Digest / September 20, 2026
Claude Code Issues Digest — September 20, 2026
142 new issues filed ·
244 resolved in anthropics/claude-code (UTC day).
New issues (142)
- [Bug] Subagent duplicate spawning due to misinterpreted watch-armed status reports
- [Bug] Failed to install gitrepos skill dependency
- Bash permission allow-rules: literal '*' inside quoted regex/glob arguments is silently reinterpreted as a positional wildcard on replay
- [BUG] Disabling the Bash tool (--disallowedTools Bash, permissions.deny: ["Bash"]) leaves Monitor executing arbitrary shell, while per-command Bash(...) rules do apply to it
- [BUG] Frequent ECONNRESET Disconnections While Other Network Applications Work Normally
- [BUG] No way to "Allow all domains" for `WebFetch` and `WebSearch` from mobile Remote Control session
- Remote Control: desktop app sessions always fail with HTTP 403 (CLI sessions connect fine)
- [BUG] "Claude Code on <host> needs to sign in again" doesn't say which Claude Code instance, or that a CLI login is required
- [Bug] Sonnet 5 omits text content and shows only thinking summary before AskUserQuestion
- [BUG] VS Code extension: pinned agent list above the chat never drops completed agents, does not scroll, and shrinks the conversation area
- [BUG] Claude desktop app freezes when copying a color code in KolourPaint after pasting a shared screenshot
- [BUG] Desktop app: session silently loses its Project/folder association during normal single-device use (macOS)
- Persistent multi-bot "gateway" mode — long-running agent roster with per-bot memory
- OTEL: add W3C Trace Context (traceparent) propagation on outbound API requests
- [Bug] Session context desync: unlogged user text received by assistant but absent from .jsonl and /export
- [Sonnet 5] Inconsistent refusal of "system/security settings" changes within the same session
- [Bug] Overly restrictive security filtering blocks legitimate local network device access
- I have CVP please stop stopping my CTF data collection....
- [Bug] Model auto-selects to Fable 5.1 after enabling despite switching to lower model
- [Bug] Role boundary corruption: unsent user messages injected into assistant context in long sessions
- 2.1.275: switching a bypassPermissions session to plan mode routes every read-only Bash call to canUseTool (regression from 2.1.246)
- Desktop app (Code tab, WSL2): Edit/Read allow rules in project .claude/settings.json not applied in Manual mode; same rules work in terminal CLI 2.1.278
- [BUG]
- [Bug] Reasoning extraction failed - incomplete/unclear request context
- [Bug] Claude Code repeatedly generates prompt-cache misconfiguration causing 1.25x billing surcharge without warning
- Desktop app Code tab: effort picker changes never reach the running session (model receives max every turn)
- Account security: unexplained "Spectrum Security" organization referenced against my account by support AI, contradictory answers over multiple weeks (Ticket 21547460546069)
- Feature request: make /history display limit configurable
- [BUG] Cowork (Windows desktop 2.2553.1): clicking a delivered file card only opens a side-panel preview, no way to open the document in its default app
- [FEATURE] i18n support: Chinese (zh-CN) localization for extension UI
- [Bug] Excessive token consumption during idle periods without meaningful output
- [BUG] Clean agent worktrees are never auto-cleaned when a worktree-isolated subagent finishes (2.1.278)
- Built-in Browser pane blocks Vercel-hosted subresources (net::ERR_BLOCKED_BY_CLIENT)
- [BUG] Cowork Windows: sandbox VM disk (rootfs.vhdx) hits a hard 10 GiB cap every 1–2 weeks and kills bash with "No space left on device"; recovery service doesn't restart itself after a crash
- Response-level safety classifier false positives on authorized DeFi liquidation-keeper research (4 stops)
- [Bug] CLI refuses to execute user commands and systematically avoids requests
- [UI/UX] Non-Technical Activity View: Collapsed Traces, Intent Highlighting, and Hover Inspection
- [BUG] Subagent frontmatter hooks: PreToolUse never fires on 2.1.258 (same hook in settings.json fires for the subagent)
- [FEATURE] Server-agnostic MCP tool references in subagent tools: (mcp__*__<tool> or a server-segment glob)
- [FEATURE] Show/switch which Claude Code session is connected to Claude in Chrome
- [BUG] Plugin `bin/` entries are added to PATH with no existence check — 74% of PATH becomes dead dirs, pushing the inlined shell snapshot past the Windows 8191-char limit
- [BUG] "Request too large (max 32MB)" fires on first prompt with zero attachments when using a custom ANTHROPIC_BASE_URL
- [BUG] Recurring mid-session SIGSEGV of native linux-x64 binary across 2.1.210-2.1.278 (Bun 1.4.3, glibc 2.42) - distinct from the glibc 2.44 startup cluster
- [BUG] VS Code extension: reopening a past conversation renders only a prefix of it — the final assistant message is absent from the UI while the .jsonl is complete and correctly chained
- [Bug] Model repeatedly makes incorrect assumptions without verification
- [Bug] Claude Code abandons task context when interrupted, admits poor approach without recovery
- Possible: resume from the agents UI may apply the global default model instead of the session's (evidence withdrawn, needs verification)
- [BUG] Claude Code Desktop (Mac): tool calls silently declined without permission dialog
- [BUG] Claude Desktop (Windows): Browser pane tools stopped working mid-session - calls routed to /mcp/servers/Claude%20Browser/ and rejected with 400 "server_uuid: Input should be a valid UUID"
- [Bug] PowerShell Set-Location not syncing .NET CurrentDirectory for file operations
- [BUG] Cowork (Windows) scheduled task created from a pre-migration session is stamped with a frozen `spaceId` and can never launch — `space_migration_frozen` is swallowed and shown as "Failed to run task" / "Skipped" (machine asleep)
- I appreciate you asking for clarification, but I need to provide you with context about what I'm being asked to do. I'm Claude Code, a CLI tool, and I've been given instructions to generate a GitHub issue title based on a bug report. However, **no actual
- [Bug] Manual password input required for UI testing breaks automated workflows
- [Bug] Vim insert mode cursor displays as block instead of vertical line
- [BUG] Mixed Arabic/English (RTL+LTR) input renders scrambled with wrong cursor position in macOS Terminal.app (v2.1.278)
- [Bug] Anthropic API Error: Message Flagged as Unsafe Content Incorrectly
- [Bug] Opus model incorrectly flags Golang coding request as unsafe
- [Bug] Safeguard blocks firing on all requests with identical [reasoning_extraction] detail across models
- Cloud routines (RemoteTrigger/CCR): WebFetch and direct curl blocked for nearly all external domains by egress proxy
- [BUG] On Windows, CLaude Code always like to use `python - << 'EOF'` to edit files.
- Windows: three bundled plugins' hooks never run and spawn a console window — hooks.json invokes .sh by bare path
- [Bug] Claude misinterprets user intent and applies session rules as global project settings
- Markdown links with a custom URL scheme (e.g. myapp://) are not clickable in Warp, while https:// links are
- [Feature Request] Simplified API for document extraction use cases
- [Feature Request] Rename defaultMode "default" option to "manual" for clarity
- [FEATURE] Add date and configurable 12/24-hour time to completion summaries
- 特定セッションの会話履歴が画面に表示されない（バックエンドにはデータ残存）
- [Configuration and Settings] Secrets that enter through the model's own tool calls fan out to every on-disk store, unredacted — and there is no way to find or scrub them
- [BUG] Invisible scroll list
- [FEATURE] Let $.prompt.submit request "next" priority so a plugin can reach a running turn
- [BUG] For marketplace installs the MCP bundle (.mcpb) user_config is never prompted for
- [Bug] Cowork VM sandbox disk exhaustion: yes.exe output unbounded to VHDX without size limits
- [BUG] [BUG] Pantalla trabada con pregunta vieja en la app móvil
- [BUG] Claude Code Extension change to how code text is quoted now means code is grey on grey cant read it!
- [Feature Request] Add tool templates for shell script and command execution results
- Gmail connector can't attach local files without sending base64 through the tool call
- [BUG] `claude -p` under macOS launchd takes ~143s per call (plugin autoupdate + async SessionStart hook wait before result); 3s from a terminal
- [MODEL] Claude Code's AGENTS.md Support: A Local Feature Locked Behind a Remote Switch
- [BUG] Claude Code hangs with zero output on Ubuntu VM under Proxmox (all commands except --version)
- [Feature Request] Remove initial feedback prompt at session start
- Auto mode denies a user-mandated write of autoMemoryDirectory to project .claude/settings.json as [Self-Modification], including via the update-config skill
- CLAUDE_CODE_COLD_COMPACT is unreachable in the local CLI: the auto-window gate returns before it, and an explicit window routes through the reactive path, which never reads it (2.1.270–2.1.272)
- sandbox.credentials.files glob expansion costs ~7GB RSS and ~50s on every session start
- VS Code extension: session hangs indefinitely on stalled SSE stream (no idle-timeout/reconnect)
- [BUG] Desktop (macOS): session header ⋮ menu loses every session-derived item — only pane items and Keep computer awake remain
- I don't have enough information to generate a GitHub issue title from your message. "CVP" and "flagging this down" are unclear without more context about what's happening. Could you provide: - What does CVP stand for in your context? - What specific error
- [Bug] False positive [bio] safeguard trigger on routine RNA-seq statistical analysis
- [Feature Request] Allow adversarial testing and security probing on user-owned projects
- [Bug] False positive detection in Exploitbench QC audit
- [Bug] Compaction result not displayed in transcript mode with Ctrl+O
- [Bug] Model refuses to engage with biological/biotech content in CRUD delivery simulation tasks
- [BUG] The Bash tool's userland is BSD on macOS and nothing says so — the model writes GNU sed/date/stat and patches around the failure
- Capability/state claims answered from generic training default instead of checking available tools
- I need clarification on the bug report to generate a GitHub issue title. Please provide: 1. What is the actual problem or error occurring? 2. What is the expected behavior? 3. Any error messages or logs? 4. Steps to reproduce the issue? Once you provide
- [Bug] Shell glob patterns incorrectly interpreted by zsh causing empty agent returns
- [Bug] Compaction summary renders above the last message pre-compaction
- [BUG] Multi-repo cloud sessions map repos to git remotes incorrectly; `origin` is arbitrary
- [FEATURE] Populate `rate_limits` in statusLine JSON before the first API call of a session
- [FEATURE] Expose subscription billing-cycle / renewal date in statusline JSON (or via CLI)
- [FEATURE] Expose remaining API credit balance in statusLine JSON (or via CLI)
- [Bug] Infinite loop in agent execution with overcomplicated instruction handling
- [BUG] Windows desktop: composer dictation randomly duplicates whole sentences (intermittent), mic auto-stops after ~1 min
- [BUG] Claude in Chrome MCP: every action denied on one domain (polat.docuware.cloud) while sibling domain works; no prompt, no local setting
- Model destroyed a 410KB user file: open(path,"w") truncated it, then the write threw
- [Bug] Model ignores CLAUDE.md rules for file context requirements despite explicit instructions
- [Bug] Unclear error message or parsing issue with user input
- [BUG] VS Code extension: model/effort picker pill is stale, disconnected from the session's actual resolved settings
- VS Code: context indicator in the bottom toolbar runs /compact on a single click — make it a Usage menu with Compact inside
- [FEATURE] PostRewind hook event: hooks that manage external state cannot observe /rewind (re-files #18858, #86283)
- Session metadata keeps a stale worktreePath/worktreeName after the worktree directory is removed
- Live, data-driven argument autocomplete for custom slash commands/skills
- Cached prState in list_sessions goes stale after a PR merges
- [BUG] PreToolUse hook `ask` resolves correctly in interactive mode but is silently denied in headless/stream-json mode, because the `PermissionRequest` hook never fires
- Desktop app: dangling plugin hook (security_reminder_hook.py missing) fires endless 'Background security review found issues' notifications
- Feature request: prompt-cache keepalive while background tasks or permission prompts are pending
- [Bug] Anthropic API Error: Safeguarding block on invoice generation workflow
- Background subagents give the user zero visibility; model waits silently for 15+ minutes
- Agent made unrequested production changes, deployed partially, reported it complete, then drafted a public report containing the user's private identifiers
- [BUG] Summarized thinking is shown in English in a fully Japanese conversation (re-report of #82785)
- [FEATURE] Inline quote-replies send immediately instead of accumulating
- [Feature Request] Add credentials verification system for relaxed biosafety content policies
- [Desktop 2.2553.1.0] Slash picker opens only at position 0, even after an accepted /skill
- Recurring report task: model orchestration got slower and more token-hungry than previous cycles
- [Bug] TUI fullscreen mode not displaying typed input on Windows
- macOS: session processes are denied Local Network access despite the grant (bundle_id: (null) at evaluation)
- [Bug] Future-dated OAuth refresh lock blocks token refresh indefinitely
- [Bug] Process termination fails when killing another terminal instance
- Bash "ask" rules stop prompting under --dangerously-skip-permissions, contradicting the documentation
- Duplicate/stale background-task completion notifications fire repeatedly after the task is already gone
- [BUG] Fable 5.1 configured at `high` effort, but model receives `reasoning_effort: 10`
- [FEATURE] Allow a non-Anthropic model provider alongside the subscription in a Remote Control session
- CLAUDE.md is present in context after /compact but no longer followed — an imperative SessionStart hook in the same context is
- If you run out of 5hr usage in a project, and then upgrade your account, the pr…
- [Bug] macOS option key combinations not recognized for diacritical marks input
- [Bug] `/compact` creates duplicate live sessions operating on same directory
- [Bug] Auto mode classifier allowed unauthorized push to main branch without explicit confirmation
- this project is stuck. others are working fine but this one says claude is busy…
- miniaturized windows to show questions
- [FEATURE] sandbox credentials mask: substitute the sentinel inside HTTP Basic auth (Base64) so git over HTTPS can authenticate
- I am not able to see my project conversation from my phone. Even when I try to …
- [BUG] Desktop 2.2553.1 (macOS): Run button on bash blocks and markdown links do nothing on click in Code tab
- Windows: per-second statusLine bash spawns turn the MSYS add_item/errno 1 race into a permanent outage of every Bash call
Resolved issues (244)
- Environment "Network access: Full" setting not applied to session containers — outbound HTTPS still blocked (403 at gateway)
- [Bug] CLI refuses to execute user commands and systematically avoids requests
- Bug / Authentication Error.
- [FEATURE] Make the Claude iOS app (Claude Code mobile client) available on Apple Vision Pro
- [BUG] I am unable to see my research history on the desk top app in mac. but is all available on web.
- [FEATURE] Warn before Console API credits expire, the way subscription usage credits already do
- [BUG] Desktop app chat sidebar stuck empty after forced re-authentication (right after chat/collaboration merge announcement) — data confirmed intact on claude.ai
- [Bug] Anthropic API Error: Legitimate Kafka/Redis integration flagged as security risk
- [Bug] Claude loses context of offline tasks after reconnection
- [Bug] Repeated UP errors after exit and re-enter
- [Bug] Previously approved code flagged repeatedly during security review
- [Bug] File copy operation fails between development directories
- [MODEL]
- [BUG] Switching in between conversation and agent session causes content load render issue
- [Bug] Unnecessary safeguard trip when reviewing agent output for internal capabilities
- [Feature Request] Add setting to restore collapsed cross-session message display
- [Feature Request] Add domain exemption or context awareness for legitimate security engineering workflows
- Desktop app: no way to pin/prompt for project folder on new session (defaults to last-used)
- Plugin relevance: cap how often a given plugin can be suggested (overlapping plugins create persistent noise)
- [Feature Request] Add safeguard override or exception handling for legitimate coding tasks
- [FEATURE] Improve semantic command search
- [FEATURE] Hook-writable context block that gets replaced instead of appended
- [BUG] Claude Code Chat isn't working
- Slash-command session titles ignore command args, and command-launched sessions cannot name themselves
- [BUG] Claude Code Has No Way To Dismiss a "Commit changes" chips.
- Agent repeatedly extends explicit instructions into unrequested git/GitHub actions (push, PR creation, branch deletion) — even right after writing its own rule against this
- [Bug] Output contains unexpected asterisks censoring text content
- Connected MCP connector (e.g. Google Calendar) not available for Cloud Routines / RemoteTrigger
- [BUG] 2.1.222 escalates #82966 to ordinary sessions: worktree isolation refuses every `source`/`.` via an interpreter denylist, with no opt-out
- AskUserQuestion: payloads containing German typographic quotes (U+201E/U+201C) deterministically fail with "questions expected as array but provided as string"
- [BUG] Cannot log in to Claude Code in VS Code — verification emails never delivered
- [BUG] Summarized thinking language does not follow conversation language — English/Japanese mixed within one session
- [Feature Request] Improve German language translation quality for technical/programming contextsw
- Stop hook config leaks across git worktree sessions
- [BUG] 1Password for Claude: autofill_credential always tab_unavailable / agenticModeNotEnabled from Claude Code sessions in the desktop app (14/14, 0 filled)
- Monitor tool: stdout-line notifications intermittently held 30-90s despite subprocess printing on time
- [Bug] Opus 4.8 reasoning regression: failed document search and tool-use persistence
- [Bug] Slash-command catalog not rebuilt after mid-session plugin upgrade in desktop client
- [FEATURE] semantic / self-monitoring triggers for skills (a "daemon" layer complementing event hooks)
- statusLine config: support position field to align custom status to the right side of the row
- [FEATURE] Anchor the viewport at the TOP of each new response (read forward, not scroll back)
- [FEATURE] LIMITS
- [FEATURE] Show shutdown progress/status when quitting takes long due to MCP cleanup
- [Feature Request] Support for official bug bounty program participation
- [Feature Request] Add text wrapping options to /export command
- Assistant answers get visually buried behind tool-call/wakeup blocks in long autonomous sessions
- You've hit your session limit · resets 8pm (America/Mexico_City)
- [FEATURE] Record the source session_ id in teleported session transcripts
- [DOCS] `claude agents` empty-state description in Troubleshooting is stale; v2.1.203 replaces the onboarding hint with the Needs input / Working / Completed sections (with descriptions)
- [Feature Request] Add expert-level security scenarios for experienced administrators
- [FEATURE] Automatic token/usage refund when a response fails to meet the request
- [Feature Request] Add customizable interrupt prompt text in settings.json
- [Feature Request] Persistent session color configuration via settings.json or CLI flag
- [Feature Request] Add transparency and steering controls for agent execution
- [FEATURE] Per-phase effort levels for opusplan (separate plan-mode and execution effort)
- [Feature Request] Mid-cycle upgrade should rescale weekly allowance immediately, not at next reset
- [Feature Request] Allow user override for Fable model selection when blocked by safety policies
- Sync local CLI session history to the cloud for cross-device resume (parity with Codex)
- [BUG] No project-level way to lift the Task/Agent ask-first default for task categories (only per-agent description "PROACTIVELY")
- [FEATURE] Expose plugin user_config values to skill script assets
- [Feature Request] Add option to disable exit prompt when monitor is actively running
- [FEATURE] Expose exact remaining-context token count to Dispatch/Cowork sessions
- [BUG] Session feedback buttons "Fine" vs "Good" are ambiguous for non-native English speakers
- [FEATURE] Include teammate sessionId in Agent Teams config.json for session→transcript mapping
- [BUG] I think it's time for Claude to implement the Self Limit Reset feature (just like Codex did)
- [BUG] Scale the Recents session indicator by number of PRs merged (multi-merge sessions should stand out)
- [FEATURE] VS Code extension: support pasting images into chat input
- Feature request: selective text copy within messages (mobile app + desktop), not only whole-message copy
- [FEATURE] Compact display option for in-process agent team panel (task text makes it tall in VS Code integrated terminal)
- VS Code: long Claude Code session tab titles are truncated in Quick Open (Ctrl+P) with no way to shorten them
- Preserve original filename when images are pasted/attached into chat
- [FEATURE]: Enterprise setting to exempt statusLine from allowManagedHooksOnly / disableAllHooks
- [FEATURE] Windows: Add toggle to disable quick entry when clicking taskbar icon
- [Windows] Terminal/PowerShell windows flash and disappear without user command
- [Feature Request] Improve iOS app modernization support and expand real-world use cases
- [Feature Request] Allow security auditing of self-generated code without safeguard triggers
- [FEATURE] Claude apps gateway: threshold notifications (80% warning / 90% critical / 100% exceeded) for daily, weekly, and monthly spend limits
- [Feature Request] Add token usage reset capability for failed operations
- [FEATURE] Add Ctrl+D in the /resume list to Remove The session
- Accessibility: /workflows and background-task progress have no screen-reader-readable text equivalent
- [FEATURE] Need to pause an agent
- [Feature Request] Overly restrictive safety filtering for legitimate hardware design use cases
- Workflow scripts: add a sanctioned sh() primitive (allowlisted) for deterministic shell hops
- Feature request: retract/sync channel permission prompts when resolved elsewhere (Discord plugin)
- Computer Use's per-app request_access model pushes users toward a strictly more dangerous standing-permission alternative when it fails to recognize an app
- Give the model provenance & timing context for user input (bang commands + queued/in-flight prompts)
- /workflows slash command not recognized in VS Code extension
- [Bug][aup] Safety block wrongly flagged legitimate aerospace HUD feature request (req_011CcjW3iHVpsQcRni1E5UFp)
- [FEATURE] Auto-mode classifier should see loaded Skill/SKILL.md content, not just CLAUDE.md
- Slash-command picker: no way to preview available sub-commands/verbs beyond the one-line description
- I appreciate you reaching out, but I'm not able to generate a GitHub issue title from this report because: 1. The message is in Czech and describes a legal/regulatory concern rather than a technical bug or feature request for Claude Code 2. It doesn't des
- Exit-time uncommitted-changes warning has no way to exclude specific paths
- [FEATURE] Non-managed System-wide `settings.json` (machine-wide defaults that users can still override)
- [Feature Request] Artifact tool missing delete/unpublish action
- [Feature Request] Add configuration option to adjust safety guardrails for testing
- [FEATURE] Feature request: open a specific session in the sidebar (URI location param or sidebar.openSession command)
- [Feature Request] Reconsider MoE architecture - performance regression compared to Opus 4 variants
- [FEATURE] copy code icon
- [Feature Request] Localize prompt suggestions based on response language and tone settings
- Feature request: agent-requestable /goal arming with one-keypress user confirmation
- Deep link: allow opening a session in the Side Bar (/open?session=…&target=sidebar)
- pr-review-toolkit: silent-failure-hunter agent has invalid YAML frontmatter, loads with empty metadata
- [Feature Request] Cyber Verification Program permissions not applied for approved operations
- /color command: support arbitrary hex colors (terminals already support truecolor)
- [Feature Request] Add safety research exemption for model security evaluation use cases
- Fan-out research skills (deep-research) inherit the session model for all subagents — no model tiering — burning premium-model quota in a single run
- [BUG] Auto mode: ask rules matching shell operators (`Bash(* >*)`, `Bash(* | tee*)`) never fire; text-matching ask rules do
- [BUG] sandbox.enableWeakerNestedSandbox silently forced to false when CLAUDE_CODE_SUBPROCESS_ENV_SCRUB is set (breaks Bash sandbox in cap-dropped nested Docker)
- Chrome extension never reaches the pairing service on Windows 11 — list_connected_browsers returns [] with a fully verified install (Chrome 151)
- Feature request: user-configurable top-bar buttons to invoke custom commands/agents
- [BUG] Claude Desktop (Windows/MSIX): repeated silent process termination preceded by PerAppRuntimeBroker DCOM activation denial; first occurrence left package unlaunchable and destroyed Code tab history
- You've hit your session limit · resets 1:10am (Europe/Moscow)
- [Bug] Agent Incorrectly Flags 3D Modeling Task as Cybersecurity Issue
- Channel events are never surfaced to the model in non-interactive (-p) sessions
- [Bug] Session ID collision causes empty scrollback and "already in use" errors in background sessions
- [Bug] Unnecessary file read causes excessive token consumption in role session
- [BUG] dark-daltonized: userMessageBackground renders at 1.18:1 on a default dark terminal, leaving user messages with no visible separator
- [FEATURE] - Create a Ticket for Suggested Tasks
- Feature proposal: first-class "bring your own mesh" (ZTNA) for MCP + Remote Control — from a phone↔PC mirror-bridge field setup
- [BUG] teammateDefaultModel in settings.json is silently ignored, while the published JSON schema declares it valid there
- [BUG]
- [BUG] tmux/pane-backed teammates have no agent_id/agent_type in hook payloads — indistinguishable from the main agent (re-file of #54785, still present on 2.1.233)
- [Bug] File mention autocomplete ranking ignores exact filename matches
- Possible: resume from the agents UI may apply the global default model instead of the session's (evidence withdrawn, needs verification)
- [BUG] Embedded CLAUDE.md intermittently reaches the API with one multibyte character replaced by three U+FFFD, changing the prompt prefix mid-session and defeating prompt caching (2.1.258, Windows)
- [BUG] Mixed Arabic/English (RTL+LTR) input renders scrambled with wrong cursor position in macOS Terminal.app (v2.1.278)
- CLAUDE.md / auto-memory block is never shared across sessions: it sits in messages[0] after the system breakpoint, with no cache_control of its own
- [Bug] Compaction result not displayed in transcript mode with Ctrl+O
- [Bug] A path-scoped rule file is never registered, while a byte-identical copy at any other path loads
- Conversation sidebar not navigable with stylus or keyboard - resizer covers scrollbar, list has no focusable container
- [Bug] Agent hallucinating and ignoring claude.md rules during task execution
- [Bug] Slow performance when sending emails with ZIP attachments
- [Bug] Unable to parse or handle "UP" command
- Claude in Chrome extension not connecting to Claude Code (tabs_context_mpc returns "extension not connected")
- [BUG] cannot install claude terminal on my windows 11
- [Bug Report: Unable to process input]
- Subagent liveness/death observability for long-running orchestrators (per-agent deadline, crash-path notification, status poll)
- [BUG] Desktop sidebar regression: per-project chat nesting removed, projects no longer expandable
- [BUG] Prompt suggestions silently suppressed whenever the client-derived rate-limit status is allowed_warning — strict-equality gate in Vxy
- Worktree isolation guard validates a pinned session's Bash/Edit/Write against another session's worktree (newest-created / shared pin key) under concurrent worktree churn
- Gateway /login reliably hangs at "press Enter to continue" (login succeeds server-side)
- /code-review ultra fails with "no commits yet" despite valid commit history
- [Feature Request] Add privacy guardrail warning for diagnostic command outputs
- Feature request: drag-to-reorder session groups + folder-linked groups (shared context, progress roll-up, concurrent-write gating)
- Feature request: collapse resolved question/reasoning history in transcript view
- Add "Waiting for Review" state to background job list
- [FEATURE] Claude Code for Visual Studio: default all permission checkboxes to ON when panel opens
- [Feature Request] Improve content policy context handling for legitimate substance research and harm reduction projects
- [Feature Request] Support "nuke" terminology in documentation/error messages
- Plugin commands/skills: support `aliases` for NL routing + optional native disambiguation
- [FEATURE] Cowork Project Context: allow custom link labels instead of raw URLs
- Expose the client surface (GUI/desktop vs TUI vs IDE) to the model
- [FEATURE] Add @conversation / @session reference to link other sessions as context
- [FEATURE] VS Code extension to support VS Code workspaces
- Malformed tool call ("could not be parsed"): expose a hook event + make the malformed-retry count configurable
- Feature request: support /bg (background session) and agent view in the desktop app
- [Feature Request] Add "continue" option when usage limits are reached to resume after reset
- [DOCS] Terminal config: Shift+Enter DOES work in JetBrains terminals when launched via the IDE plugin button
- Docs: fallbackModel behavior for subagents with frontmatter-pinned models is unspecified
- [BUG] Scrollback buffer cleared when streaming output arrives while user is scrolled up
- Session name label in app UI doesn't update after renaming via Ctrl+R in resume picker
- [Bug] Bash-tool grep wrapper still silently returns nothing on text files with a stray NUL byte (2.1.220) — #56644 was auto-closed as stale, not fixed
- [BUG] Background Agent tasks persist as "running" in /workflows after full process restart on both dispatch and viewing hosts
- Edit tool false-positive JSON validation failure on .claude/settings.json hook commands
- [BUG] Typeface contrast variation in Desktop Code sessions
- [BUG] Whole message renders as raw markdown when the first markdown construct starts after character ~500
- [BUG] Conectar github con claude como editor
- Feature request: persistent voice conversations with Claude Code (no idle disconnects)
- [BUG] macOS desktop restart permanently orphans all pre-restart sessions from the Android app — archived, then stuck on "connecting" forever
- [BUG] iOS Simulator panel (Japanese locale): "Resolution" mistranslated as "解決" instead of "解像度"
- [Bug] Anthropic API Error: Internal Server Error (500)
- [BUG] WebFetch fabricates a live job posting for a URL that is actually a 500 error page (Cowork Scheduled Tasks)
- Background agent panel resizing shifts the entire transcript and input box vertically while typing
- [BUG] Routine-fired runs that create a fresh cloud session crash at Claude Code startup (turns=0); interactive sessions and into-existing-session deliveries work
- Headless -p "/skill" with a failing !-preamble silently aborts pre-turn and reports subtype:success, num_turns:0
- [BUG] MCP server environment variable gets corrupted/redacted with bullet characters before reaching subprocess
- [FEATURE] Permission dialog: make the digit for "No" stable across the 2-option and 3-option variants (implementation TBD)
- [BUG] Claude in Chrome: navigate fails with "Navigation to this domain is not allowed" for every domain, including after a full browser restart and re-pairing; no consent prompt ever surfaces
- claude-in-chrome never connects to a healthy native-messaging bridge (Linux)
- [Bug] Fork on completed /btw response re-submits original prompt instead of continuing session
- [BUG] Claude Desktop macOS: filesystem MCP server unusable in both package generations - new schema never dispatched, old schema dropped at registration
- [Bug] Anthropic API safety filter incorrectly blocks legitimate developer documentation
- Claude-in-Chrome: browser downloads silently fail to reach workspace (no JS error).
- [BUG] Subagent rate-limit failures are surfaced as a user prompt after the limit has already reset
- [BUG] Claude Code Violates Settings
- MCP elicitation dialog blocks scrolling the conversation behind it
- [BUG] NZBGet config-write during extension install silently drops existing settings; session self-verification (queue "resumed") missed it, mislabeled as success in project notes
- [BUG] Auto refusing dangerous command that are explicitly whitelisted
- [Bug] Claude Code fails to complete basic tasks reliably
- [Bug] Claude model failing to read and follow documented instructions and memory context
- [BUG] Built-in bwrap sandbox: launch races transient git lockfiles under a shared `.git` → bwrap exits 1 "Can't get type of source .../config.lock"
- [DOCS] Docs promise repo-declared plugins install in cloud sessions; CLI deliberately skips them
- [BUG] Parent resumes as if subagent completed while subagent is waiting in Monitor for a background process
- [BUG] Skills-dir plugins don't load in subfolders of a trusted directory — every launch dir silently gets hasTrustDialogAccepted:false, no dialog ever shows (Linux, no git)
- [FEATURE] Emacs/tmux-style keyboard pane splitting (split-right / split-down) in desktop Code tab; new sessions shouldn't replace an open pane
- Session/project sidebar empty after app crashed, relaunch blocked by file lock, force-kill + relogin
- security-guidance: sg-python.sh discards probe stderr, misreporting any failing interpreter as 'no working Python 3'
- [BUG] Desktop app: chat pane silently omits a contiguous ~25h window of turns (local transcript is complete)
- [BUG] Lack of tool output fidelity
- [Bug] Anthropic API Error: Overly aggressive content filtering blocking benign messages
- /reload-plugins re-injects full skill roster without evicting the session-start copy (+~7.5k tokens per reload)
- [BUG] Parking a mid-flight turn leaves the interactive session's probe stuck at status: "busy" forever (2.1.220)
- Archived session's background process keeps running after archive
- [Bug] False positive detection in code analysis
- archive_session("self") deletes the worktree but resumes the session onto it instead of terminating
- [Bug] Claude Code fails to identify application errors and code quality issues
- send_message to a running session can silently fail to persist - content confirmed sent, never appears in target's transcript
- [BUG] Interactive CLI fails after tool calls while Desktop Code and claude -p work reliably
- [BUG] Apple in-app Pro subscription recorded as claude_free — Dispatch unavailable on Windows desktop
- [BUG] Weekly + Fable quota jumped 0% -> 50%/100% with account idle (Max 20x, after Jul 31 reset)
- /auto-mode-setup always fails: "removeFromPermissionsAllow[0] is not a rule string the removal offer could have produced"
- [Bug] OTel metrics: skills from non-official plugin marketplaces report skill.name="third-party" with no opt-out — teams lose analytics for their own skills
- Opus 5 built a parallel implementation of a complete existing subsystem, then denied it when asked directly
- [BUG] [Cowork] Plugin-bundled stdio MCP servers silently dropped in cloud sessions (skill loads, MCP does not)
- otelHeadersHelper headers silently dropped for gRPC OTLP exporters (metrics/logs/traces) — falls back to no auth
- [BUG] remote-control --no-create-session-in-dir prevents session resume
- Vertex: `thinking.display` dropped from request body under CLAUDE_CODE_USE_VERTEX=1 — Sonnet 5 thinking text invisible on Vertex
- You've hit your session limit · resets 6:20am (America/New_York)
- [MODEL] Opus advising Opus
- Worktree cleanup deletes live worktrees when a repo is opened from both Windows and WSL (gitdir path-spelling mismatch makes every cross-side worktree look prunable)
- [BUG] autoMode in project settings is silently ignored — settings schema does not mark it user-only, unlike sibling restricted keys
- PushNotification skips even when the terminal/VS Code window is closed or minimized
- [Bug] False Positive in Opus Security Scan
- Memory-file naming: distinguish 'role' (capability/function) from 'relationship' (behavioral norms)
- [Bug] Anthropic API Error: Persistent 429 rate_limit_error despite low usage (1%)
- [BUG] https://github.com/anthropics/claude-code/issues/69354..Issue still exists
- [Bug] Conversation compacting hangs indefinitely during token reduction and uses huge tokens
- [FEATURE] Add voice/audio output for Claude's responses in Claude Code
- [BUG] Any link generated by the CLI that contains a URL inside parentheses is unusable
- Sessions sidebar: option to limit sessions shown per project/folder group
- [BUG] Sessions that exit with live background tasks stay "active" forever — orphan reconciliation only runs on resume
- [BUG]vm_workspace_diagnosis
- [BUG] Claude Code installer hangs indefinitely at "Setting up Claude Code..." on Windows Server 2019 — process spins CPU with no network activity
- Remote Control app: file delivery via SendUserFile fails silently (red error triangle), tool reports success
- Sandboxed Bash fails with "bwrap: Can't create file at .git: Is a directory" in normal (non-worktree) git repos
- `API Error: The socket connection was closed unexpectedly` — packet captures show server-initiated FIN mid-stream (10 incidents, requestIds included)
- Feedback prompt is a single non-scrolling line — text disappears off-screen after a few words
- Agent SDK bundled CLI stalls after final message: transcript has complete end_turn message, but stream-json result is never emitted
- [BUG] Approving a forwarded teammate permission request with a message silently drops the message
- [MODEL] [MODEL] Repeated inefficient actions during browser automation (stale click coordinates, wrong keyboard shortcut, keyword-guessing) burn excessive session budget
- [Feature Request] Add official read-only transcript viewer for session logs
- [BUG] Plan mode: after "No, keep planning" feedback, the revised plan is never re-shown and edits start without a second approval
- [Billing Bug] Unauthorized Pro→Max upgrade
- Cross-session message: sender label shows generic string instead of source session title