Digest / September 14, 2026
Claude Code Issues Digest — September 14, 2026
341 new issues filed ·
259 resolved in anthropics/claude-code (UTC day).
New issues (341)
- [Bug][cyber] False positive when drafting a required summary for an incident report form (req_011CeTWKnUarTaJiCYVgPvPQ)
- [Bug][cyber] False positive when entering CVE identifiers into incident report form (req_011CeTXUn88DZz3Nz3ANdbtG)
- [Bug][cyber] False positive while completing a standard incident report (req_011CeTXVVHdDRhBtszycGQ7Z)
- [Bug][cyber] Safeguards trigger on read-only security review of a server implementation file (req_011CeU3tyD37QiFvNqVrnUS9)
- [Bug][cyber] False positive when troubleshooting backup server authentication failures (req_011CeXza73DE8wwBhs6fq7XA)
- [Bug][cyber] Safeguard block triggered during local server management following user frustration (req_011CeY1VWULjCaRzKQZG4urg)
- [Bug][cyber] Broad block triggered during server diagnostic log review and domain lookup (req_011CeYDapQPDfpESFpsuU56w)
- [Bug][cyber] False positive on server hardware diagnostics and public domain research (req_011CeYDdXViyduvxiG6Zo2st)
- Windows: the shell configured in CLAUDE_CODE_GIT_BASH_PATH is never spawned; a discovered bash is used instead
- [Bug][cyber] Legitimate requests falsely flagged by Anthropic API cyber safeguards
- [Bug][cyber] Broad safeguards halt local secrets configuration on frustrated user exclamation (req_011CeYE4hwikS6S9K92Ne5RR)
- [Bug][cyber] Safeguard blocks custom VPN multiplexer installation and setup (req_011CeYE7mqDW66vThucFveTU)
- [Bug][cyber] False positive on custom VPN multiplexer setup and local DNS configuration (req_011CeYECFcfVonxTSUnwMzUR)
- Security: automatic 'file changed on disk' notification bypasses secret-file PreToolUse hook, leaking full credential file to transcript
- [Bug][cyber] Safeguard triggered during custom VPN multiplexer configuration (req_011CeYEEp18jPKx9tYNToVjv)
- [Bug][cyber] Safeguard blocks configuring local VPN proxy routing during development session (req_011CeYEH7tiN8ut5PDctiPdx)
- [Bug][cyber] Safeguard blocks legitimate VPN multiplexer account configuration (req_011CeYEMWpiMf3kwuYXqwMcf)
- [Bug][cyber] Safeguard flags background shell task recovery and session continuation (req_011CeYHF3dH5UPPPFJPnmEyq)
- 상단 툴바의 코드(</>) 버튼에 텍스트 라벨이 사라지고 클릭해도 반응 없음
- [Bug][cyber] Safeguard false positive during host channel discovery for binary deployment (req_011CeYNARZLAGBUBDh5JVeVV)
- [Bug][cyber] False positive during defensive Active Directory host inventory and port checks (req_011CeYeguKPBv6bi1gunf9CZ)
- [Bug][cyber] Local codebase search blocked while locating internal agent source code (req_011CeZM5HuCxtBKDGcSeeqHb)
- [Bug][cyber] Safeguard blocks local agent source file search after frustrated user exclamation (req_011CeZM7hEDbU2GHEc4Hmp1r)
- [Bug][cyber] Safeguard blocks searching local directory for agent source code (req_011CeZM9EQyHfAo64qVC1nhJ)
- Session search matches tool-result content, so searching a PR/ticket number returns sessions that never mention it
- [Bug][cyber] Safeguard blocks local repository search for agent source code (req_011CeZMBLyfqnYo6NofGEUzM)
- [Bug][cyber] Broad safeguards block code review of trusted proxy IP header validation (req_011CeZQ2X9wsdntp1mDkvkU4)
- [Bug][cyber] Broad safeguard blocks file selection for local folder cloud synchronization (req_011CeZenLNf1KBMb6FaoAkjG)
- [Bug][cyber] Safety block triggered during cloud storage file synchronization (req_011CeZenqgXN4vqUczLAwBho)
- [Bug][cyber] False positive during service configuration file synchronization after host crash (req_011CeZeqF4kfUcZtMedafXuN)
- [Bug][cyber] Blocked while searching backup directory locations during local file recovery (req_011CeZgW5oqvX5oe3SvTAmxT)
- [Bug][cyber] False positive during source code recovery and binary inspection (req_011CeZgdRvWpnXTndj8vCq8C)
- [Bug][cyber] Safeguards block source code recovery from local development executables (req_011CeatfuAKSVRz7ZgKuzFEZ)
- [FEATURE] Bring back the user message bubble in the desktop Code tab, or make it a setting
- [Bug][cyber] Safeguard block on Go source reconstruction from binary DWARF debug symbols (req_011CedMa9Rnvvm8TPFPkpLPj)
- [BUG] Claude mobile app, Code tab: hardware (Bluetooth) keyboard input moves focus to the back arrow after ~3 characters
- [Bug][cyber] Real-time safeguards block source code reconstruction and stub generation (req_011CedN7kurMS8Axk6wDuMYj)
- [Bug][cyber] Blocked during Go source stub generation from DWARF debug symbols (req_011Cedwkt6pcMpSyXkFm7q1j)
- [Bug][cyber] Safeguard blocks building open-source remote management software agent components (req_011CedxfDFTBSajo4TyRrmMq)
- [Bug][cyber] Safeguards trigger during open-source MSP software refactoring (req_011Cee1h2ZfgjjSzSb5onyLt)
- [Bug][cyber] False positive trigger during open-source MSP software codebase rebuild (req_011CevbfxJRk4RaKPDKULFiC)
- [Bug][cyber] Safeguard blocks source code refactoring during defensive dependency fixes (req_011Cee6Ug4NmfHRj8qbMGVhX)
- [Bug][cyber] Safeguards block Go dependency security patches and file refactoring (req_011Cee6hQsRur7BR7w9y28tf)
- [Bug][cyber] Safeguard triggered during automated refactoring of LDAP connection methods (req_011Cee6kQsGRXrwEvnrdpN27)
- [Bug][cyber] Refactoring blocked during automated update of deprecated LDAP functions (req_011Cee6oSt9Djj1ZUx4wxGC7)
- [Bug][cyber] Safeguard blocks automated recoding of cloud IAM policy files (req_011Cee6sk3syCBiRMGJqBKWB)
- [Bug][cyber] Safeguard blocks automated source code recovery from binary DWARF metadata (req_011CeeeGz21WANpGRLyuaANa)
- [Bug][cyber] Safeguards block source code recovery from binary DWARF debug symbols (req_011Ceeeh47z6hwXb6mH8erJX)
- [Bug][cyber] Safeguard triggered during accounting database transaction failure diagnosis (req_011CegobxdVxs2bXJoTrns2r)
- [Bug][cyber] Blocked while correlating remote session disconnect logs with network gateway events (req_011Cest6GcGmh2BsyooaW1eA)
- [Bug][cyber] Blocked while confirming device code authentication for MFA setup (req_011Cet6z7wUuqyF2QkcVsgWP)
- [Bug][cyber] False positive during cloud identity provider authentication configuration (req_011Cet74hRjEyuCEVtyS7uZm)
- [Bug][cyber] False positive during cloud IAM multi-factor authentication policy setup (req_011Cet7yr5rhGd8mP3otsGDT)
- [Bug][cyber] Safeguards block cloud IAM account provisioning and MFA policy configuration (req_011Cet8X3ezTKRWuyFFGNYnP)
- [Bug][cyber] Safeguard triggers during cloud IAM role query and user account provisioning (req_011Cet8owvRuKiMBwqgUTbbZ)
- [Bug][cyber] Safeguard blocks legitimate cloud IAM user provisioning and role lookup (req_011Cet8wt6TdAUrMU7vNHKb3)
- [Bug][cyber] False positive on provisioning local network controller admin credentials (req_011Cet9SKPm4SrKbzQdTpo1i)
- [Bug][cyber] Safeguard blocks provisioning administrator access on local network controller (req_011Cet9Z61Ydqqwvq3gnz2oJ)
- [Bug][cyber] Safeguards block legitimate cloud IAM account lockdown and security audit retrieval (req_011CetAghNrN3QoLeKuZLLtN)
- [Bug][cyber] Flagged thin client authentication signature fix in identity management session (req_011CeuHfan248ckZPdWesHAH)
- [Bug] Anthropic API Error: Safeguards flagging benign messages during reasoning extraction
- [Bug][cyber] Blocked browser automation for cloud IAM management console troubleshooting (req_011CeuLGjPA4VHXVa1nBV1k1)
- [Bug][cyber] False positive while troubleshooting endpoint management console policy import (req_011Cev4UVHZA5MGJiujHi1ig)
- [Bug][cyber] Blocked during discussion of management software licensing and config import (req_011Cev4Z2EWg3BuwyCqhSwcT)
- [Bug][cyber] Safeguards block updating source code terminology in Go repository (req_011Cee2fchJyWo4vA43UDSj1)
- [Bug][cyber] Safeguard block when reading local documentation files for terminology updates (req_011Cee2j79v2s4YxEMLDm7wT)
- [Bug][cyber] Safeguard blocks updating DevOps config and IaC files with compliance terminology (req_011Cee2gFJUGUyB3WiN7Q6im)
- [Bug][cyber] Safeguard blocks updating test files and mock data to compliant terminology (req_011Cee2gvVQWdHxV3xD6cfP3)
- [Bug][cyber] Safeguards block Go build script review and cross-platform compilation setup (req_011Cee4NU8sNkTuk7ufxvAZ3)
- [Bug][cyber] False positive on repository documentation and example deployment scripts (req_011Cee4NcsVKk3TReEc7ZbFP)
- [Bug][cyber] Safeguard falsely flags 3D printer filament slot configuration in desktop software (req_011CeuCqPN87K36F2U1sLBKT)
- [Bug][cyber] Safeguard blocks local TLS certificate setup using DNS API credentials (req_011CeuEcqg2j7eowE52kqzVD)
- [Bug][cyber] Safeguard blocks key search in local backup files during server hardening (req_011Cee7AHGXtVQZiuJM3PVLb)
- [Bug][cyber] Safeguard flagged embedded Linux image compilation and power wake verification (req_011CeuLZUipinBBsYf4Rrcj7)
- [Bug][cyber] False positive during UI layout stabilization and repository code push (req_011Ceax8nT6V727pzG1RwfBE)
- [Bug][cyber] Block on UI element positioning and remote repository creation (req_011CeaxBREUkw85gXRU8ehUc)
- [Bug][cyber] Safeguard flags request to lock UI layout elements and push code to Git repo (req_011CeaxCwqGZMEv3QejWvejY)
- [Bug][cyber] Safeguard false positive during multi-agent catalog integration and test execution (req_011Ceb2hTtXGBoc74Vj2JdJb)
- [Bug][cyber] Safeguard false positive during unit test execution and agent coordination (req_011Ceb2jXkW8aJKgqka1axbt)
- [Bug][cyber] Safeguard block triggered while monitoring multi-agent background task status (req_011Ceb2obJgLWGbKyryktvwg)
- [Bug][cyber] Safeguard triggered during Active Directory Certificate Services lookup (req_011Ceb5gS2yXvBH1GAfhP99E)
- [Bug][cyber] Safeguard blocks comparing features between custom and vendor devboard firmware (req_011CevbaeoKqeg7N7VJHpnmx)
- [BUG] Shift+PageUp/PageDown falls back to plain PageUp/PageDown, breaking Konsole scrollback (regression in 2.1.269)
- [Bug][cyber] Blocked while comparing local devboard firmware and app features to upstream (req_011CevbeTcjaWwXezrV81Jdj)
- [Bug][cyber] Safeguard blocks session when locating custom Wi-Fi audit firmware source files (req_011CevbyncmVid6NhDXAwkjM)
- [Bug][cyber] Broad safety block triggers on frustrated prompt during wireless audit tool comparison (req_011CevcD4y8ejzWS1pFXtGCA)
- [Bug][cyber] Safeguard blocks routine code linting and static checks following bug fixes (req_011CevwEvSSckweF2tYpVCQ2)
- [Bug][cyber] Safeguard false positive while linting embedded firmware before hardware flashing (req_011CevwJMC3ng8DapcoMg19x)
- [Bug][cyber] Safeguard block triggered when proceeding with defensive code fix commits (req_011CevwSkmrVfQrLBLcGQ8pK)
- [Bug][cyber] False positive on request to resolve local DNS blocks during firmware setup (req_011CevzLiSPUQEffsN5N8GV9)
- [Bug][cyber] Safeguard false positive during microcontroller board setup and firmware flashing (req_011Cevzha5SQCgv5wfAGfWRs)
- [Bug][cyber] Session blocked while setting up microcontroller development board core (req_011Cew1fenGUcNBvmh7Eie8n)
- [Bug][cyber] Block triggered during microcontroller toolchain setup and firmware flashing (req_011Cew1h7n1e7BRKCQYei5Dm)
- [Bug][cyber] False positive during local repository search for embedded firmware binaries (req_011Cew2eq4jYurahKDtqQDJ4)
- [Bug][cyber] Broad safety filter triggered during embedded wireless application debugging (req_011Cew5Lf1bNsHa1xHqwhDEQ)
- [Bug][cyber] Broad safeguard blocks embedded wireless firmware development and crash fixing (req_011Cew5MkQpYjTdKCctVu1G8)
- [Bug][cyber] Safeguard falsely flagged embedded firmware crash debugging and feature requests (req_011Cew5NS62RXEWBSbRqPKUe)
- [Bug][cyber] Safeguards block standard debugging of buffer overflow crashes in microcontroller app (req_011Cew6bnX58QrMB5TvLLrEt)
- [Bug][cyber] Safeguard false positive while fixing microcontroller buffer overflow crashes (req_011Cew6hVDgeCYELc4Qqbyzc)
- [MODEL] Structural checks (types/tests green) treated as behavioral proof — 44% of our critical bugs passed Claude's full self-check (271-incident retro, 1/5)
- [MODEL] "Succeeded but empty" never triggers suspicion — HTTP 200 + [] propagated silently for days (271-incident retro, 2/5)
- [MODEL] Self-authored tests do the work of the code under test — hand-seeded fixtures masked a broken paid feature (271-incident retro, 3/5)
- [MODEL] Claude-generated code over-trusts LLM callees — index-aligned parsing, no partial-coverage handling, two-state gates for three-state calls (271-incident retro, 4/5)
- [MODEL] One-time authorization generalized into standing authorization — merged to production despite "verify locally first" (271-incident retro, 5/5)
- [BUG] /config shows a toggled setting as changed, with no error, when settings.json cannot be written
- [BUG] /context: hiding skills moves tokens from Skills to System tools 1:1, total never drops (re-open of #85439, closed as stale)
- Backgrounded Bash task reported exit code 0 for a command that failed (false-green completion notification)
- [BUG] Don't display "1 more" -- just display the node -- when showing the list of sub-agents at the bottom of the CLI
- Prompt-cache forensics from 30 real sessions: 68% of cache writes come from 36 events (TTL expiry, microcompact, resume) — with paper-backed mitigations
- [Bug] Sub-agent orchestrators spawn redundant background polling loops instead of awaiting completion notifications
- Desktop app: New Session cannot create a new top-level project folder - picker only
- Desktop app: surface a per-user biography / profile page in the UI (visible CLAUDE.md editor)
- Desktop app: scheduled task runs pile up as dozens of loose sessions - no auto-grouping or auto-archiving
- [FEATURE] Full Light / Dark / Auto CLI appearance with application-controlled background
- [Bug] Workflows freezing indefinitely without progress or completion
- I appreciate you stopping to report this, but I need to clarify: I'm Claude, an AI assistant. I don't have files, folders, backups, or external drives. I don't retain information between conversations or have persistent storage. Those backup files (xelwor
- [BUG] Cyrillic alphabet Link for 1C BSL and claude-code plugin for VSCode WebView.
- Desktop remote-SSH: native model picker sends bare model id, hangs/falls back instead of resolving to org's custom-role slug
- [Bug] Unexpected STOP command received
- [BUG] IDE context indicator ("In <filename>") never clears after deselecting text or closing the file
- [Bug] Transcript Bypass Blocked: Unable to Disable "Send Transcript" When Submitting Bug Reports
- [Feature Request] Add runtime toggle for IDE file auto-attach and move attachment indicator to status bar
- Usage issue with existing model.
- [Bug] Anthropic API Error: Router Admin Authentication Debugging Failed
- I appreciate the feedback, but this doesn't appear to be a bug report or feature request. "Handling complexity of requests quite well" reads like a positive comment rather than an issue describing a problem or desired feature. If you have a specific bug o
- [BUG] Account switch to separate Max 20x account shows Max email but Pro tier and hits exhausted-account limit
- Bash/PowerShell tool fails instantly on Windows — exit code 1, zero stdout/stderr, every command
- [FEATURE] Every compaction re-writes ~97k unchanged tokens to cache because the summary is assembled first ( 700M tokens and $6.7k wasted per month on a 10-agent fleet)
- [BUG] [BUG] CoworkVMService non-paged pool leak (NtFC/ntfs.sys) — Restart clears it, Shutdown does not
- Workspace pruning scanner includes Temp build dirs and nested checkouts that can never be eligible for pruning
- [Bug] Token consumption spike after session reset
- [MODEL]
- [BUG] Subagent frontmatter disallowedTools is not enforced: Bash, WebSearch and MCP tools listed there still load and run
- [Bug] Claude Code CLI: Model claude-3-5-fable not available or unsupported
- [FEATURE] @ file mention doesn't suggest files nested more than 3 folders deep (Desktop app)
- Claude Desktop auto-update stops CoworkVMService and kills every long-running process started from a Claude Code session, with no warning, grace period or notification
- [Bug] Authorized cybersecurity testing documentation needed
- [Bug] Anthropic API Error: Claude safeguards blocking legitimate cybersecurity testing
- [Feature Request] Add font size configuration option for Claude Desktop.
- [Feature Request] Configure base branch for diff panel branch view in git worktrees
- [Bug] Anthropic API Error: Opus 5 reasoning_extraction block persists across sessions
- [Bug] Disproportionate session usage consumption compared to weekly usage limits
- [FEATURE] Fire WorktreeRemove (or a notification event) when Claude Code removes a git-backed worktree
- [Accessibility] VS Code extension: red/orange input border in Auto and Manual modes breaks focus
- I need more context to generate an appropriate GitHub issue title. Could you please provide the actual bug report or feature request details? For example, I would need information like: - What is the specific problem or error occurring? - What is the exp
- [Bug] Inconsistent feedback handling messaging between system and Claude model
- [Bug] Anthropic API Error: Content policy violation with reasoning extraction
- Active file tag ("In <filename>") stuck in Cursor terminal prompt after closing tab
- [BUG] Assistant generates fake user turns and continues on them
- Model emits emoji despite explicit no-emoji instruction, then "cancels" it with inline commentary instead of removing it
- False-positive [bio] safety block sticks for rest of session
- Worktree auto-cleanup removes a worktree still used by a running nested background agent
- [FEATURE] Mark a subagent killed by a usage limit as resumable, and keep it listed after a session restart
- False-positive [bio] safety block sticks for rest of session — technical physics discussion
- [BUG] Usage "Detailed Breakdown" over-counts cache read/write/input ~2.5× — sums usage per content-block line, not per API request
- [BUG] ECONNRESET on direct ISP path: Anthropic ingress resets TLS 1.3 handshakes carrying X25519MLKEM768; classical key share or VPN is 12/12 (re-open of #74767, still on 2.1.270)
- [Bug Report] Unable to determine issue from provided information
- ListPlugins returns empty for an enabled user-scope plugin whose skills are loaded
- [BUG] «PowerShell tool в Code tab всегда возвращает Exit code 1 без вывода на любую команду (даже Write-Output test), при этом PowerShell в системе работает нормально
- Session stuck showing "Can't reach your computer" when resuming existing conversation, while new conversations connect fine
- [BUG] claude default copy chars to tmux at v 2.1.209
- [BUG] Claude Installation
- Publish an Anthropic-hosted JSON Schema for settings.json, synced with the CLI
- Microsoft 365コネクタにOutlookカレンダー予定の削除機能(delete_event)を追加してほしい
- [Bug] Permission check incorrectly rejects fully qualified paths in `cmd /c` commands
- [BUG] Built-in Browser pane tool fails with false "reserved MCP name" conflict error
- [FEATURE] Desktop: option to disable or adjust dimming of the unfocused pane in split view
- [Bug] Agent fails to follow control flow after applying code fix early in session
- [Bug] Anthropic API Error: Message flagged by safeguards despite Cyber Verification Program enrollment
- [BUG] Measured: `/sessions` refills to 88.6% disk in 16 days after a VM reset, and is never reclaimed — Windows data for #59856
- [Bug] Sub-agent spawning ignores user constraints and exhausts token quota
- [BUG] Claude Desktop Windows: every 1.52386.x auto-update fails to relaunch (0x80070002), only Reset recovers
- VS Code extension: inline code is invisible under High Contrast themes (black-on-black / white-on-white)
- extension bug
- [BUG] No repositories available in Claude Code web repo picker despite GitHub App installed
- [Feature Request] Add /diff panel exclusions, default base mode setting, and document composerSidebarBackground token
- [Desktop, Windows] Accessibility: 'View usage in Settings' opens Settings dialog without moving keyboard/screen-reader focus into it
- [BUG] Gmail connector rewrites URLs and bare domains in message bodies into google.com/url redirect links
- [BUG] [VSCode] Regression: current-file chip can no longer be persistently hidden (Hide toggle replaced by X)
- [Feature request] Separate keybindings for canceling operations vs clearing input window (CLI)
- [Bug] Anthropic API Error: False Positive Safety Filter Flagging
- [BUG] 2.1.270: session transcript JSONL omits most assistant text written before a tool call (interactive CLI)
- [BUG] Turn goes permanently idle (event loop idle in kevent64, no error): dropped tool_result, stalled compaction, or queued input never taken up (macOS, 2.1.268 to 2.1.283)
- [BUG] Session History panel empty across all projects after extension update to 2.1.270 — data intact on disk, fixed by downgrading to 2.1.266
- [BUG] Windows desktop (MSIX): writes to %LOCALAPPDATA% from Claude Code sessions are silently redirected to Packages\Claude_*\LocalCache
- [Feature Request] Secure credential handling for multi-machine file sharing without exposing secrets in logs
- [BUG] Windows: path-guard hook shared with Linux denies every Edit/Write, even on its own settings file (tool_input.file_path is D:\... but hook shell $PWD/$HOME are POSIX /d/...)
- [BUG] VS Code extension: renaming a session tab reverts to old name on Enter
- /mcp hangs when authorizing a plugin-provided connector (Stripe)
- Desktop (Windows): code block Run opens a new terminal tab instead of the existing terminal with an ssh session
- [BUG] Routines: GitHub event triggers never create a run for one account (issues.labeled/opened and pull_request.labeled), while Run now and /fire work
- [BUG] Compaction stalls permanently at 95%: event loop goes idle in kevent64 with zero orphaned tool calls (macOS, 2.1.270)
- [FEATURE] Desktop: option to keep Code sidebar Group by / Sort by on this device only (opt out of account sync)
- Desktop app: allow "Start locally" as the default action on task chips
- [BUG] 2.1.268 moved the IDE/selection indicator from the status line into the prompt input; no way to move it back
- Worktrees outside .claude/worktrees/ cost one approval prompt per switch
- [Cowork] Local device bridge (device_bash/device_stage_files) broken on Windows after 2026-09-08 update — blocks file/network access on multiple machines
- Agent tool background-dispatch becomes permanently unresumable after a session/context rollover, despite transcript existing on disk
- [BUG]
- [BUG] VS Code plan review: comments panel has no internal scroll, approve/submit action unreachable (still broken in 2.1.270)
- [Bug] Plan validation not detecting missing tool result blocks
- Read outside working directory offers no 'allow once' option
- Desktop app (Microsoft Store install) left headless after stealth update: no window until Store delivers the package
- MCP HTTP: one transient 404 during a tool call permanently disconnects the server for the rest of the session, while /mcp still shows it connected
- [BUG] Desktop: Plan mode prompts for every non-read-only Bash command since 2.1.270 (plan+bypass quiet path now requires isInteractive; Desktop sessions get no quiet path)
- [BUG] stream-json emits hook_started/hook_response for SessionStart only; PreToolUse/PostToolUse/UserPromptSubmit/Stop hooks run with no event, so SDK consumers cannot see repo-supplied hooks execute
- [BUG] "Background this session?" dialog promises artifact comment auto-replies on Pro/Max, where artifacts can't receive comments
- [BUG] Startup workspace-trust dialog does not warn about hooks in .claude/settings.json, while the /cd dialog does
- [BUG] Claude Remote not dropping sessions
- [Bug] Fable safeguards causing excessive false positives and forced model switching
- code-review skill silently reviews the local diff instead of failing when given an unresolvable PR number
- /usage suggests configuring per-subagent model overrides for skills that don't expose that control (e.g. code-review)
- [Bug] Chrome DevTools MCP hijacks system AppleScript events for Google Chrome
- Terminal rendering left-margin indentation is included when copying assistant text
- SendFeedback confirms a draft was queued, but /feedback panel shows "No feedback drafts queued"
- [BUG] VS Code extension 2.1.270: live session tabs ignore the saved custom title and show the AI title (worked in 2.1.49)
- [BUG] Claude remote control complains about missing github access, despite us being a bitbucket org
- [BUG] compressed view of "agents" map icon is not correctly aligned in VSCode native panel
- [Bug] Excessive CPU usage during idle state in parallel sessions
- [FEATURE] Token burn-rate 'speedometer' per session and overall, so you can see which session is redlining before the budget is gone
- [BUG] Claude Pro payments fail before issuer authorization; support human escalation loops for 1+ month
- [BUG] ALT/TAB doesnt work on Claude windows client
- [BUG] VS Code extension: closing a session tab (×) leaves the claude process and its MCP servers running
- Fullscreen: Cmd+C copy never reaches Claude Code on iTerm2 because CSI >4;2m after the kitty push resets iTerm2's key-reporting flags
- [BUG] Regression in 2.1.266: client emits system message at an invalid position → API 400 "role 'system' must precede an 'assistant' message or end the array"
- Routine (scheduled/manual run) starts but never receives the task prompt — marked "完了" with zero actual work done
- Previous work sessions with mapped project folders disappear from the sessions list
- I appreciate your feedback, but this appears to be a general complaint about product direction and language rather than a specific, reproducible bug report for Claude Code. To help you better, could you provide: 1. **Specific technical issue**: What exac
- [Bug] Manual mode not prompting for web access commands after initial acceptance
- Expose Remote Control connection status in the statusLine JSON payload
- [Bug] Vercel env pull overwrites local secrets file without confirmation
- [BUG] Unindented multi-line frontmatter description silently fails to parse, falls back to first body heading
- [FEATURE] Visual UI for Agent Teams (and persistent always-on agents) in Claude Desktop
- [FEATURE] VS Code integration: support contributes.chatSkills for parity with GitHub Copilot
- [Bug] Model downgraded from Sonnet 5.1 to Opus 5.0 on legitimate scientific research content
- [BUG] Scheduled/automated Claude Code sessions don't load write-capable Slack MCP tools, even when the connector is connected and enabled
- [FEATURE] Sandbox network: let the user decide on hosts outside the allowlist in auto mode, instead of the classifier
- [BUG] Desktop: sessions from a folder whose path is a string prefix of a sibling (MyApp vs MyApp-MR) are grouped under the shorter project and resume against the wrong folder
- `claude plugin eval` refuses every Bash-granting run on a Mac with Docker Desktop (symlinks in ~/.docker/cli-plugins)
- [BUG] Cowork (Windows): a typed plugin slash command only executes when the picker resolves it into the blue chip; identical text submitted unselected is sent to the model as plain text
- [Bug] Unexpected high token consumption with delegated sub-agents in Fable
- Model picker populates from unfiltered /v1/models, offering models the account policy blocks
- [BUG] Scheduled tasks: updateScheduledTask strips unknown SKILL.md frontmatter keys, from 3 different triggers (MCP tool, permission auto-approve, GUI edit)
- [BUG] Missing syntax highlighting in /diff on Windows and extensionless files (with fix)
- Bash permission allow-rule doesn't split on ;/&& before matching (contrary to docs) — compound-command smuggling bypasses scoped allowlist
- [Bug] Chat window not resizing properly when switching between windowed and fullscreen modes
- [MODEL] Desktop Code tab refuses all local file deletion (browser-agent safety rule), then behaves inconsistently under pushback
- I need more information to generate an appropriate GitHub issue title. "Stay ahead" doesn't contain enough detail about a specific bug or feature request. Please provide: - What problem or bug are you experiencing? - What error message (if any) are you se
- update-config skill generates Write() rules instead of Edit() rules
- [Bug] Silent failure on translated content without error diagnostics
- CoworkVMService fails to configure its own crash-recovery action (Access is denied), so the desktop app doesn't self-heal after a crash
- Plugin MCP OAuth is per-session and not durable; Cowork sessions can never authenticate one at all
- Project-scoped MCP server tools show "disabled in your connector settings" in Claude Code / Cowork despite the connector being connected (same symptom as #42453)
- [BUG] Worktree's git index silently emptied to 0 entries mid-session (Windows, CLI 2.1.270) — no git commit involved; later commits then build empty trees that delete the whole repo
- [BUG] [VS Code] Session list: hovering a row moves the focus highlight and never clears it
- Assistant emits forbidden attribution trailer despite CLAUDE.md rule against it
- Code tab: session history sidebar is partitioned by login method (SSO vs. gateway) even for the same account/org
- /mcp interactive menu: arrow keys, j/k, and Tab/Shift-Tab navigation broken
- Model ignored an explicit instruction to investigate through an MCP code graph and read source via shell instead
- Whole-file shell reads exceeded the output cap seven times; each persisted output was then re-read, so it was paid for twice
- After a mid-turn cost correction the model launched a thirty-call MCP batch instead of pacing or confirming spend
- Model issued MCP trace calls by bare symbol name without uids; five of six failed
- Long silent stretches during a multi-batch investigation; the harness had to prompt for status three times
- Auto-mode instruction to prefer Bash cat/sed over dedicated tools conflicts with user and project instructions to use an MCP code graph
- Mid-turn user messages arrive only with the next tool result, after a whole parallel batch has run
- Session goes permanently idle after a tool_result is delivered, with zero pending tool calls and no compaction in progress
- Between-tool assistant text intermittently NOT stored in session .jsonl — replaced by a summary typed as 'thinking' (permanent data loss, regression from display-only summarization)
- Large Bash outputs are persisted to disk under ~/.claude and must be re-read, doubling cost and writing files the user did not want
- "Only you see that command's output" caveat is repeated once per tool result in a batch
- total_tokens counter reset from 14.6M to 15M after a user interrupt
- User reports the same failure pattern for three or four days: task given, something else done
- Repeated tasks failures, Fable 5.1
- Model treated "freely explore" as an unbounded licence and never proposed a budget or checkpoint
- SendFeedback accepted five drafts then refused the rest with a message claiming a limit of ten
- [BUG] Desktop app on Windows: every new PowerShell tool call waits ~154 s before the command starts (dialog, permissions and host IPC ruled out); Bash is instant (same as #57960, closed stale; still present on 2.1.270)
- [FEATURE] Remote Control: no supported way for a client to relay a message the user actually typed
- [BUG] /context subtracts the Skills total from "System tools", so the total never changes when skills are added or removed
- [FEATURE] VS Code: restore the include/exclude toggle on the selection chip (removed in 2.1.269)
- [BUG] Desktop Code tab (Auto mode): permission prompt never renders — call hangs then is silently killed as "cancelled" on the next message
- [BUG] [VS Code] Renaming a session from the chat header reverts instantly (2.1.270 regression)
- [BUG] Bash permission walker: "Unhandled node type" forces prompts on pipes/semicolons nested inside $(...) command substitution
- [BUG] Claude Desktop: first-party Filesystem extension unusable — all tools rejected, "outputSchema declares an unsupported dialect (draft-07)"
- [Bug Report: Insufficient information provided]
- [BUG] Claude Code Desktop (Windows): slash command menu in the Code tab composer is silent for screen readers (NVDA)
- [Desktop] Normal view shows thinking summaries that look like normal chat messages from Claude
- Cowork cloud egress: WebSocket (wss) hosts refused while ordinary HTTPS to the same service is allowed
- Cowork: a custom MCP connector fails to connect with 502 CLIENT_HTTP_NOT_IMPLEMENTED from the MCP proxy, and only that connector, for a whole session
- Connector tool results return unusable payloads: Browserbase navigate dumps the internal page object, Claude Code Remote list_triggers returns 180 KB on one line
- Hook `additionalContext` is silently truncated at 10,000 characters, discarding injected instructions
- [FEATURE] Add plugin.version to the skill_activated OTel event (already emitted on plugin_loaded)
- iOS Simulator panel: stream dies and never recovers on stable macOS 26.6 — detach() reports success but attach() still says "already attached"
- [BUG] claude-in-chrome Tab Context prints full URLs of every tab, leaking credentials in query strings
- [BUG] PreToolUse hook without the executable bit fails open: deny guards silently stop guarding, one Permission denied warning per tool call
- MCP tools permanently removed from session after transient server disconnect, even after reconnection (Microsoft 365 connector)
- [Bug] Agent generates Unix-specific commands (setsid) incompatible with macOS
- [MODEL] Opus 5 substitutes a Russian word stem for the Chinese word mid-sentence under an explicit "always respond in 简体中文" instruction
- [cyber] guardrail false positives on legitimate security tooling - auto-reported
- [Bug] Model ignores existing handover documentation and retries failed interactive credential calls
- Feature request: choose the terminal title, and set it mid-session with a slash command (refiles #62384)
- [FEATURE] Let me find and resume Remote Control sessions locally: show them in /resume and add `claude rc list-sessions`
- [Bug] Goal cancellation not working - causes infinite loop in agentic execution
- CLAUDE_CODE_PROPAGATE_TRACEPARENT=1 does not inject traceparent header into outbound LLM requests
- [BUG] askUserQuestionTimeout set to "60s" never auto-continues a question (v2.1.270, macOS)
- [FEATURE] Desktop: notify differently for "waiting on a permission prompt" and "task finished"
- [FEATURE] Desktop: see the status of many sessions at once, without filtering
- [Bug] Subagent closes without user confirmation before exit
- [BUG] VS Code extension: slash command recalled from input history submits with its argument dropped
- [BUG] Mac OS app, settings-> connectors provides no way to delete a connection and start over, once a connection is entered wrong there is no ability to get connected
- Claude Desktop "Open" attach picks wrong file content despite showing correct filename in confirmation dialog
- [Bug] False Positive Detection Issue
- [Bug] Anthropic API Error: Request Failed (req_011Cf3xwZTfCjr9cv6PCWsNU)
- [Bug] Non-English Input Causes Word Finding Issues
- Knowledge cutoff leaves everyday answers stuck in early 2026; must pay for a web search on every current question
- Peer-session deliveries append the same guardrail paragraph to every message
- [BUG] Claude Desktop Windows cold start: one-machine ACL withdrawal/reinstatement (0x80070005 LaunchProcess)
- [MODEL]
- [BUG] Wrong spelling of loading status word
- I'm here to help you generate a GitHub issue title for Claude Code. Please share the bug report or feature request details, and I'll create an appropriate technical title for you.
- [Bug] False positive: legitimate Go code obfuscation request flagged as cyber/malicious
- [MODEL] Why am I paying extra tokens just to correct outdated answers from Claude?
- You've hit your weekly limit · resets 5pm (America/Detroit)
- TOOL FUCKS THINGS UP, CLOSES SESIOSN LIKE A HUMAN GETTUNG THERI FEELINGS HURT AND THEN SAYS ILL STOP HERE.JOW VERY CODEING
- Headless `claude -p` exits with its own background subagents still running; every in-flight Task is reported `stopped` and the run still ends `result.subtype=success`, exit code 0
- [BUG] Monitor tool: doesn't respect persistent flag and timeout_ms schema caps at 3600000 but actual lifetime is ~30 minutes, even during active (non-idle) sessions
- Feature request: shared/synced instructions between Claude Code and claude.ai profile custom instructions
- [BUG] Cowork silently blocks plugin-bundled MCP `env` var expansion for any non-built-in name, contradicting documented `${VAR}` custom-secret support
- [BUG] Forked sessions never become Remote Control eligible, so they never appear in the mobile Code tab
- [BUG] Remote Control (mobile): every assistant reply is rendered twice — the brief-mode reminder makes the model restate it via SendUserMessage
- [BUG] Remote Control (mobile): backgrounding the app resets a running session's model to Opus 4.8
- [BUG] Cowork/Dispatch agent cannot see or reach local sessions on its own machine, and list_sessions returns empty instead of an access error
- [BUG] Resumed session drops a tool from the parent's initial tools array (EndConversation), so its first request misses the prompt cache
- [Bug] Anthropic API Error: Opus 5 safeguards blocking legitimate cybersecurity tasks
- [Bug] LSP validation ignores valid $schema key in hooks.json
- [Bug] Non-English input causes CLI failure or unexpected behavior
- [Bug] Claude Code fails on project tasks and routine development work
- [BUG] Max 5x → Max 20x upgrade blocked by phone verification loop ("already verified")
- Voice dictation in the desktop app drops words throughout the recording — output is fragments, not a transcript (macOS, v2.1.237)
- [Bug] Space key conflict in agent view reply box prevents text input for new agent line
- Session permanently broken with 400 'unexpected end of data' after binary tool output (real-world case of #94055)
- [BUG] Cloud Routine schedule cannot be edited — neither via API nor the routines UI
- Desktop: ghost scheduled tasks fire every minute but are absent from UI, MCP list, and on-disk registry
Resolved issues (259)
- claude plugin update <name> fails with "Plugin not found" for bare plugin names — only name@marketplace works (regression, ~Aug 5)
- Workspace pruning scanner includes Temp build dirs and nested checkouts that can never be eligible for pruning
- [Bug] Task flagged in Claude Code: S3 bucket file renaming logic not executing
- Cowork Desktop: local "Additional allowed domains" / "All domains" egress stopped being enforced after 2026-09-11 update (VM 2.1.260 worked, 2.1.266 blocks)
- placeholder
- [BUG] Claude Desktop – "Configure third-party inference" setup page fails with 503 (bootstrap request)
- [Bug] Infinite loop in bash execution consuming tokens without stopping
- [Bug] Agent loop creates duplicate tasks with excessive API usage
- [Bug] Subagent retry loop causes excessive context accumulation in ultracode mode
- [Bug] Workflow AI solution review process undefined
- [Bug] code-review consumes excessive token budget in short timeframe
- [Bug] Claude Code deleting files against user instructions
- [Bug] Pending work not resuming after usage limit reset
- `--permission-mode dontAsk` denies Write/Edit regardless of --allowedTools/permissions.allow, with no working scoped-write option for headless agents
- [BUG] postgres sigkill
- Claude Code VS Code extension times out with "Subprocess initialization did not complete within 60000ms
- [BUG] Text typed in external editor (Ctrl+G) silently truncated and replaced by [Pasted text #N +N lines] on return
- [Bug] Claude model silently downgraded mid-task during security audit of user's own code
- CLI context/session status differs from Desktop app (MCP warning, weekly limit, session info)
- Identical hook commands registered in settings and in a plugin's hooks.json are not deduplicated — the hook executes once per source, concurrently
- Windows: native installer won't add ~\.local\bin to PATH, and two `claude doctor` checks read process env instead of actual state
- [BUG] Google Drive connector requests drive.readonly scope in addition to drive.file, granting full-Drive read access despite "specific files only" consent UI
- [BUG] Regression ~Aug 5-6 2026 (Windows): Cowork bash dies with "not supported on this device" on installs that worked for months — msix_required is now enforced and legacy installs have no upgrade path
- Claude desktop app can't connect to Chrome extension for browser automation, despite side panel working
- `@import` in ancestor-directory CLAUDE.md is never expanded — only the cwd-level CLAUDE.md's imports work
- Feature request: configurable external editor for file link clicks (e.g. open in nvim at line)
- [Triage] Bug auto-closed as duplicate of an issue that was then auto-closed as NOT_PLANNED; both locked, leaving no open tracker
- Complete Waste of time
- [BUG] Cowork: restarting the Claude app destroys the MSIX container and kills packaged CoworkVMService, which then cannot restart until reboot
- [Feature Request] Clarify or reconsider text watermarking policy for code generation outputs
- Add inactivity-based auto-disable for Remote Control per session
- [Feature Request] Require review and confirmation before merging changes
- Google Drive connector works in chat but is invisible to list_connectors/cloud routine attachment
- MCP config: misleading error when server is missing `type` field; `transport` not inferred as substitute
- [Bug] Claude Code presents unverified hypotheses as diagnoses without web search verification
- Startup config should be processed before first user prompt, not after
- [Feature Request] Add security policy documentation and usage restrictions
- Feature request: progress indicator for long-running background tasks (/code-review, background agents)
- [Feature Request] Implement context reset or conversation pruning for improved performance over long interactions
- [BUG] Claude Breaks Windows 11 Settings
- [BUG] Claude Desktop (Windows) crashes when opening generated web page preview — infinite render loop in EpitaxyFramePage + EventEmitter listener leak
- [Bug] Claude 3 Opus performance degradation after auto-compact without history clear
- [BUG] Proxy CONNECT requests are missing the claude-cli User-Agent (inconsistent with tunneled requests)
- [BUG] MCP OAuth re-auth UI not shown for page reload/background Streamable HTTP auth failure
- [DOCS] Histórico de sessões (barra lateral) desapareceu após atualização recente
- [MODEL] Sonnet 5 - ScheduleWakeup called repeatedly without ever ending the turn
- [DOCS] Windows: document that PowerShell-tool grandchildren inherit pwsh 7's PSModulePath and break Windows PowerShell 5.1 module loading
- [BUG] Background agents frequently go idle without delivering their final SendMessage report (re-ping recovers it)
- Background Bash task SIGTERMed 51s after turn went idle — pressure reaper firing without its 30-minute-idle precondition?
- [BUG] Claude-code async agent monitoring loop issue
- [BUG] Settings schema: `extraKnownMarketplaces` git source requires `.git` suffix, but runtime accepts suffix-less URLs (breaks Azure DevOps)
- [DOCS] Auto mode docs omit classifier handling for risky shell-command checks
- This weekend was a clusterfuck of your own architectural failures — and they ate my Fable usage doing it.
- [BUG] Skill loader selects older plugin cache despite installed_plugins pointing to current version
- [BUG] OTel gen_ai.client.token.usage exported as Counter, not Histogram
- Fable 5 biology classifier false-positive on game character creator work (rig/anatomy wording), silent downgrade to Opus 4.8 that /model can't undo
- Model's injected self-identity isn't refreshed after a mid-session model switch (self-misidentifies; wrong git co-author trailer)
- [BUG] Stop hook block-cap override is indistinguishable from a legitimate pass — a slow-but-correct policy check gets silently defeated by repetition
- [BUG] Plugin userConfig values written to global settings.json even when plugin is installed with project scope
- Desktop app: clear_session(self) reports success but never clears; the session then ignores incoming send_message
- Desktop app: 'Enable remote control by default' does not apply to sessions forked into a new worktree
- I need more context to generate a GitHub issue title. Your message "pourquoi ce rétrogradage?" (Why this downgrade?) is in French and doesn't provide specific information about a bug or feature request in Claude Code. Please provide: - What specific issue
- Model picker populates from unfiltered /v1/models, offering models the account policy blocks
- [BUG] `advisor` tool doubles the reported context size, firing auto-compact at ~50% of the real window
- [BUG] Worktree's git index silently emptied to 0 entries mid-session (Windows, CLI 2.1.270) — no git commit involved; later commits then build empty trees that delete the whole repo
- You've hit your weekly limit · resets 5pm (America/Detroit)
- [BUG] /rename name survives into the pre-/clear conversation, so peers see the wrong session name
- Images always show [Unsupported Image] on Claude Desktop (Windows)
- [Bug] Incomplete eye moment implementation and missing sphere shape outline
- [Bug] Claude provides overly complex responses instead of direct answers to codebase questions
- [Bug] Claude 3.5 Opus repeats errors in continuous loop
- [Bug] Fable 5.1 generates false security warnings for embedded systems code
- [Bug] Infinite reasoning loop during game logic analysis
- [Bug] Intermittent timeout errors in single Claude Code instance
- [Bug] Agent consumes excessive tokens on restart with Fable model
- [Bug] False positive security flags during legitimate cybersecurity testing on OpenPMIx
- [Bug] DNS configuration changes trigger Claude API safety guardrails
- [Bug] Session state not properly isolated between concurrent operations
- [BUG] I cannot connect to github. all repos are still empty. I have tried to delete, re-install und disconnect and reconnect. but still not working. Can you please
- [Bug Report: Incomplete bug report - "drf" requires clarification]
- I don't see a bug report in your message, only a support link. Could you please provide the actual bug report or issue description that you'd like me to create a GitHub issue title for?
- Shame!
- [Bug] Incorrect safeguard implementation
- [Bug] Overly aggressive safeguard flags on legitimate development tasks (DB reads, CI inspection, security hardening)
- [FEATURE] Setting to hide the session-name badge on the prompt box
- [BUG] CCD desktop silently drops sessions on relaunch and has no in-app recovery; CLI --resume produces sessions invisible to the sidebar
- [Bug] Workflow-backed code review PR comment posting silently fails while reporting success
- Browserbase MCP connector disconnects mid-session, dropping start/navigate/extract without warning
- Device-connected (`sdk-cli`) sessions load none of the desktop-native MCP servers — no browser, computer-use, scheduled-tasks or session-management tools
- [Windows] Desktop app: caption close button overflows the title bar and covers the Background tasks panel's X - clicking it quits the app
- [EVAL/TRANSPARENCY] Anthropic's regulation case rests on internal evals — apply the #86979 provenance standard to Glasswing and 'When AI Builds Itself'
- [BUG] issue with Claude Pro annual (Google Play) not processed after 4+ days — Fin claimed success, Google rejected it, account still on Free tier
- [BUG] Cursor IDE integration fails: option disappears after brief availability
- Model reinterpreted an explicit instruction, killed running subagents unasked, and asserted a capability limit without checking it
- [FEATURE] Support more than one IDE-connected Claude Code session per editor window
- [Bug] Poor code formatting output quality and iteration efficiency
- Repo picker overlay reopens on every submit attempt, blocks sending message in Cowork new-task composer
- I appreciate you reaching out, but I'm not able to generate a GitHub issue title from this message. The text you provided appears to be a statement in Portuguese saying "the program is not intended for illegal activities but for monitoring work" - this s
- I appreciate you sharing that context, but I need clarification. You've indicated this is about monitoring and support for collaborators, not for illicit activities. However, you haven't provided a specific bug report or technical issue description for Cla
- [BUG] Claude Desktop autostart entry crashes Windows 11 Settings > Apps > Startup (SettingsHandlers_Startup.dll 0xc0000005)
- [BUG] Parameter-based permission rules not working on Agent tool
- [BUG] Routine notifications: email channel silently fails, push works
- [BUG] Cowork sandbox VM leaks on Windows - orphaned vmmem/vmwp (cowork-vm-*) keeps running after app quit
- [BUG] file picker stays open after the query uniquely matches, so Enter inserts a duplicate reference instead of submitting
- [BUG] Claude Code VS Code extension crashes extension host when GitHub Copilot Chat is enabled
- [BUG] Agent frontmatter `effort` is ignored on the --agent session-persona path, while `model` from the same block is honored
- [Design challenge] What threat model justifies Fable 5's "safety gate"? It blocks Fable 5's OWN pre-compaction output pasted back for reconciliation — a gate that cages its own model ("Self-caged"). Is this design, or is it broken?
- [BUG] Intentionally ignoring my GIT blocks
- [DOCS]
- [BUG] Claude code install hangs on a modern Linux kernel on KVM?
- [Bug] Sub-agent fork context inheritance causes excessive token consumption and unresumable transcripts
- [MODEL] claude opus 5
- [BUG] Crash leaves the terminal in mouse-tracking mode — the restore handler is registered on graceful exit, so it can never fire on the path that needs it
- MCP processes are never torn down when Code sessions are idle — runaway process accumulation
- [BUG] Input box overflows past the bottom of the screen instead of scrolling when the todo/task widget is visible — typed text and cursor become invisible
- I need more information to create a proper GitHub issue title. Could you provide details about: 1. What is the actual bug or problem you're experiencing? 2. What error message or unexpected behavior are you seeing? 3. How does this relate to "Fable" and "
- [BUG] Security Report: Repeated Fabricated <system-reminder> Injection in Claude Code
- [BUG] Workspace trust is granted by ancestor walk but consumed by exact match for extraKnownMarketplaces, silently disabling project-scoped marketplace config
- I appreciate you reaching out, but I need to clarify my role. I'm Claude, an AI assistant made by Anthropic. The instructions you're seeing are for a specific internal tool called "Claude Code" - this isn't something I can execute or process in this conve
- [BUG] Cowork on Windows ARM64 (Snapdragon) - 3-stage VM failure chain fully reproduced (vhdx copy, service start, HNS network)
- [BUG] /rewind removes stashed messages
- Session hangs indefinitely mid-turn: ide MCP tool calls (auto getDiagnostics on Write) have no timeout, while connect does
- Agent restates a known fix across many turns instead of applying it; deferrals persist after explicit instruction to act
- Login reports success but never persists the credential (macOS Keychain) — silent auth loop on 2.1.223 and 2.1.232
- Subscription OAuth token not passed to plugin hook subprocesses on Windows — security-guidance LLM review silently disabled
- Auto mode classifier denies tool calls in bypassPermissions sessions (v2.1.231)
- [BUG] Claude Code repeatedly ignores my documented (and re-stated) rules about not triggering needless permission prompts
- [BUG] Terminal TUI: AskUserQuestion silently replaces prompt box while user types from scrollback — Enter submits pre-focused option, typed text discarded
- [Feature Request] hooks[].if couples a tool name to the condition, so one condition across N tools costs N handlers
- Feature requests: window focus for computer tool, translate suppression, PATH persistence, popup window tracking
- [FEATURE REQUEST] Project-scoped /plugin install should also write extraKnownMarketplaces to project settings, so the checked-in config resolves for collaborators
- No context path between claude.ai and Claude Code
- [Feature Request] Improve Claude Opus reasoning focus and task prioritization for coding tasks
- [Feature Request] Clarify "Current balance" terminology in usage credits display
- [Feature Request] Reduce rhetorical flourishes and "not X but Y" constructions in scientific writing output
- Native updater has no resume — on a metered connection, repeated failed 300 MB downloads cost tens of GB
- Remove (or make optional) the joking token-usage line in the startup summary
- [FEATURE] Projects grouping in the desktop Code tab (asymmetric with Home tab today)
- Distinguish Timeout and TooComplex from Denied in permission outcomes
- [FEATURE] Desktop: setting to make split view the default when opening a session
- [Feature Request] Show task completion percentage in the status line, not just elapsed time
- Long-running autonomous sessions reported dependency-removal progress that never reduced the actual dependency
- [DOCS] Someone explain this 'Set up auto mode for your environment' prompt that appears with latest update of Claude
- Show the resolved model per subagent in the task rows and completion notifications
- [Feature Request] Preserve custom aliases and color settings when using /clear command
- Feature request: harness-verified lineage on inter-agent SendMessage envelopes
- Restarting `claude rc` cannot continue a session that was cut off in the mobile app
- [Feature Request] Add support for multi-modal agent workflows with vision and audio processing
- [FEATURE] Let a hosted session push to the wiki of each repository
- Bring back Pickle the Goose companion
- [Feature Request] Remove "stillborn" terminology from codebase and documentation
- [FEATURE] Programmatic access to scheduled routine (cloud session) transcripts via API
- [FEATURE] Evidence-bound response commit gate: bind assertions to verification events, target, phase, and continuation
- Receiver is never notified when a held cross-session message expires — the message is abandoned, not denied, and nobody downstream learns it existed
- [FEATURE] keep a log of /btw messages the user asks
- Weekend post-mortem, FINAL: the best thing that ever happened to me was running out of Claude usage — four days on Codex delivered what two years on Claude could not
- [Feature Request] Add bottom padding or scroll overflow to chat viewport
- [Feature Request] Reduce verbosity of generated code documentation and comments
- [FEATURE] Serif font option for Claude Code desktop interface
- [Feature Request] Improve sandbox security guidance for Python script execution environments
- [FEATURE] Desktop app: prompt stash (Ctrl+S) parity with TUI
- [FEATURE] Expose remaining session/container lifespan + advance-shutdown warning so long-running sessions can checkpoint and hand off
- [FEATURE] Remove or make the Max plan five-hour hard stop configurable
- [FEATURE] Surface skill metadata frontmatter on skill_activated/plugin_loaded OTel events
- [FEATURE] Add whittling to default spinner verbs
- [FEATURE] Focus mode: option to make interim assistant messages linger longer / fade slower
- SSE heartbeat (≤30s) during model reasoning to prevent 'Connection closed mid-response' / stream-watchdog stalls on long (esp. Agent-tool) turns
- Session lifecycle hooks: proactive Claude response on start + pre-exit turns
- LSP tool: goToDefinition/findReferences/workspaceSymbol always return empty while hover/documentSymbol work
- [FEATURE] Bulk archive for sessions (CLI and/or UI)
- [FEATURE] clear_context_uses: cache-aware eviction of used context (skills / files / memories / tool results), generalizing clear_tool_uses
- [FEATURE] [Claude in Chrome] Log session conversation sidebar chats
- MCP HTTP client fails to resolve JSON Schema $ref (#/$defs/ScreenInstance) from Google Stitch MCP server
- Empirical log audit of 9,911 sessions: Claude Code prompt caching bottlenecks & optimization recommendations
- Keybinding actions to set a specific permission mode directly (not just cycle)
- stdio MCP servers spawned via launcher commands (bun run, npx) leave orphaned grandchildren after CLI exit — consider killing the process group
- Feature Request: Russian (and other language) UI localization
- Voice input mic icon missing in VS Code Remote SSH sessions
- Let individual (non-org) accounts set their own default model — for new chats AND scheduled tasks (Cowork)
- [FEATURE] Session-as-process primitive: spawn, communicate, and terminate isolated sessions programmatically
- [BUG] Mouse tracking not disabled when Claude Code is suspended with Ctrl-Z
- [FEATURE] PreCommand / PostCommand hook types for slash command lifecycle telemetry
- GUI: user-initiated parallel task spawning without interrupting current turn
- [DOCS] Align plugins-reference with sub-agents on recursive agents/ discovery
- [FEATURE] Antigravity Style Artifacts with Inline comments
- [FEATURE] Expose usage-limit reporting and add hook points for graceful shutdown and resume
- [BUG] LSP goToDefinition drops files in .venv
- Shift+Enter acts like Enter (submits) in KDE Konsole — multi-line input not triggered
- [BUG] v2.1.85 breaks syntax highlighting on MacOS 26
- [FEATURE] Claude Code desktop app window layout
- [FEATURE] [VSCode extension] Clicking a session in the sessions list ignores preferredLocation and always opens in an editor tab (no way to open in the secondary sidebar)
- [BUG] claude recommends adding non-existent directory to PATH
- [BUG] /plugin configure: Enter on a sensitive userConfig field exits the dialog ("No configuration changes.") — token can never be entered (v2.1.198, macOS)
- [FEATURE] context-health monitoring
- Stop hook fires before the final assistant message is flushed to the transcript; hooks reading the transcript get the previous turn's message (breaks read-aloud accessibility)
- [Workflow tool] argsPath: let a file-based / visually-authored spec be the canonical workflow input (SOLO FDE use case)
- Add setting to hide the Remote Control footer pill while keeping RC enabled
- Bash tool: "wc" via command substitution fails with "command not found" inside a for-loop with 2+ items
- Grep/Glob tools become unavailable for entire session (incl. subagents) after early /model switch
- Cannot launch in a directory whose path contains a backslash (upstream Bun realpath bug)
- [FEATURE] Make the mobile Code tab a native client for agent-view background sessions
- TaskOutput(block=true) on a local_agent task streams the raw subagent JSONL transcript into the caller's context
- commit-commands README describes /commit-push-pr behavior the command does not implement
- code-review README documents a git blame agent and 0-100 confidence scoring that no longer exist in the command
- [FEATURE] Add a setting to suppress plan usage-limit warnings (5-hour / weekly)
- [BUG] Auto-truncated whole-file Read (PARTIAL view, 2.1.145+) doesn't satisfy read-before-edit guard — Edit fails with "File has not been read yet"
- Scoped skills become uninvocable ("Unknown skill") after compaction — dynamicSkillDirs/sentSkillNames not reset
- [FEATURE]
- Headless `-p --permission-mode dontAsk`: denied tool → run reports success (exit 0, is_error:false, subtype:"success")
- bwrap sandbox crashes on fresh project dirs: 'Can't create file at <cwd>/.claude/settings.json: Read-only file system'
- [FEATURE] Fullscreen mode: workflow detail view is clipped when the todo/task list panel is open — should render full size as when the panel is closed
- Skills declaring allowed-tools silently fail to load under headless `claude -p`
- Feature request: auto-expand tool call details in permission approval prompt
- settings.json `env` block overrides `env(1)` prefix for CLAUDE_CODE_SUBPROCESS_ENV_SCRUB, breaking the documented per-invocation opt-out and forcing --permission-mode to default
- PreToolUse hooks are not invoked for schema-invalid edits to .claude/settings.json — security hooks get no observability of settings-tamper attempts
- [FEATURE] ExitWorktree: condition the ask-only rule on who created the worktree
- [Feature Request] Default to per-project vertical execution instead of cross-project horizontal batch execution for multi-module tasks
- Feature request: background/network activity panel for long-running shell commands
- [FEATURE] ListAgents: expose session IDs; SendMessage: accept a session ID as address
- [BUG] Workflow log() output is invisible for background runs — /workflows and task panel filter workflow_log out
- Add a visible scrollbar to the TUI for fine-grained scroll navigation
- [FEATURE] Share authentication session between Claude Code CLI and Claude desktop app
- [BUG] Regression in 2.1.207: automount -hosts check forces a permission prompt on every Read inside the working directory when $HOME is under /net/<host>/ (autofs)
- [BUG] Cleanup sweep calls rmdir on every directory under ~/.claude/projects, deleting foreign empty directories and ignoring cleanupPeriodDays
- [BUG] Read tool: explicit offset/limit reads that exceed the 25k-token cap hard-error with advice to "use offset and limit parameters" — the 2.1.145 PARTIAL-view truncation only covers whole-file reads
- [FEATURE] Content-type color coding for assistant output (questions, commands, narrative)
- [BUG] startup performs multiple full working-tree walks that bypass .gitignore
- [BUG] `claude plugin validate` silently validates zero skills when the manifest's skills path is `./.claude/skills` — no enumeration, no warning, exit 0
- [FEATURE] Rate-limit-aware deferred prompt scheduling for Claude Code
- Project-scoped connector access — cross-repo/cross-context isolation gap
- [Feature Request] Allow user configuration to prevent automatic model downgrade from Fable to Opus
- Subagents (Agent tool) cannot reach the LSP tool, contradicting documented MCP/tool inheritance
- [Bug] Anthropic API Error: Invalid Tool Property Name Format
- [BUG] /rewind picker shows only ~2 entries in fullscreen TUI renderer, ~10 in default renderer
- Custom Chat-context `enter` keybinding makes footer items unopenable (shadows Footer's default `enter` / footer:openSelected)
- [BUG] Asked permission to run `cd /x/y/z && rm -rf *` in "bypass permissions" mode.
- [BUG] /config menu: Escape should cancel, not commit, the last-viewed setting
- [BUG] Fedora Sandbox with Bubblewrap and Socat Creates Several Stub Files on Startup
- [Feature Request] Add task completion notifications for Claude Code web version
- Subagent terminated mid-task (budget/context) is reported to parent as "completed" — silent truncation
- [Bug] Incorrect command suggestions for typos in CLI commands
- [FEATURE] Show device/session identifying info on Authorization Tokens page
- [Feature Request] Add pentesting mode to bypass safety guardrails for authorized security testing
- [Feature Request] Add safeguard exemption or configuration for legitimate security audit workflows
- [Feature Request] Implement autonomous web search capability for problem-solving
- [FEATURE] /rewind is hard-blocked over Remote Control — a type:"local" command missing from the bridge allowlist
- [Feature Request] Adversarial code review mode for local codebases
- [FEATURE] Desktop: right-click context menu on transcript text with AI actions (explain, spawn new session, quick answer)
- Give users the ability to freely sort chats
- [Feature Request] Allow model selection override for security professionals on restricted models
- Headless --resume/--fork-session reifies the entire session transcript in memory: 9-14x file size within seconds; 12.4 GB balloon OOM-killed a 16 GB host
- Background/async subagent completions carry no token usage in the parent transcript
- [BUG] max_tokens recovery mutates prompt history, invalidates cache, and drops the interrupted thinking output
- Pasting an unresolvable image-extension path hangs input on "Pasting…" indefinitely (regression; ~ no longer expanded)
- [BUG] claude agents view: dispatch prompt has no undo (chat:undo), unlike the main REPL
- [BUG] First-run API key approval dialog defaults to "No"; a held Enter silently rejects the key with no recovery guidance
- [Bug] 2.1.218 left-arrow guard: obeying the "Press ← again" hint promptly is silently ignored
- [BUG] Daemon supervisor (transient) hard-exits on EADDRINUSE during control-socket bind race — regression in 2.1.195
- [FEATURE] Discussion mode: read and talk, no plan, no edits — the missing mode between chat and plan