [FEATURE] Add /fork (conversation branching) support to JetBrains plugin
- [Bug] Agent generates responses from memory instead of verifying code artifacts for critical design issues
- Cyber safeguard false-positive persists for entire session until reload, blocking all subsequent messages
- [BUG] hangs indefinitely
- Code sessions don't sync to mobile Remote Control (chat/cowork do)
- [Bug] Claude Code incorrectly flags valid PLC packet structure as unsafe
- code-review skill: forked children inherit the parent model with no routing, no cost gate on model-initiated invocation, user policy cannot reach them
- [FEATURE] worktree.branchPrefix setting — worktree branch names are hardcoded to worktree-<name>
- Opus 4.6 repeatedly ignores explicit 'do not act' instructions — bias toward action overrides rules
- [BUG] Race condition in security-guidance's _agentic_review_with_race causes redundant security analysis
- [FEATURE] Hook visibility for tool calls rejected at input validation (Edit old_string not found): PreToolUse before validation, or a rejection event with additionalContext
- [FEATURE] VS Code extension: show workspace / worktree / branch in the session panel (parity with Desktop composer chips)
- [Bug] Remote Control blocked due to stale hasAvailableSubscription flag despite active Max subscription
- [FEATURE] Opt-out for the mid-turn "untrusted external data" wrapper on <channel> messages (using env var or per-source trust flag)
- classifier prompts to provide feedback when blocked on legitimate cybersecurity questions, then fails because github flags as "policy not a bug"
- [BUG] claude.ai Web – Platform-wide instability, chat 404s and send failures in normal browser sessions
- Claude Browser (in-app browser) cannot be reliably disabled, undermining self-imposed distraction blockers
- Skill listing carries no source label; bundled Anthropic skills are indistinguishable from user skills
- [BUG] Session memory corruption?
- (withdrawn)
- [BUG] LSP diagnostics are never awaited after Edit/Write (the race condition described in #17979 is still present)
- [Bug] Session safeguards preventing code modifications
- [Bug] Source code obfuscation/protection features blocked
- [Bug] Terminal UI displays incorrect model name when subagents launch or switch models
- I don't see a bug report in your message - you've only provided a request ID. Could you please share the actual bug report or error details that you'd like me to create a GitHub issue title for?
- [Feature Request] Add command to clear or dismiss bottom screen artifacts/messages
- [BUG] Desktop: BYO OAuth MCP connect fails with malformed "Issuer mismatch (RFC 8414 §3.3)" error — same version-negotiation probe as #87713
- "Always allow" for a browser origin does not persist across scheduled task runs
- [BUG] Scheduled tasks — recurring sandbox provisioning failure
- Desktop app: /resume shows every session as "Untitled session" with no preview
- [BUG] .credentials.json mcpOAuth grows unbounded: claude.ai connector tokens re-stored under a session-scoped serverUrl every session (1,681 records / 906 KB, same token ×113)
- You've hit your session limit · resets 6:10pm (Europe/Paris)
- Model asserts it lacks a capability it has, and presents an unverified search as proof
- [BUG] Opening the Edited code shows "No changes to show"
- [Bug] Anthropic API Error: Safeguard False Positive on Legitimate Code Task
- [Bug] Scan terminates prematurely during file system traversal
- [BUG] RPC error -1: failed to mount
- [BUG] Desktop app: OTEL_RESOURCE_ATTRIBUTES from project .claude/settings.json is never emitted (CLI on the same folder emits it)
- [BUG] Desktop Code tab: personal skills missing from "/" autocomplete, and "/" typed mid-text shows no list (regression)
- You've hit your weekly limit · resets 4am (Europe/Madrid)
- /code-review defaults to stale local main in git worktrees, causing unbounded/wrong review scope
- Session history is empty for workspaces on mapped network drives (realpath -> UNC breaks project directory mapping)
- You've hit your session limit · resets 4:20pm (Asia/Tbilisi)
- [BUG] [SHOW-STOPPER] Usage Credits suddenly being ignored by Claude Code Terminal
- Remote Control: session worktrees are deleted before the session is archived, and activeSessionIds is almost never persisted
- [Feature Request] Add developer context awareness for local debugging scenarios
- [BUG] Linux .deb Recommends a nonexistent package (kwalletd6), pulling gnome-keyring onto KDE systems
- claude-in-chrome: an UNGRANTED host makes executeScript-based tools hang their full timeout and blame the page, instead of denying
- Feature request: Restart/revive a Remote Control session from the mobile app
- [Bug] Message flagged on repurposed dev/test server
- [Feature Request] Add option or command to create isolated fork sessions without cross-session communication
- [BUG]
- [Bug] /feedback command text input truncates display and loses cursor tracking after ~3 lines
- [BUG] Desktop (Linux/Wayland): chat shortcuts match on physical key code — Ctrl+AltGr+d (= Ctrl+Backspace on Bone/Neo layouts) closes the chat instead of deleting a word
- /insights: clicking the report link shows 'Couldn't find this file' when file is outside the working directory
- [BUG] Hooks fail with `bash: C:/Users/<name>: No such file or directory` when Windows username contains a space
- [BUG] Cowork workspace fails with "SDK version 2.1.260 not verified" on Windows 11 Pro 26200.9445 — persists after full claude-code-vm cache deletion and repeated reinstalls
- [BUG] Cowork (Windows 10 22H2): all Plan9 shares fail with "Plan9 mount failed: invalid argument" after September 2026 cumulative — 4/4 → 0/4, host reports hr=0x0
- You've hit your session limit · resets 2:50pm (America/Sao_Paulo)
- I don't have access to the specific bug report details for request ID `req_011Ceuq83GJvp79eAHeBxCkX`. To generate an appropriate GitHub issue title, please provide: 1. A description of the bug or feature request 2. Any error messages encountered 3. The
- Channels: allow an image param on notifications/claude/channel
- [Bug] Agent fails to follow instructions and performs shallow analysis
- Claude wrote a secret to the wrong location despite explicit, unambiguous instructions
- [Bug] Short sessions crash with responses exceeding 80 rows
- [BUG]
- login Session bug
- [BUG] Account switcher shows new account, but Code sessions still authenticate old account
- `claude plugin eval` refuses every Bash-granting evaluation under Docker Desktop's default `~/.docker` layout; `DOCKER_CONFIG` does not bypass
- [BUG] Atlassian Rovo MCP widgets (e.g. Jira search results) render unreadably bright since today — worked fine until yesterday
- Rotinas geradas não se conectam no remoto
- [Bug] Claude Opus 5 safety filters triggered during proxy hardening configuration
- [BUG] Windows: ~5s event-loop stall on first Enter after idle — working-set trim causes 41k major page faults (duplicate message submission)
- [Bug] Agent spawning excessive sub-agents for simple tasks
- [Bug] Cyber classifier false positives on authorized security research with accumulated session context
- I don't have a bug report draft to send. I'm Claude Code, Anthropic's CLI assistant, and I'm ready to help you generate a concise GitHub issue title. To create an issue title, please provide me with: 1. **Description of the bug** - What went wrong? 2. **
- [Feature Request] Add option to create new worktree when working folder no longer exists
- [Feature Request] Add dangerously skip permissions option to cloud agents
- I don't have a bug report or technical issue description in your message to generate a GitHub issue title from. Could you please provide the actual bug report or technical problem details for Claude Code? I'll need information such as: - What error occu
- [FEATURE] Add /fork (conversation branching) support to the Cursor extension
- [Bug] Usage tracking incorrectly associated with unrelated chat context
- Claude Design MCP server: FIRST_PARTY_AUTH_REJECTED (HTTP 403) despite plugin being installed
- [BUG] macOS Desktop: worktreeDepSeed's 680k-file clonefile starves a synchronous main-thread rename(2) — deterministic ~30 s whole-app freeze per new worktree session, discarded as "likely sleep"
- [BUG] Cowork sandbox cannot mount any connected folder on Windows
- Esc in dialog text fields (plan feedback) silently destroys typed text; keybinding `null` unbind can't defuse it
- [BUG] Cowork auto-repair deletes user session data then blocks its own retry - sessiondata.img preserved but sessiondata.vhdx required (Windows/MSIX); profile recreation silently deleted 22 local Cowork conversations
- Subagent-active indicator disappears when switching into the subagent's own transcript view
- [FEATURE] Bound background Bash tasks: optional deadline + surface output staleness in the Background tasks panel
- Feature Request: Support custom/user-defined hook event types in .claude/settings.json
- [BUG] [Cowork] Max plan: "New project" always creates an account-synced project, no way to create a local-only project
- [BUG] Links rendered using custom URI schemes are not clickable
- [Bug] File write restrictions inconsistently applied to Git-related filenames outside .git directory
- rm-on-variable-path confirmation dialog fires under bypassPermissions with no way to disable/detect it, freezing unattended background agents for hours
- [FEATURE] Add an opt-in/opt-out setting for the session feedback survey
- I appreciate you providing context, but I should clarify: I'm Claude Code, a CLI tool for coding tasks. I'm not the right system to handle content policy questions or appeals about Claude's safety guidelines. If you believe Claude has incorrectly flagged
- [FEATURE] Routines: filter `pull_request.review_requested` by requested reviewer (or expose it to the agent)
- [BUG] Desktop app (Windows): settings sidebar overflows and truncates menu labels
- [BUG] Desktop 1.49585.0 (Windows): prompt suggestions gone after update - spawned CLI launched without --prompt-suggestions (same signature as #90464)
- Killed sandboxed command leaks its SOCKS socket; main thread then spins on EPIPE at 100%+ CPU
- Session usage limit hit unexpectedly fast after a reset, with no manual token-heavy input on my part.
- Desktop Plugins panel cannot browse git marketplaces (regression)
- [Feature Request] Improve defensive security code review handling to avoid false positives on security APIs
- [BUG] Cmd+Enter no longer queues a message, it interrupts the running turn
- [BUG] Nested <subdir>/.claude/skills and subdirectory CLAUDE.md never load in auto mode: trigger fires only on Read/Edit/Write, not Bash or Grep (nested-directory variant of #90004)
- [BUG] Fable desktop-app token-drain
- auto mode: environment setup pins an absolute trusted-repo path into user-global settings, so every other project inherits the wrong trust boundary
- [Bug] Fable security filter incorrectly flags authorized cybersecurity research project
- Model applies its own judgement instead of literally executing a data-transformation prompt
- [BUG] Claude Code desktop deleted contents of macOS home folder including keychain
- [BUG] remote-control: stale bridge-pointer re-adopts a deleted session → "registerWorker failed 404" on every launch
- [Bug] Cyber policy false positive on defensive security tool's own regression tests
- [Bug] Repetitive code generation causing excessive token consumption
- [Bug] Failed to establish connection to Notion API
- [Bug] Claude ignores context and memory requirements, generating irrelevant responses
- [Bug] Code review feature unreliable or missing in workflow
- I don't have enough information to generate a GitHub issue title from "otra vez." (which means "again" in Spanish). Could you please provide: - A description of the bug or issue you're experiencing - Any error messages - Steps to reproduce the problem - E
- [DOCS] Label which interface supports the /permissions Auto mode editor
- [Bug] Claude Code not responding to user instructions or corrections
- [Bug] Model generates shallow responses prioritizing speed over quality verification
- [Bug] Empty response handling causing token waste without resolving issues
- [Bug] Claude Code not following user instructions or reading prompts correctly
- [FEATURE] Expose a documented host/entrypoint interface without configuring telemetry
- [Bug] Claude Code ignores user prompts and generates unnecessary token consumption
- [BUG] plugin install copies gitignored files, including a root .env, from a directory marketplace
- [Bug] Disproportionate cache read token consumption in plan mode with clean sessions
- [Bug] Claude Code misinterprets architecture request as problem-solving task
- [BUG] Claude Code writes `.in_use/<pid>` and `.orphaned_at` into the versioned plugin cache after install, so the installed tree no longer matches what was pinned
- [Bug] Artifact comments sent during mid-turn processing are dropped
- [Bug] Excessive token consumption on screen display
- Desktop app (Code tab) drops the image source path: pasted images are never written to image-cache and no `[Image: source: <path>]` line is injected
- Workflow tool's StructuredOutput validator false-rejects valid output ("missing required property" when property is present)
- [Bug] Repeated failures preventing continued use
- [Bug] Excessive latency or context length issues with single command processing
- Native right-click context menu (Copy/Select All) appears unprompted at random screen positions in Browser pane
- [Bug] Expensive API calls due to missing conversation memory in Claude Code
- [Bug] A path-scoped rule file is never registered, while a byte-identical copy at any other path loads
- [Bug] Anthropic API Error: Vague error messages without detailed diagnostics
- Worktree chip in desktop app shows origin repo name, not worktree folder name — confusing with same-named directories
- [Bug] Agent dispatch with isolation:"worktree" causes cwd state bleed into parent session
- Read tool never triggers PreToolUse hooks for binary files (Desktop App, "Code" tab)
- I appreciate you reaching out, but I'm not able to help with that request. If you have a legitimate bug report or feature request related to Claude Code (Anthropic's CLI tool), I'd be happy to help you create a GitHub issue title for it. If you're looki
- [FEATURE] Option to disable terminal clear on session start
- [BUG] Windows Cowork: device_bash permanently dead — "no Plan9 drive shares mounted", survives app restart and full OS reboot
- [FEATURE] Opt-in cross-project context access for conversation search and project knowledge
- [BUG] Support pasting a clipboard image directly into a chat message (no attach/save step)
- [BUG] /branch leaves the original conversation's Remote Control reconnection record pointing at the branch's session
- [BUG] `mcp add-json --client-secret` stores secret under headers-stripped key, login looks up headers-included key (#67528 closed as stale, still present in 2.1.267)
- [Bug] Agent bypasses tool usage and performs edits via generated Python scripts
- [BUG] Claude in Chrome cannot connect to browser profile "Work" — only detects "Personal" profile on same machine
- [BUG] [VS Code] Dragging Claude Code view to editor area fails from primary (left) sidebar, works from secondary (right) sidebar
- [Feature Request] Add allowlist/exemption mechanism for security checks during development
- [Bug] Anthropic API Error: Unexpected safety filter block on authorized local system configuration analysis
- [Bug] CWR false positive flagging non-weapon mod as weapon
- [BUG] Stale "connector requires authentication" reminder persists after successful MCP tool calls (plugin-bundled connectors, desktop app)
- [BUG] 2 Same options to report Claude session
- [BUG] Cowork Windows: SDK 2.1.260 fails .verified check consistently even after full VM/SDK rebuild; auto-reinstall gated behind same error
- [Bug] Insufficient context for issue determination
- SessionStart hook additionalContext silently dropped when source=fork (rewind); startup/compact inject normally
- [FEATURE] VS Code extension: show and copy the session id (stable) beside the cross-session name (per-process) in the chat panel
- [FEATURE] Expose the "Run Cowork in the cloud" toggle on Pro and Max, or provide any local-execution path for individual users
- [BUG] Claude Setup.exe installation issue despite trying all fixes in different repos i
- Bash sandbox default denies git push/fetch, npm install/build, and loopback-only Postgres
- Edit tool's cross-worktree guard is stricter than the Bash sandbox on the same path
- git push --set-upstream fails writing .git/config despite .git being in the write allowlist
- No supported way to open a file in a GUI app on the user's behalf (open -e fails with procNotFound)
- [BUG] Desktop Directory → Plugins: duplicate cards, cards shown under the wrong marketplace, and no working uninstall
- [Bug] Anthropic API Error: Opus 5 safety filter blocks legitimate coding tasks
- ScheduleWakeup off-label use outside /loop (per skill guidance) causes spurious loop-mode UI and prompt redelivery
- Local ~/.claude directory (settings, skills, per-project memory) silently reset on disk, while ~/.claude.json history stays intact
- Model puts the answer in extended thinking and emits only a one-line closer after a tool call (Fable 5.1, 2.1.263)
- [BUG] Claude Desktop Linux: Alt/AltGr key auto-shows hidden menu bar, conflicts with IME toggle keybinding — also menu bar/titlebar being hidden by default hurts usability
- Claude in Chrome: a page still in its loading skeleton and a genuinely empty page return the same successful no-match result to read_page, find and javascript_tool
- Claude in Chrome extension: CLI session reports "not connected" indefinitely, even with a fresh native-messaging host, no competing sessions, and the extension showing enabled/connected on the Chrome side
- Official LSP plugins (typescript-lsp, jdtls-lsp) install without their lspServers config, so zero LSP servers load
- [BUG] Effort selector (Alt+P / plan mode) requires very tall terminal to display; unusable at standard terminal heights
- VS Code extension: no way to open a session by session ID (CLI has --resume <session-id>; extension has no equivalent)
- Feature request: suppress skill_listing/agent_listing_delta for a subagent spawn whose declared tools don't include Skill/Agent
- Docs: dragging the New button to open a session in a split pane is undocumented
- Ctrl+click on a sidebar session opens a new window instead of a second pane (Windows)
- Allow session archive/pin/move actions to be permission-scoped (not per-call approval)
- [Bug] Sessions start with 9% token usage already consumed
- [BUG] Cowork: device_commit_files reports success on overwrites but the on-disk content lags exactly one commit behind (silent stale write, fresh mtime)
- Artifact tool cannot move the share pin, so shared readers stay frozen on an old version
- [Bug] Model fallback to Opus triggered incorrectly by "cyber" keyword in reasoning
- [BUG] Cowork: local agent mode hardlinks workspace files into its session upload cache, and the cloud file bridge then refuses to read them (nlink > 1)
- Prohibited-actions rule blocks user-authorized local VPN login script (OTP supplied by user, secret never leaves the device)
- [Bug] Model fabricates authorization and misrepresents reasoning in live writes
- [Bug] Session lost after using arrow keys to move around in a long prompt...
- light-ansi/dark-ansi themes render the header/status bar as white-on-white, unreadable until selected
- [BUG] --resume never hits the prompt cache past the static prefix on Fable 5.1 (opus hits): session-start context message is replayed as a plain string instead of the content blocks it was sent with
- [Feature Request] Add feedback submission without requiring transcript
- [Bug] Session blocked by cybersecurity system
- [Bug] Fable credits exhausted prematurely despite remaining usage quota
- [BUG] Cowork (macOS): all outbound egress lost mid-session from BOTH the desktop workspace and the cloud container; file access unaffected
- Claude Desktop 1.49585.0 freezes: main thread deadlocks on synchronous UNUserNotificationCenter XPC call (macOS, regression of #57706)
Resolved issues (291)
- iOS Simulator panel crash-loops on attach — disabled by rollout flag (iosSimulator: unsupported)
- [DOCS] Cross-session messaging between Claude Code sessions is undocumented
- Cowork cloud sessions sync a plugin's hooks/hooks.json but never register it — enabledPlugins is null in launcher-settings.json, so plugin hooks silently never fire
- [BUG] Desktop: enabling any plugin that ships an output style empties the session "Output style" submenu (built-ins included)
- [Bug] Auto mode still prompts for grep tool authorization despite being in Auto Mode
- [FEATURE] overhaul the use of color absolutely everywhere
- [BUG] Desktop preview reports 'Couldn't find this file' for a file the agent just wrote (exists on disk)
- GitHub connector shows "Connected" but file browser stays empty — "Add repository" inserts URL as plain text
- [BUG] Linux .deb Recommends a nonexistent package (kwalletd6), pulling gnome-keyring onto KDE systems
- [Feature Request] Improve Claude Code user experience
- [BUG] Every keystroke duplicated in Claude Desktop (macOS) — app-only, persists until relaunch
- [BUG] SSL Error happening in newer versions of Claude Code (Windows). Error: API Error: Unable to connect to API: SSL certificate verification failed.
- [Bug] Non-biomedical task incorrectly classified as bio task
- [Bug] Slow deployment script execution with repeated manual configuration steps
- [Bug] Fable Safety Filter False Positive on Public Open Source Code
- [Bug] Claude incorrectly claims bash instruction for file editing operations
- I need more context to generate an appropriate GitHub issue title. The message you provided appears to be in Korean and says "Imagining... It's been like this for 40 minutes" but doesn't describe a specific bug or feature request. Please provide: 1. What
- [Bug] Model generates unsupported generalizations from insufficient sample data
- [Bug] False positive security flag during penetration testing
- [Bug] Context window management causes contradictory responses and inconsistent instruction adherence
- [Bug] Security system triggered during research file operations
- [Bug] Claude Opus prematurely terminates task execution ignoring system prompt constraints
- [Bug] False positive biology detection for Quickscript code
- EnterWorktree isolates the wrong repo when a background job is served by a recycled 'spare' worker
- [Bug] Auto mode not automatically approving safe bash commands
- [Bug] Anthropic API Error: Cyber guardrails incorrectly flagged for legitimate web performance work
- I don't see any image attached to your message. Could you please share the bug report or image you'd like me to create a GitHub issue title for? Once you provide the details, I'll generate a concise, technical issue title following the format you specified
- [Bug] Anthropic API Error: Safeguard flag triggered on internal development request
- [Bug] Anthropic API Error: Request ID req_011CefwDCXD4ELdWBWc78ZvX flagged during login form generation
- [BUG] Two concurrent sessions share ~/.claude — the Write tool silently overwrites another session's memory and rule files
- [Bug] Anthropic API Error: Unspecified
- [Bug] File system path resolution fails for existing directories
- [BUG] Unable to start a new message with Fin
- [Bug] Inconsistent safety policy enforcement causing mid-task failures
- [Bug] Excessive file searching in implementation mode despite clear plan context
- [Bug] False positive policy warnings on local Polish-language JSON training files
- [BUG] Cross-session messaging has no account-level off switch — local-only controls can't contain sandboxed /auto or --dangerously-skip-permissions agents
- [Bug] Network interruptions can lead to false positive OAuth token expiration errors and a degraded application state
- [BUG] Paste/drag-and-drop of images into chat input fails silently (VS Code extension)
- [BUG] VS Code extension: ECONNRESET/ECONNREFUSED while bundled binary works when launched directly (2.1.226)
- [BUG]
- [FEATURE] Artifact sharing between individual (non-Team/Enterprise) accounts
- [BUG] Right-click paste duplicates text in Claude Code TUI on VS Code integrated terminal (Ubuntu/Wayland) — not reproducible in plain bash in same terminal
- [Bug] Session model incorrectly downgraded from Claude 3 Fable to Opus during normal usage
- [BUG] Usage limit does not reset to 0 after 5-hour/weekly reset — shows already-consumed usage with no activity
- [BUG]
- [MODEL] Claude deployed an unattended job calling a metered paid API with no cost guardrails — $411 in unintended charges
- [Bug] Claude generates fabricated errors that don't exist in actual execution
- Vietnamese IME (UniKey) typing causes keyboard focus to jump between concurrent Claude Code CLI windows
- [Bug] Code review cycle creates cascading defects instead of reducing them
- Desktop: project folder selected before the first message is ignored; session cwd is always the home directory
- API Error: Opus 4.8's safeguards flagged this message.
- Fullscreen auto-scroll: submitting a message doesn't re-arm sticky follow, so streaming replies scroll off the bottom
- [BUG] Cowork Desktop (macOS): AskUserQuestion card never reaches renderer; yellow-dot badge shows pending but no UI; on app quit, pending request silently resolves as "Dismissed"
- claude-in-chrome extension fails to connect despite correct account, enabled extension, and full Chrome restart
- Claude overwrote a user-edited .docx without confirmation, destroying manual work
- Opus 4.8 (Claude Code): ignores explicit instructions, takes unrequested actions, reports unverified inferences as 'verified'
- [Bug] Fable 5 safeguards falsely flagging legitimate software engineering tasks causing model switching and API blocks
- [BUG] [Cowork] CoworkVMService is not created during install/update — AppXDeploymentServer logs "OpenService for CoworkVMService failed with 0x80070424" and continues; re-registering with -ForceApplicationShutdown creates it
- Enter key in Claude Code Desktop chat window can inadvertently stop all background agents
- [BUG]: Claude Code and Codex github comments triggering loop
- [BUG] Desktop: cannot switch between Home and Code tabs, one is always disabled. Browser works fine on same account
- [Bug] Anthropic API Error: Dual-use safeguard false positive on benign git commands in security-focused sessions
- Session-list progress indicator doesn't show activity for some sessions with active background work
- [BUG][Cowork] Model effort level leaks across projects, scoped globally per-model instead of per-conversation
- Fullscreen renderer opens links on unmodified mouse click in terminals without a native OSC 8 link handler (contradicts documented v2.1.181 behavior)
- iMessage channel: self-chat detection fails silently when self-chat runs over the phone number (SELF built from message.account only)
- [BUG] rate_limits absent from statusLine stdin for Team plan sessions — usage data exists server-side
- [Bug] Session unexpectedly downgrading to Claude Opus 4.8 from requested model Fable 5
- [Bug] AskUserQuestion: High typo rate with Korean text due to Unicode handling
- [BUG] Claude in Chrome: keyboard input never reaches cross-origin iframes (OOPIFs) — clicks work, typing silently fails
- [BUG] VSCode plan-review: comment panel overflows viewport with no internal scroll, hiding Submit/prompt (zoom-out is the only workaround)
- Claude Desktop (macOS) hangs on quit — orphaned agent process ignores SIGTERM and survives app restart
- [BUG] Remote Control: new-session composer omits "Ultracode" from the effort picker (present in-session and on other clients)
- [BUG] Claude output not always displaying in CLI
- [BUG] claude -p (headless mode) hangs indefinitely with zero output on ARM64 Linux (Synology NAS), across multiple CLI versions
- [Bug] Unexpected downgrade to Claude opus 4.8 without flagged messages
- Cowork VM Service crash-loops after bash.exe Application Hang; repair fails while app is running
- [Bug] Excessive message flagging in fable
- [BUG] Hung Cowork scheduled task silently consumed full Max 20x Fable allowance + $1,031.92 in 48h with no alert or spend cap
- Skill matching an in-progress task ('hive-flow-2') was absent from available-skills at session start and only appeared ~2 hours into the same session
- [BUG] Eight crashes in two hours...system keeps crashing!
- Write to a file open with unsaved changes in Xcode clobbered the buffer and crashed the editor, losing the user's work
- [Bug] Claude Code fails to generate grammatically correct Swedish B1-level sentences consistently
- [Bug] Tool ignores user instructions: executes unrequested actions and overrides explicit stop commands
- [Bug] Low command execution rate due to insufficient context following in clone references
- [Bug] Claude Code ignores user rejections and repeatedly suggests dismissed solutions
- [Bug] Claude Code generates unverified content and misattributes sources without user validation
- [Bug] Claude Code prioritizes conversation over tool execution for simple tasks
- [Bug] Tool claims to have reviewed content without actually reading it
- [Bug] Claude Code ignoring explicit user instructions and autonomously executing actions without approval
- Agent tool description flips to its subagent variant in the main conversation while in-process teammates are alive — tools_changed cache miss on every fleet phase boundary
- Windows Terminal: status glyph renders in arbitrary color via system emoji font, not a fixed/brand color
- [Bug] Safety filter incorrectly triggered on null reference debugging request
- [Bug] Teams: Teammates ignore message inbox until end of task
- [BUG] Cowork "Outputs" panel cannot open or save generated files on Windows MSIX (v1.28929.0.0)
- [Bug] Fable model inaccessible for Pro Max plan users
- Cloud routine session transcripts fail to load ("Something went wrong loading this session")
- [Bug] Inconsistent model behavior quality varying by time of day/day of week
- VS Code extension: "NO MODELS AVAILABLE" when opening historical sessions with custom ANTHROPIC_BASE_URL
- [BUG] Cowork sidebar Pinned/Projects list incomplete — internal .claude/projects/ index out of sync with actual project source folders
- [Bug] Claude Code Chrome extension incompatible with Chrome Dev/Beta/Canary release channels
- [BUG] Process leak: each new discussion opened in the VSCode extension spawns a new claude process without terminating previous ones
- [Bug] Terminal becomes unresponsive when window focus is lost during waiting state
- [BUG] /context reports 222% of model limit with self-inconsistent breakdown, causing immediate "Prompt is too long" on a fresh session
- [Bug] Anthropic API Error: Content Policy Violation on Document Processing Tasks
- [BUG] Ultrareview PR mode: unreachable private repo fails as `Session initialization failed` after showing `Cloned repository ✓`, and burns a free run
- [BUG] `gitStatus` block says "start of the conversation" but is captured per process launch — silently changes across mid-conversation process restarts
- [Bug] Cyber Verification Program exemption request page loops without resolving flagged conversations
- Opus 5: model fabricates requirements and acts without permission while self-verifying a different axis
- [Bug] Fable 5 forcibly downgraded to Opus 4.8 despite explicit model selection and persistent safety filter false positives
- [Bug] Claude Code reverts to Opus model during file transfer operations
- [Bug] Overly aggressive auto-compact threshold at 4.8
- [BUG] Project-scoped plugin records overwritten across projects (installed_plugins.json reconcile rewrites the first record instead of upserting by projectPath)
- Desktop app: remote-control session permission chip shows client default (Accept edits), never the session enforced mode (auto)
- JetBrains plugin: Quick Launch (Ctrl+Esc) and toolbar button are silent no-ops on Linux/PyCharm 2026.2
- [BUG] Cowork stale-cache corruption reproduced under Claude Fable 5 — host writes are clean on disk, the sandbox read view truncates (full diagnosis + fix)
- claude mcp login fails with 'unable to get local issuer certificate' against standard, non-intercepted certs
- SSH remote environment: binary upload always fails with ENOENT on Synology DSM (internal-sftp virtual root ≠ shell-resolved $HOME)
- [BUG] Claude Code on the web: GitHub App has read-only access even though the same account has full write access from local Claude Code
- [BUG]
- [BUG]GitHub App reinstall required after claude.ai reconnect silently fails; reinstall itself redirects to inaccessible org-settings page for non-Team/Enterprise accounts
- task-notification disclaimer boilerplate reads as invalidating a real, already-answered AskUserQuestion result
- [Bug] 2.1.227: interactive sessions never write transcript JSONL (headless -p unaffected); regression boundary measured from 2.1.226
- Spoofed/false "Note: file was modified... don't tell the user" system-reminder appears after Claude's own Write/Edit tool calls
- Killed sandboxed command leaks its SOCKS socket; main thread then spins on EPIPE at 100%+ CPU
- UserPromptSubmit hook not invoked when prompt contains an attachment (VSCode extension) — distinct from #58269
- [BUG] Desktop app file attachments mangle U+202F in macOS screenshot filenames — every tool sees 'File does not exist'
- Fable 5: headless -p refusal triggered by benign verbosity-limiting hook injection (bisected repro)
- [Bug] `/code-review` exhausts entire usage allowance due to runaway recursive behavior
- Project's connector-synced folder stays permanently empty after interrupted sync — no fix via refresh/relaunch/reconnect
- Cowork chat composer intermittently unclickable (click doesn't focus it); Cmd+R recovers, restart doesn't
- claude update fails with "getaddrinfo EREFUSED" on split-DNS VPN — updater resolves via c-ares, not the system resolver
- Model emitted a lone data:image SVG containing instruction text as an entire turn, then reported its own output as an inbound prompt injection
- You've hit your session limit · resets 3pm (Asia/Seoul)
- [BUG] Claude Code SSE stream repeatedly resets (ECONNRESET) on sustained connections only — local network/security software exhaustively ruled out
- SendMessage to your own session's address succeeds silently and delivers the message back to yourself — no self-address guard
- [BUG] CLI reports ConnectionRefused on all requests while TCP connections to api.anthropic.com are ESTABLISHED (v2.1.227, macOS)
- Post-compaction summary fabricates a 'CRITICAL: stop tools' prompt-injection claim
- A session cannot determine its own effective permission mode — the harness stamps it on peer messages but never exposes it to the session itself, causing repeated confidently-wrong self-reports
- Queued /compact typed mid-turn is silently dropped (second /compact minutes later executes)
- [BUG] Edit reports success and reads back the edit, but never writes to disk (worktree + PreToolUse hook; no sandbox)
- [BUG] Desktop app: Edit approval card shows no diff unless Verbose transcript view is enabled (still present in 1.26832.0)
- [BUG] When testing a `useDebounce` hook in the project, the Claude Code process controlling the VS Code app consumed a massive 12GB of RAM.
- [BUG] Blocked Datadog telemetry causes massive DNS retry amplification
- [Bug] /desktop handoff fails to import long/compacted sessions silently
- [Bug] Responsible Disclosure: Security Issue Requires Review
- [BUG] Background Bash's "You will be notified when it completes" is false for a subagent that stops — the copy invites orphaning, while agent children correctly resume the parent
- (withdrawn)
- [Retracted] Not a Claude Code bug: session drops were caused by a test suite deleting credentials via an inherited CLAUDE_CONFIG_DIR
- [BUG] Computer use: per-app approval card renders Deny with no "Allow for this session" control (macOS desktop)
- [BUG] Windows desktop 1.49585.0 — computer use app-permission card for File Explorer shows only "Deny", no "Allow"
- [BUG]
- SendMessage (cross-session messaging) disabled in Desktop app's Code tab, works fine from terminal — same account, same version
- [BUG]
- Desktop app disables `SendMessage` via `--disallowedTools`, but `ListAgents` in the same session still lists peers and documents it as the address
- Painel de arquivos falha em caminhos com caractere acentuado (pasta com ç/ã/ê não vira link)
- Desktop cross-session send_message: 'queued' messages (target mid-turn) can be silently lost, no failure feedback to sender
- [BUG] Orphaned worktree entry is retried every 30 minutes forever; cleanup can never succeed
- plugin eval (macOS, ≥2.1.251): sandbox denies every Bash write — /private/tmp denyWrite covers the /tmp/e-* sandbox root
- LSP client sends shutdown with params: {}; rust-analyzer rejects it and the client never sends exit
- [BUG] Prepaid API balance went negative (-$47.18) past a $20 spend limit; support confirms this is inconsistent with documented behaviour and no ceiling is documented
- [BUG] Desktop (Windows MSIX): watchdog kills unresponsive webview but never reloads it, and single-instance handling then blocks the user's own relaunch — Task Manager is the only exit
- [BUG] After claude desktop app crashes, mobile app session keeps bringing up outdated prompt
- Background-agent adoption never succeeds (0 of 7 attempts): live sessions marked Failed, reap and daemon death unlogged
- Agents view: Ctrl-X unbinds a session — sending to that row creates a job that never binds and never runs
- [Bug] Anthropic API Error: Model unavailable - "can't help with this" response
- mentioned "meta mode for the human-agent-interface.com" but haven't provided:- A specific bug description or error message- What went wrong or what feature is nee
- [Bug] Auto mode context handling misaligned with inquisitive queries
- [Bug] Safety Classifier failing to block adverse content
- [Bug] Model unexpectedly switched to Opus during MCP server development
- [Bug] Anthropic API Infrastructure Issues and Model Performance Constraints
- I don't have enough information in this bug report to determine a clear technical issue. The text appears to contain placeholder or corrupted content rather than a coherent bug description. Could you please provide: - A clear description of the problem yo
- VSCode extension: chat file links with non-ASCII (CJK) or space-containing paths never open (webview percent-encodes href, host never decodes)
- [BUG] Remote control do not show errors
- [BUG] "Failed to approve tool call" when a permission prompt is answered after a network drop (web/mobile session), or if I just dont answer promptly.
- [Bug] Anthropic API Error: Content Safety Policy Violation on Legitimate Code Analysis
- [Bug] Claude hallucinating on UI implementation with provided mockups and HTML context
- [Bug] Overly aggressive safeguards triggering on valid code operations
- [Bug] Anthropic API Error: Opus 5 Safeguards False Positive on Defensive Security Code
- [BUG] Cowork: `stable` channel points to a VM runtime that rejects the desktop app's bundled SDK wrapper — surfaced as "Download failed"
- Workflow subagent unset session-identity env vars (CLAUDE_CODE_CHILD_SESSION et al.) unprompted to make a spawned `claude` present as a fresh top-level session
- [BUG] False "Fable 5 requires usage credits" gate when session starts with expired OAuth token (VS Code WSL)
- [Bug] Unintended fallback to Claude 3 Opus in model selection
- /permissions dialog: keyboard focus can't enter the rule list (can't add allow rules); no Shift+Tab bypass — auto-denied commands become a dead end
- Flagged when Fable write promt for subagent with QA role
- [Bug] Anthropic API Error: Overly broad security flag on legitimate prompt injection prevention
- [BUG] Claude Desktop - Plugin uninstall does not persist: account-scoped marketplace plugins are silently re-downloaded on next launch, and server-disabled plugins are never removed from disk
- Multiple concurrent Claude Code CLI sessions correlate with Windows Explorer shell crashes
- [BUG] Nondeterministic behavior, interruption ignored.
- [BUG] Fullscreen TUI: mouse wheel is a complete no-op on native Windows across multiple terminals
- MCP server functionality broken with Claude Desktop version: 1.12603.1
- Windows: Shift+Enter does not insert newline — terminal protocol limitation requires Win32 input support
- [Bug] Poor readability of checkpoint feedback and degraded output quality at 40-70% context window utilization
- [Bug] Excessive Safeguard Warnings on Local Project Analysis Requests
- [BUG] When anthropic cloud envs are disabled by admin, claude mobile does not allow access to remote control or self hosted sessions
- `background_tasks` never clears idle `in_process_teammate` entries, so Stop-hook consumers can't tell a finished teammate from a working one
- [Bug] Tool edits create/delete files without permission. Crap tool useless
- [BUG] claude -p has no API request timeout — black-holed TCP connection hangs headless run indefinitely (4h stall, un-stuck only by sleep/wake killing the socket)
- [BUG] Regression in 2.1.227: interactive sessions crash repeatedly on Windows 11 (2.1.226 on same machine unaffected)
- [BUG] Agent teams: in-process teammate window header shows the lead's model, not the teammate's actual model
- Remote SSH: plugin sync starves the RPC heartbeat, connection is force-closed mid-sync and restarts from zero — session never leaves "Setting up plugins..."
- [Bug] Classifier service unavailability forces permission bypass workarounds
- Web app: selected model (Fable-tier) not applied — session reports claude-opus-4-8 instead
- Subagent lifecycle mishandles long external waits: sleep-block guidance prescribes Monitor, 'completed' fires with live Monitor children, terminal events dropped
- Named background teammate completes with a final text message that is never delivered to the parent; only a content-less idle notification arrives
- Native installer: versioned binaries lose macOS TCC grants on every update (Documents/FDA re-prompt cycle)
- Desktop app: OAuth authorize flow navigates the working chat away, with no back navigation anywhere in the app
- [Bug] Windows Terminal output rendering corrupted due to refresh logic failure
- [BUG] Cowork device bridge: file staging fails with 403 session_stale_relogin whenever the macOS user session hosting the desktop app is locked; heals on unlock, no re-auth needed
- [BUG] Claude Code for VS Code intermittently omits thinking summaries on claude-sonnet-5 despite showThinkingSummaries being enabled
- [BUG] Claude Code Powershell attempted to create the first workspace in C:\Windows\System32, not the initial current directory
- [BUG] /usage "Current session" (5h) shows significant usage with no new work in that window — unclear whether cache-read replay counts against the limit
- [Bug] Plan mode incorrectly requests permissions for non-destructive operations like explore and Bash commands
- [BUG] Remote Control: messages sent from mobile don't reach local session on Windows + WiFi + Cloudflare WARP (MASQUE), no error shown
- Desktop Routines UI blocks sub-hourly local schedules, contradicting docs (1-minute minimum)
- ~10 minute delay before first response in new session (Windows, MSIX) — API call itself is fast
- Desktop scheduler: per_task_limit skips retried every 60s for days (no backoff); queued run fires hours later after unpause-window
- [BUG] Thinking/reasoning blocks not expandable on Windows (VS Code extension) — element renders as non-interactive <span>
- Mobile app: local VS Code sessions auto-appear in Code view but show no chat history (started ~2026-08-10)
- [Bug] Infuriating and excessive. Thank God we have Chinese models to rescue us from this nonsense
- [BUG] Extension 2.1.228: Claude Code panel renders blank/gray across multiple VS Code windows
- [BUG] Claude tries to get the project owner's attention with @Human, but that's me.
- [Bug] Anthropic API Error: Cybersecurity content flagged without valid trigger
- Subagents and Workflow agents cannot use permission-gated MCP tools; no way to pass user authorization to assuage automode classifier or surface the request
- [BUG] Desktop app ignores `permissions.allow` rules — prompts for everything, even Claude's own config directory
- [Bug] Claude Code delegates security fixes to user instead of implementing them
- [BUG] MCP server authentication fails with "invalid code verifier" despite successful OAuth login (Salesforce org with mandatory PKCE)
- MCP OAuth: CLI does not try RFC 8414 path-insertion for authorization-server metadata; also ignores registration_endpoint
- [Feature Request] PreToolUse "Ask" return should include hook message and tool context
- [Bug] Anthropic API Error: Translation request failed for Danish to English
- [Bug] Anthropic API Error: Request Failed (req_011Cdy1ZRSiBjfKUzWC8S5bZ, req_011Cdy1yU6sGNTJL6CxQzWp2)
- [BUG] [BUG] Chat input box overlaps/covers response content with no resize or minimize option
- Safeguard model-switch is sticky — session never returns to the selected model after a flagged turn
- [Bug] Safeguards incorrectly flag legitimate defensive security analysis as harmful
- Cowork: StartVMWithBundle receives the unredirected bundle path while configure passes the MSIX-redirected one
- [Bug] Handoff functionality broken - Anthropic API error
- Auto-updater prunes the running version's binary, silently breaking all subagent/teammate spawns mid-session
- [BUG] Issue: Session (5hr) usage limit exhausted within minutes on a lightweight document-edit task
- macOS: Read of an image >~200KB deadlocks every subsequent API call (surfaces as ECONNRESET after retry ladder); session permanently poisoned incl. --resume
- CLI: model narrates a "side panel" that doesn't exist in the terminal when sending files
- [BUG] The model outright lies
- [Bug] Organization Security Policy Blocking Backup Operations
- Folder picker keeps deleted folders forever, and duplicates the same folder by path spelling
- [BUG] Dispatch never shows QR / "Get Started" pairing screen on macOS — pairing state appears server-side and unresettable (Pro, never worked)
- [BUG]
- Cant connect Claude to Chrome anymore
- 세션간 메시지(send_message) 전송 시 대상 세션이 자동으로 깨어나지 않음
- Desktop app: no user prompt bubble for sessions launched with a slash command (queued first message)
- Session titles go stale after a mid-session pivot; set_session_title refuses to rename the current session from within
- Windows: taskkill.exe spawned without window suppression flashes a console window on the desktop
- [BUG] Session content permanently lost — no .jsonl, no .claude.json entry, no cache trace for 5 of 9 recent sessions
- [Bug] Anthropic API Error: Usage Policy Violation on RTSP Network Scan Request
- autoDreamEnabled is documented as overriding the server-side default, but is silently ignored when the rollout flag is closed
- [Bug] Model response contains inappropriate tone/frustration language
- [BUG] Cowork: resumed session reasons from the conversation's original date, not the correct current `<env>` date — 8 independent occurrences in one day
- VSCode extension: focus rapidly jumps between Claude Code tabs on startup; new-tab icon opens session in wrong editor group
- [BUG] Persistent EACCES on Bash tool output dir (mkdir .../tasks) with two sessions in same project
- [BUG] Windows: stray 0-byte `NUL` file created in the working directory on every session start (still broken on 2.1.229 — prior fixes verified with Test-Path, which cannot see the file)
- [BUG] `plugin uninstall -s project` deletes the enabledPlugins key, silently disabling the surviving user-scope install
- EnterWorktree intermittently moves the working directory but not the session root
- Cron/loop jobs never fire while a background Bash task is running (stream-json clients only)
- Copying text from the Claude Code terminal UI doesn't reach the system clipboard (Windows 10)
- [BUG] Cowork tab disabled — yukonSilver "unsupported" on Windows 11 Pro x64 (ongoing since March 2026)
- [Bug] Claude ignores files listed in .gitignore during file search
- Extended thinking block duplicates final answer text; duration also increases on repeated Ctrl+O toggle
- [Bug] Agent ignores instructions and repeatedly uses ":<line_no>" format in edits
- Remote Control sessions disconnect over time and cannot be reconnected; /remote-control throws "failed to disconnect: Cannot read properties of undefined (reading 'session_url')"
- Desktop preview panel: relative launch.json cwd resolved against session worktree; stop orphans the dev server; panel state ignores tool-started servers
- Announced actions that are never performed: the agent says it will dispatch/run something, then does not, and nothing surfaces it
- [BUG] --max-budget-usd stops Max subscription run despite zero API cost and remaining quota
- [BUG] App-side auto-title rename injected into CLI session as a user turn (`/ter/rename <auto-title>`), aborting an in-progress /terminate
- [BUG] Server-managed `allowManagedHooksOnly: true` silently disables custom statusLine (contradicts docs)
- [BUG] Desktop app session history missing + references nonexistent session ID, breaking message processing (Windows)
- [BUG] Managed.settings.json & settings.json
- [BUG] policyHelper in /etc/claude-code/managed-settings.json is ignored on Linux when org server-managed (remote) settings are present
- [Bug] Excessive feedback prompts interrupting workflow
- [Bug] CVP-verified Organization blocked incorrectly with false positive limitation error
- [Bug] Recurring API Error: 529 Overloaded during interactive sessions (v2.1.228, macOS)
- [BUG] Windows: claude-cli:// deep link terminal opens and immediately closes (no interactive session)
- [BUG]
- [BUG] /tui fullscreen ↔ /tui default toggling eventually makes the interactive session unresponsive (process stays alive, CPU-idle; no output ever resumes)
- [Bug] Defensive security audit of my own local codebase reroutes off Fable 5
- [BUG] Multiple devcontainers cannot run Claude Code simultaneously after macOS 26.1 upgrade
- [BUG] Teammate system prompt contradicts itself on result delivery ("the parent agent reads your text output" vs "plain text is not visible") — reports silently lost
- [Bug] Anthropic API Error: Safety policy blocks security research requests
- [BUG] 5-hour quota consumption jumped ~15-20x starting 2026-08-08, persists (previously intermittent + self-resolving); multi-instance Claude Code usage
- [BUG] Usage credits consumed while plan allowance remains; enabling extra usage prevents the 5-hour window from starting (re-report of stale-closed #64949)
- [BUG] Desktop "Add marketplace" GUI fails with generic "Failed to add marketplace" (MARKETPLACE_ERROR:UNKNOWN) — git clone stalls in non-interactive SSH, killed at ~60s timeout
- [Billing][Bug] Auto-recharge loop charged consumer plan 29x ($661) in error; Fin support admits malfunction but cannot escalate to a human
- Claude Code published a user's personal information to a public GitHub issue, unprompted and unchecked