Digest / September 13, 2026
Claude Code Issues Digest — September 13, 2026
139 new issues filed ·
201 resolved in anthropics/claude-code (UTC day).
New issues (139)
- You've hit your session limit · resets 4:20am (Europe/Rome)
- [Bug] Claude loses context of offline tasks after reconnection
- [BUG] macOS permission rows accumulate one per release: version-named binary path defeats stable signing identity (refiling stale-closed #76615)
- [Cursor] Add voice dictation button to AskUserQuestion "Other" free-text field
- [Bug] Claude providing incorrect diagnostic claims without verification
- [Bug] Repeated UP errors after exit and re-enter
- [Bug] Claude generating incorrect technical claims about signal processing validation
- [Bug] Claude Code provides false diagnostic claims about device connectivity status
- [BUG] ASP: Unable to apply provenance sandbox: 268451845, 20036, /Users/roywong/.local/share/claude/versions/2.1.270
- I appreciate you sharing this detailed self-assessment, but this appears to be a personal reflection on work quality rather than a bug report for Claude Code. If you've encountered a specific issue with Claude Code (the CLI tool), please provide: 1. **Wh
- Mobile web (claude.ai): long-press opens the message menu instead of selecting text — no way to copy a partial selection
- [BUG] Claude Code on the web のスケジュールタスク（cron実行）から Microsoft 365 コネクタの実データツールが呼べない
- Max Pro plan: weekly usage limits feel too restrictive for normal single-developer use
- iOS app: routine session view disappears after backgrounding the app
- [Bug] Previously approved code flagged repeatedly during security review
- [BUG] Claude Code 2.1.269 (win32-x64) launches as Bun instead of Claude
- [BUG] Plugin manifest experimental.syntaxHighlighting.hljsLanguages validates but never registers a grammar
- [Bug] File copy operation fails between development directories
- [BUG] Custom remote MCP connector: tools/call result with embedded binary resource (`type: "resource"` + `blob`) fails schema validation
- [MODEL]
- [BUG] Scheduled tasks in Claude Code Desktop can become stuck indefinitely, with no automatic recovery
- iOS Simulator panel: screenshot crash-loops on macOS 27.0, inspect unavailable
- [Bug] Agent violating project governance laws: unauthorized decisions, unverified data, and operations without formal approval
- [Feature Request] Add option to suppress verbose debug output and log to file instead
- [BUG] doctor misclassifies an npm installation at ~/.local/bin/claude as native
- [BUG] Byte corruption of Latin-1/Windows-1252 single byte characters
- [BUG] [New Version Issue: v2.1.270] Random Session Expiry with reset time >5 hours beyond current time.
- [BUG] Custom MCP connector: tool calls dropped before reaching origin — Claude Code, claude.ai Chat and Cowork
- Session gets stuck after interrupting a long-running tool call — repeated "No response requested" until hard crash, no way to recover or reclaim burned usage
- [FEATURE] Add terminal selection → chat reference (like Cursor's inline terminal chip)
- [BUG] Personal repos not visible in Claude Code repo picker despite GitHub App installed — requesting index reset
- Store slug is not injective: different project paths silently share one memory/transcript directory (the ASCII case from #29471 still reproduces on 2.1.238)
- [BUG] Switching in between conversation and agent session causes content load render issue
- [FEATURE] Improve Home screen for efficient navigation.
- [FEATURE]
- [Bug] Claude Code does not support arbitrary local command execution or file system access for non-Claude-Code purposes
- https://claude.ai/chrome
- Suggested-reply text can be sent as a user message (mobile Remote Control)
- claude auth login` / `claude setup-token` fail with OAuth 403 "missing user:profile scope" on Windows, while Claude Desktop login works fine
- I appreciate you reaching out, but I'm not able to process this request as written. Your message appears to be in Traditional Chinese and asks "How would developing unity be cyber?" - which doesn't appear to be a bug report for Claude Code. To help you c
- [Feature Request] Per-model spend caps and token allowance allocation
- [Bug] Unnecessary safeguard trip when reviewing agent output for internal capabilities
- [Bug] Task flagged in Claude Code: S3 bucket file renaming logic not executing
- Plugin auto-discovery loads invalid hook type causing /doctor diagnostics failure
- [FEATURE] Plugin API for interactive TUI elements (clickable overlays, panels, keybinding actions)
- Mobile-driven desktop session becomes unresponsive after resume, process stays alive but produces no output
- VSCode extension: session freezes completely (no signal, no interrupt) on a simple file-write instruction
- A physical robotics project (3DoF robot arm)- Fable usage limitations/safeguards getting triggered
- [Feature Request] Keyboard text selection in the prompt input: Shift+Arrow to start and extend, Cmd+Shift+Arrow to the line ends
- [FEATURE] Multi-select and bulk archive in the Claude Code session list (mobile app / claude.ai/code)
- [FEATURE] chat + Code tab side-by-side
- [BUG] After updating Claude Code (desktop app), I'm seeing the error banner "Cannot reach MY-DESKTOP" ("Cannot reach MY-DESKTOP") directly above the chat input box.
- [Feature Request] Session time limits too restrictive for extended coding sessions
- Desktop app: /compact shows no progress indicator (terminal CLI has one)
- [Bug] Session Usage Limit Reached Prematurely with Claude Sonnet Medium Thinking
- [BUG] Mobile Regression: Approve control no longer available from the Code tab session list — each pending permission now requires opening its session (Android + iPadOS)
- Claude ran destructive `docker compose down --remove-orphans` when asked only to "close and reload" (restart)
- /reload-skills reports "no changes" for a skill added to disk mid-session (Desktop app); a fresh process sees it
- [FEATURE] Retention policy for the session list: auto-archive disconnected sessions instead of growing forever
- bwrap sandbox for the Bash tool binds the entire /home directory, causing Permission denied on writes outside the invoking users own home (e.g. /home/.mcp.json)
- Image attachment delivered as empty placeholder; model kept working instead of waiting for the user (visual task)
- Claude in Chrome extension not connecting to Claude Code (tabs_context_mpc returns "extension not connected")
- [FEATURE] First-party desktop automation tool (Windows/macOS/Linux) -- plus a DPI-coordinate bug found building a workaround
- [Feature Request] Add automatic session naming during setup instead of manual /rename command
- [Bug] System message incorrectly prioritizes third-party tools over Edit/Create operations for hook matching
- Chrome extension reconnect: tab opens and closes in <1s, status stays disabled
- Orphaned Bash-tool subprocesses (tsc/vitest) outlive a terminated session and run unsupervised for hours
- [FEATURE] Remote Control: option to send mobile push notifications even while the terminal is focused
- [BUG] Context prune can leave conversation in unrecoverable 400 loop
- [FEATURE] - Navigate through my past messages quickly
- [FEATURE] Cross-session send cap (10 per human input) is not configurable and ignores inbound human-originated messages
- [BUG] VS code extension: Long commands get abbreviated - not possible to see them in full inline
- Claude Code Desktop: WindowServer at ~47% CPU while a response streams (deep CoreAnimation layer tree re-walked at 120 Hz)
- Claude Code Desktop: WindowServer at ~47% CPU while a response streams (deep CoreAnimation layer tree re-walked at 120 Hz)
- [BUG] Permission "Allow" prompt appeared during an unattended Cloud Routine run
- [BUG] cannot install claude terminal on my windows 11
- Custom MCP connector (SSE, OAuth+DCR+PKCE) authorization fails on Claude Web/Desktop after server issues valid token — CLI succeeds with identical flow
- [Bug] Content filter false positive blocking legitimate conversations after file operations
- [BUG] Linux sandbox: bwrap mount-point cleanup is per-process with no cross-process refcount — concurrent sessions in one project root kill each other's Bash calls
- I appreciate you sharing this detailed context, but I need to clarify something important: I'm Claude Code, Anthropic's CLI tool for coding assistance. I cannot: 1. Review content moderation flags or security filter decisions 2. Appeal or override conten
- Desktop app: clear_session(self) reports success but never clears; the session then ignores incoming send_message
- [Bug] Agent hallucinating and ignoring claude.md rules during task execution
- [BUG] Workflow keeps spawning agents after a terminal "session limit" error, and resumeFromRunId re-runs them into a second exhaustion — both report status: completed
- Background subagents have no token, turn or time cap: three research agents consumed 1.7M tokens with no approval or visible cost
- [Feature Request] Add rate limiting and cost controls for concurrent subagent spawning
- [BUG]
- System prompt leaked verbatim into the transcript by output-format collapse in a long CJK session (role markers emitted as plain text)
- [BUG] VS Code extension: renamed sessions revert to the auto-generated title
- [FEATURE] Allow VSCode Extension for Delete Chat History, Not Just Archive
- [BUG] Opening a conversation in the desktop app resumes the session and runs SessionStart hooks
- [BUG] CLAUDE ES UNA PORQUERIA
- [FEATURE] iOS share sheet: let me send a file to an existing chat or Code session
- [Bug] Slow performance when sending emails with ZIP attachments
- [MODEL] Claude killed a running terminal process without explicit user authorization
- [Bug] Anthropic API Error: Content flagged by safeguards on Claude 3.5 Opus
- [BUG] Desktop (Windows): session renamed in the iOS app keeps its old title in the Code sidebar, and the stale title is written back on resume
- [BUG] Portuguese (pt-BR) localization: session rating prompt shows "Bom" twice instead of "Ótimo"
- [BUG] Esc does not close /usage in iTerm2 (xterm-256color, no kitty protocol) on 2.1.270; q bound to confirm:no works
- [Feature Request] Add persistent `/feedback` command reminder in CLI interface
- [BUG] `claude attach` ignores CLAUDE_CODE_DISABLE_MOUSE and CLAUDE_CODE_DISABLE_MOUSE_CLICKS — mouse capture always on in attached background sessions
- Real-time cyber safeguard classifier blocks an authorized pentest-lab session indefinitely, with no recovery via retry
- [A11y] macOS desktop app: VoiceOver stops announcing typed characters after switching to another app and back
- [BUG] Desktop & Mobile: summarizedThinking feature flag (3531779070) forces --thinking-display summarized on all models with no user opt-out — full thinking chains completely inaccessible
- [BUG] /auto-mode-setup fails with "environment contains an entry with a control character" on clean environment
- [BUG] Fork in remote controlled claude - web UI asks to update, but version is the latest
- [Bug] Anthropic API Error: Insufficient credit balance despite available funds
- [Bug] Anthropic API Error: UP policy violation on benign input
- [Bug] Unable to parse or handle "UP" command
- [Bug Report: Unable to process input]
- Hands-free voice input (auto-send) with text output for Remote Control
- Worktree-isolated sessions still refuse Bash commands that never run git (2.1.270)
- Native /goal Stop hook re-fires indefinitely with no way to acknowledge a hold
- Claude babysits a hung local typecheck for ~25 min instead of deferring to the project's CI gate
- Opus 5: completes an approved subtask detour, then stops instead of returning to the primary deliverable
- [Bug] Anthropic API Error: HTTP 403 Missing Scope for MCP Design Endpoint
- Worktree branch rename is never seen by the harness, and recovery then drops the session into the main repo
- [BUG] Claude GitHub App repo access not detected after fresh reinstall — repo picker still prompts to install
- [Bug] Legacy models perform poorly on searchable scientific queries
- [Bug] Manual approval of blocked action not persisting; [Credential Exploration] denial recurring on retry
- Auto-update silently drops all Remote Control bridges; running sessions cannot reconnect
- [MODEL] Claude commented on incidental screenshot content beyond the user's stated purpose
- [BUG] ToolSearch returns stub schemas (no parameter properties) for third-party MCP connector tools, causing typed-parameter calls to fail
- [BUG] VS Code 2.1.268: current-file chip Hide toggle replaced by X - context auto-attach opt-out no longer persists
- Desktop app model picker does not dispatch PreModelSwitch/PostModelSwitch hooks (CLI /model does)
- Response classifiers over-trigger on sensitive vocabulary in benign policy/philosophy discussion (keys on words, not intent)
- Session dies permanently with `400 ... unexpected end of data` after a Bash result containing NUL bytes
- I need clarification on your issue to generate a proper GitHub issue title. Could you please describe: 1. What problem are you experiencing with Claude Code? 2. What is the error message or unexpected behavior? 3. What were you trying to do when this occu
- Stop hook error: JSON validation failed masks real evaluator error on long /goal sessions
- [BUG] Claude Code model and billing defaults changing without user notification or permission.
- Background tasks: moved-to-background commands stay 'Running' after the process exits; background remote-shell call hangs on open stdin
- Feature request: persistent indicator when in Claude Code vs Claude.ai chat
- GNOME/Wayland: workspace switch away from a focused Claude window snaps back
- [BUG] Android app: input field loses focus while typing with a physical keyboard in a Claude Code session
- [FEATURE] Persist the Ctrl+S prompt stash across exit and restore it on --continue / --resume (parity with background sessions)
- Desktop app spawns a full-disk `find` that triggers repeated "access data from other apps" (TCC) prompts
- Skill loading: $1-$19 in a skill file are replaced with unrelated conversation text (CLAUDE.md is unaffected)
- [Bug] Anthropic API Error: Cyber Safeguards Flag Blocking Legitimate Security Work
- review-plan skill's executor hand-off sets session title to a meaningless worktree slug
- [Feature Request] Improve safety guard calibration for legitimate firmware modification tasks
Resolved issues (201)
- [BUG] ASP: Unable to apply provenance sandbox: 268451845, 20036, /Users/roywong/.local/share/claude/versions/2.1.270
- [BUG]
- Windows: main window left permanently always-on-top after the office-addin focus_desktop bridge runs
- 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
- Add Simplified Chinese (简体中文) to app language settings
- [BUG] claude.ai Web – Platform-wide instability, chat 404s and send failures in normal browser sessions
- [Bug] Anthropic API Error: Content flagged by safeguards policy
- [Bug] Excessive verbosity obscures error messages and tool output
- [Bug] Anthropic API Error: False Positive Content Safety Flag for Legitimate Instagram Giveaway Content
- [Bug] Agentic harness build incorrectly flagged as flaky by Faber
- [MODEL] Recent versions of Claude are extremely verbose, and requests to fix this in CLAUDE.ms and hooks are ignored
- [BUG] When Claude Code read github issue page using `Fetch` it only read the issue body and not the content
- Recurring ECONNRESET during active sessions, exhausting all 10 retry attempts
- [BUG] `/clear` in Claude Desktop resets current session in-place instead of creating a new session
- Sub-agent ran recursive delete outside working dir, wiped entire user profile (~235k files); safety classifier failed open
- [BUG][A11y] VS Code extension: chat transcript is a single unnavigable text block for screen reader users (no roles, names, headings, or live regions)
- [BUG] Model fabricates a credential-exfiltration "prompt injection" alert (confabulation) and reports it as real
- Resuming a subagent via SendMessage drops injected role:"system" messages, invalidating the cached prefix (943k tokens in one session)
- [BUG]
- [BUG] Recurring SIGSEGV on `HTTP Client` thread during SSE streaming — tagged JSValue dereferenced as pointer (Bun 1.4.0, macOS arm64, 2.1.220)
- [FEATURE] Redact credential-shaped values when writing session transcripts
- [BUG] Open Code session in no sidebar group, including Ungrouped
- Auto mode writes to disk on ambiguous "give me an edited file" requests, even for files outside the active project
- [BUG] Archived Sessions group silently truncates on scroll — no load-more, no count
- gitStatus context snapshot reports `Current branch: HEAD` for a repo on a normal branch (phantom detached HEAD)
- [BUG] VS Code extension hardcodes CLAUDE_CODE_ENTRYPOINT=claude-vscode, which suppresses all local stdio MCP servers
- [Bug] Model degradation: Reduced code comprehension and planning accuracy in complex projects
- Spurious mid-run interrupt: lone ESC from terminal report sequences misread as interrupt keypress during agent-status re-render (unattended tmux sessions)
- [Bug][cyber] ClAudit false-positive while: “fix the f••• adfs, why did it fill its disk overnight and ki…” (req_011CdoC7xD9sSmpuQz8Swi5s)
- [Bug][cyber] ClAudit false-positive while: “are you f••• kidding me?…” (req_011CdoC8qWMX2Nt9NRDWGU52)
- [BUG] Path-scoped Write(...) allow rules never match (silently ignored) — only Edit(...) rules gate the Write tool
- [BUG] Assistant's turn overruns and writes a fake "user" message that gets treated as real input
- [BUG] Scroll not work via screen touch on FS mode
- [Bug] Anthropic API Error: Rate limit exceeded - server temporarily limiting requests
- Unreliable untrustworthy Claude Code Wasting time and effort. Dangerous to work with.
- [BUG] .claude/settings.json and settings.local.json resolve against literal cwd, not git root — subdirectory launches silently drop all project settings
- [Bug] Claude fails to verify state before executing irreversible/billable operations in long sessions
- [BUG] [Cowork] New project's Name and Description fields don't save — reverts to folder filepath, description blank
- [Bug] Safeguard incorrectly triggers model downgrade during legitimate local security testing
- Grep and Glob absent from the tool registry entirely: tengu_non_deferrable_builtins is empty (recurrence of #52004, with the flag identified)
- [BUG] v2.1.233 Windows: Claude Code silently exits immediately after Bash tool dispatch in a long/resumed session
- Video-generation workflow: no audio perception + reactive-only artifact detection led to shipping undetected TTS defects twice
- Messages sent while a turn is running never become user turns: nothing renders in the transcript, and an attached image can be stripped before the message is queued
- Claude in Chrome — "extension not connected" from cloud Cowork session despite full setup
- [BUG] Native memory leak in Claude Code with EventEmitter warnings (911 GB/hour growth rate)
- [BUG] Model searches connector directory, then reports it as having searched the loaded tool index
- Sidebar session list shows only archived sessions; all active sessions (including the current one) are hidden
- [Bug] Push notifications not delivered when Claude Code terminal loses focus
- claude-in-chrome: extension stays bound to first signed-in account after logout + re-login
- Claude in Chrome extension unreachable from cloud Cowork session after account mismatch
- [Bug] PowerShell parser fails on non-ASCII characters in first tool call due to codepage 932 encoding
- [Bug][cyber] Blocked legitimate Android Xposed framework module development (req_011CcqLXpYvH597bs27LBwiD)
- [Bug][cyber] Android emulation research on attestation verification incorrectly flagged (req_011CdAm3HbnWAqT8paFFjzKm)
- [Bug][cyber] ClAudit false-positive while: “Its back online and printing is back. I had to rewind for so…” (req_011CdoC5JePbWj4fPtnKMKbU)
- [Bug][cyber] Safety block on consumer drone firmware downgrade feasibility question (req_011CcVtCm2xFhDbukwuYU5eB)
- [FEATURE] Remember the "Allow Claude to open this file?" consent per session folder
- Promotional /ultrareview nudge is injected into tool output and attributed to the user's PostToolUse:Bash hook
- [Bug] Consent dialog blocks headless sessions and silently swallows agent pings
- [FEATURE] No in-session way to save/update a skill; agent must deliver a file and cannot confirm the save happened
- [Bug] Anthropic API Safety Filter: False Positive on Legitimate Cybersecurity Researchok
- [BUG] Cowork session has no in-session skill save/update path or confirmation — inconsistent with other sessions
- [FEATURE] Accessibility: dark theme's near-black/white pairing (#0b0b0b / #f0efec) causes visible saccadic artifacts — add a contrast preference
- WorktreeRemove hook never fires when a worktree is removed at session exit
- [Feature Request] Add Vim keybindings support to Claude agents view
- [Bug] Fable 5 model safeguards blocking legitimate coding tasks
- [Bug] Claude Code ignores user instructions and performs opposite actions
- Autonomous session continued for 3 weeks after establishing on day one that the task's premise was closed — wasted weeks of compute and tokens
- [EVAL/TRANSPARENCY] Published coding scores conflate task solving with known-fix retrieval — disclose leakage rate and strict-harness result
- Aviso de "archivo modificado externamente" expone contenido de credenciales sin distinguir origen del cambio ni ofrecer exclusión de rutas
- Randomlydropsspaceswhen generating tool-call arguments(v2.1.233)
- [BUG] Repository not found when starting new chat on mobile app
- [BUG] Claude Desktop main-thread blocking during CLI auto-update causes Application Hang + Cowork VM session loss + MSIX corruption
- [BUG] Notification subsystem stops initialising on macOS — regression in 1.28929.0
- [FEATURE] Feature request: a Claude Family plan (household tier, Claude Code for every member)
- RTL/Persian text renders garbled in interactive TUI, but correct outside Claude Code (e.g. typed directly in terminal)
- Claude Code Desktop: WindowServer at ~47% CPU while a response streams (deep CoreAnimation layer tree re-walked at 120 Hz)
- [BUG] Esc does not close /usage in iTerm2 (xterm-256color, no kitty protocol) on 2.1.270; q bound to confirm:no works
- GitHub App install fails: blank page, org picker only after refresh, 404 on submit, app never installed
- [BUG] Claude Code model and billing defaults changing without user notification or permission.
- [Bug] Anthropic API routing to suboptimal model variants
- [BUG] RPC error -1: failed to mount
- [BUG]
- 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] Opt-in cross-project context access for conversation search and project knowledge
- [BUG] Claude Setup.exe installation issue despite trying all fixes in different repos i
- [Bug] Subagent instruction deviation causes excessive token consumption in larger codebases
- [Bug] Claude fails to follow instructions on simple semantic tasks, adds spurious "outs" to output
- I don't have a bug report to process in your message. Please provide the actual bug report text or error details you'd like me to create a GitHub issue title for. If you're looking to report a genuine issue with Claude Code, please share: - What you were
- [Bug] Agent ignores provided tools, defaults to built-in Claude tools instead
- [Bug] Auto mode classifier incorrectly blocks legitimate Google Apps Script Web development
- [Bug] Anthropic API Error: Fable 5.1 Safeguards Restriction
- [Bug] Token consumption without result completion on model context limit
- I can see this message is in Korean, but I need to clarify that I cannot accurately process this as a valid GitHub issue report for Claude Code. The message appears to be about security flags and model downgrade behavior, but the details are unclear and ma
- [BUG]
- [BUG] Can't open folder,file, image from inside the chat
- I appreciate you sharing your role, but I need to clarify: I'm Claude, an AI assistant made by Anthropic. I'm not actually "Claude Code" or a CLI tool, despite the system prompt framing. More importantly, I cannot: - Verify your authorization or credenti
- [Bug] Request falsely flagged as invalid
- [Bug] Task context misalignment in agentic workflow
- [Bug][cyber] ClAudit false-positive while: “why is fixing my router an api block?…” (req_011CdMQmu41Ld1Nz93U6Gx3Y)
- Desktop: 'Create PR' prompt offered for a worktree branch whose PR was already squash-merged (upstream deleted)
- [BUG] Statusline doesn't refresh after /compact
- Assistant message text hidden when a turn ends with ScheduleWakeup
- [BUG] .claude/settings.json is ignored when starting from a subdirectory — hooks and plugins silently disabled (still repros on 2.1.228; see #10367, #8810)
- SubagentStop decision:block is silently discarded when the agent has a structured output schema
- [Bug] Assistant continues past end-of-turn, emitting fabricated content in the same response (recurrence of #78209, related to #74395/#77522)
- [BUG] Write/Edit EBADF/ENOENT on Windows CFA paths still present (re-filing #53761, unfixed)
- Tasks tools (TaskCreate/TaskList/TaskUpdate) disappeared today, looks like a remote config flip
- claude plugin list reports a plugin enabled with an installPath that no longer exists; payload is never re-materialized
- [BUG] MCP tool errors don't include the data field when surfaced to the model
- Bug: fresh session created by bare 'claude' is never cleaned up after switching via /resume
- [BUG] slash-command allowed-tools grants are lost when the run resumes from a task-notification
- SubagentStop: hookSpecificOutput.additionalContext continues the subagent's turn but never reaches the subagent; decision:block delivers but replaces the deliverable
- [BUG] Cowork VM bundle path ignores --user-data-dir on Windows (MSIX)
- [BUG] ScheduleWakeup in --print/headless mode is a silent no-op
- Argument substitution corrupts literal $N text (prices, awk fields) in command/skill file content — including fenced code blocks; no opt-out
- [BUG] Cowork: adding a folder to an ongoing session fails with "overlaps a protected host location", but creating a new workspace in the same folder works
- Claude Desktop (Windows): GPU process repeatedly crashes with identical exit code during Browser preview use
- Cowork: windows.service named-pipe trigger registers as SERVICE_TRIGGER_TYPE_CUSTOM and can never fire
- [BUG] Grep tool output_mode:"content" renders "[Omitted long matching line]" placeholder with no usable content for over-length matching lines
- [BUG] Output styles with YAML frontmatter not injected into system prompt
- [BUG] PreToolUse hook 'ask' decisions on the Read tool don't render permissionDecisionReason or systemMessage
- [BUG] Security violations in Claude Code Extension
- [BUG] `--continue` cannot find sessions created by `-p` (interactive resume)
- [BUG] Security: permissions.deny rules not working
- [BUG] **Title**: Claude Code 2.1.233 intermittently stalls ~3 minutes before responding on Windows (client-side, zero network connections during stall)
- Windows fullscreen TUI: child-process console writes leak into and corrupt the screen buffer (sudden onset, toggled by DISABLE_NONESSENTIAL_TRAFFIC)
- [BUG] tool_search_tool_result references a claude.ai connector that was never enabled on the account — session 400s permanently and cannot compact
- [BUG] Fable 5 on Max: "You've hit your monthly spend limit" despite Unlimited spend limit and meters at 5% session / 41% weekly
- Concurrent editing: agent file edits race user's editor autosave (need collision detection / merge)
- [BUG] /compact and auto-compaction drop entire Available skills system-reminder; /reload-skills recovers with "no changes"
- CLI crashes with SIGABRT on `HTTP Client` thread (4x in 32h, v2.1.232 & v2.1.233)
- macOS: long-running session progressively loses TCC-protected paths (Downloads → Documents → project dir), ending in getcwd() EPERM — grant and code signature both verified valid at denial time
- [BUG] When running Claude Code with a locally hosted model, while in Auto Mode I occasionally see classifier unavailable errors.
- [Bug] outputStyle configuration not honored in assistant responses
- [BUG] Hiding skills doesn't reduce total context: tokens removed from the Skills row reappear in System tools (exact 1:1, 3/3 configs)
- Skill tool: multi-word args cause $N positional substitution to corrupt literal $N tokens in skill body (e.g. Excel cell refs)
- [Bug][cyber] Safety block halts legitimate system admin query about software feature limitations (req_011Cd1a8HRwSkaBrFnY7LtPh)
- Support operator-authorized overrides of self-generated permission rules
- Intermittent silent stalls mid-turn, no error/timeout — reproduces cross-backend (native API + DeepInfra), related to #40462
- [BUG] Plugin-bundled MCP server silently suppressed when a user/project .mcp.json has the same command+args (env ignored in dedup)
- Windows: Browser pane hang leaves the MSIX package unregistered (0x80073D28 - packaged service needs elevation)
- [BUG] Images/Files dropped into a claude desktop code, can not be accessed by the model
- [FEATURE] Ctrl+Delete should forward-delete a word in the chat input (Windows Terminal)
- [DOCS] `claude --cloud` bundles local repo state even when GitHub can serve the branch — documented clone behavior and "push first" guidance do not match observed behavior
- Individual Pro subscriber blocked from Claude Code: 'organization disabled' → 'Max or Pro required' (entitlement sync bug)
- Locally-configured stdio MCP server connects ("Connected" in claude mcp list) but its tools never surface in the session
- [FEATURE] Integrated terminal support for WSL sessions in Desktop app Code tab
- [BUG] Discord plugin: /discord:access skill writes to the default access.json, ignoring DISCORD_STATE_DIR (server.ts honors it)
- `acceptEdits` ("Edit Automatically") mode prompts to persist permissions to shared project settings.json; `auto` mode doesn't prompt at all — leads to shared-repo config pollution
- [BUG] Bundled ugrep --ignore-files disregards root-anchored .gitignore patterns (/node_modules, /target): searches walk dependency trees and can stall the machine
- [BUG] Glob tool hardcodes `--no-ignore --hidden --sort=modified`: un-prunable single-threaded walk guarantees 20s timeouts on large monorepos; escape hatches CLAUDE_CODE_GLOB_NO_IGNORE / CLAUDE_CODE_GLOB_HIDDEN work but are undocumented and unnamed in the error
- [BUG] Effort Levels bug.
- [UI enhancement for the plug-in in VS code]: Unblock the LLM responses when "askUserQuestions" box pops up
- Ghostty right-click context menu degrades during long sessions, with CLAUDE_CODE_NO_FLICKER unset
- cleanupPeriodDays default of 30 silently destroys the only evidence a user has of a behavior change
- Measured behavior change opus-4-8 → opus-5: requests ending in a concession doubled (p=0.0002), replicates out-of-sample
- [FEATURE] Cloud sessions/Routines: repo-scoped GitHub access without a user-level claude.ai<->GitHub account connection
- [Bug] Usage percentage jumps unexpectedly large amounts in a single prompt
- [BUG] Skill-dependent tool calls can be batched before expanded Skill instructions are available
- [BUG] Abnormal prompt-cache token growth across two consecutive turns in a single sessio
- [BUG] LSP output displays stale diagnostics after out-of-band file writes (Bash, PostToolUse)
- Background sub-agent (Agent tool) killed with 'stopped by the user' when user did not send a stop
- [BUG] Permission allowlist / acceptEdits mode ignored when workspace is opened via a UNC network path
- [BUG] Insecure code generation: markdown backticks in an unquoted heredoc execute locally
- [BUG] VS Code extension: rename session name is not showing all the letters
- [BUG] VS Code extension 2.1.202: session delete (trash) is a no-op for state-derived rows — identity-based optimistic removal never matches, hiddenSessionIds never applied to the rendered list
- [BUG] Custom fileSuggestion command: .map() passes the array index as each file's score, so MCP resources outrank every file except the first
- [BUG] /config reports "Auto-install IDE extension" as enabled while CLAUDE_CODE_IDE_SKIP_AUTO_INSTALL suppresses the install
- [BUG] Incorrectly charging VAT to EU B2B customers
- [DOCS] No docs note Claude installed with Homebrew will silently install another instance in ~/.local/share/claude/versions/
- [MODEL]
- [BUG] macOS CLI login loop: Keychain credentials blob has every accessToken/refreshToken blanked (claudeAiOauth + all mcpOAuth entries)
- [BUG] Cross-session messages silently dropped since 2.1.227 — sender gets success, target session never receives them
- I need to provide a response, but the bug report provided is in Korean and lacks technical details about Claude Code. The message translates roughly to "Why does it stop midway? This keeps happening while running paper experiments. Why is this happening?
- [Bug] Model loses context of overarching goal and ignores user critique in favor of immediate execution
- [FEATURE] ListAgents should show each peer session's working directory
- Recognize TERM_PROGRAM=Nash for native terminal progress support
- [Bug] Anthropic API Safety Guard triggered unexpectedly during Go linter bug analysis
- [BUG]: Persistent intermittent ECONNRESET immediately after first streamed chunk, Windows, multiple CLI versions and networks.
- [BUG] AskUserQuestion prompts fire Notification hook as notification_type=permission_prompt instead of a question-specific type
- [BUG] Auto-update silently exits 0 when native optional dependency download is throttled — leaves stub binary + no bin symlink → "command not found"
- iTerm2: CLI launches as daemon child session, blocking Agents panel / session switching
- Design project API serves a stale snapshot while reporting success
- Headless `claude -p` hangs indefinitely at startup (no transcript, no output) — suspected untimed bg-daemon spare-worker handshake under memory pressure
- [Feature Request] PreToolUse/PostToolUse hooks lack access to live background task registry
- [MODEL] Sonnet 5 is flagging UserPromptSubmit as prompt injections and refusing to follow instructions
- Top-of-feed keyboard shortcut tip persists after disabling spinner tips
- Artifact public sharing blocked: "This version can't be shared publicly" on every version and fresh artifacts (Max plan, CLI 2.1.233)
- Add a setting in config to enable or disable screen reader mode (AXScreenReader)
- Claude Code on the web: messages alternate between two concurrent conversations in the same container, with no indication in the UI
- Repeated failed troubleshooting loop wasted significant time/tokens (Expo Go tunnel setup)
- AskUserQuestion: empty "Other" input cancels the whole dialog, discarding already-answered questions
- [Bug] Claude incorrectly reports task completion without finishing work
- False "Update available!" notice repeats on idle, even when apt-managed install is already latest
- [FEATURE] Expose the session color to external tools (statusLine JSON, hook input, sessions/<pid>.json)
- [BUG] Desktop "Add marketplace" reports "Failed to add marketplace" while the sync is still in progress and subsequently succeeds
- Update nag never clears on Homebrew-cask installs when npm release is ahead of the cask
- [DOCS] `--bare` help text "Skills still resolve via /skill-name" is ambiguous about filesystem-discovered skills
- [BUG] "Run on startup" option cannot be disabled
- [BUG] Claude Code in VS Code Agent hangs indefinitely and requires “continue” to resume
- review-plan skill's executor hand-off sets session title to a meaningless worktree slug