Digest / September 19, 2026
Claude Code Issues Digest — September 19, 2026
121 new issues filed ·
245 resolved in anthropics/claude-code (UTC day).
New issues (121)
- [BUG] "Last updated" refresh does not update the remaining credit
- [BUG] Windows/MSIX Desktop: bundled Claude Code engine retries a failing fswatch-probe open ~38,000x/s forever, leaking ntfs.sys NtFC nonpaged pool ~230 MB/min until reboot (workaround: CLAUDE_CODE_TMPDIR)
- [Bug] Content filter blocks "cyber" keyword in project names
- [BUG] Remote Control is not restored for sessions killed by a stealth auto-update
- [Bug] Claude provides inaccurate scientific analysis with false confidence and resists correction
- [Bug] Claude degraded performance: inconsistent error corrections across codebase
- Agent repeatedly overrode explicit scope, presented fake test results as real, and committed without permission
- Agent stated unverified work as fact, and hedged about itself to deflect accountability
- Chats/Projects sidebar section disappears after updating to 2.2553.1 (Windows)
- WebFetch/WebSearch fail with "400 adaptive thinking is not supported on this model" under Sonnet medium
- [Bug] Claude Desktop loses Local environment after Windows reboot
- [BUG] Desktop/Code: MCP widget sandbox frame blocked by CSP renders nothing — no error surfaced, and the tool result still reports "rendered"
- [Bug] Fable 5 safeguards chain-flag entire sessions: children's song lyrics and routine Korean e-commerce instructions all flagged (6 request IDs)
- [FEATURE] Mobile app: more context to Code chats
- Desktop-hosted session never restarts a stdio MCP server after it exits, and keeps reporting it as connected
- [BUG] Small image (558x162px) incorrectly rejected with "exceeds max allowed size... 2000 pixels" error
- [BUG] MCP tools silently dropped when inputSchema uses root-level allOf/if/then, which MCP 2026-07-28 permits
- Model reports a written instruction back as a completed fact, and writes the stale state into live files
- [BUG] Desktop session reset and vanished from list/archive after clicking a Claude-generated link — data still intact on Android
- LSP: hint-severity (tagged unnecessary) diagnostics are pushed into model context as new-diagnostics — token waste, no way to filter
- [FEATURE] Desktop app (Code tab): render hover hints from markdown link titles and <abbr title> in assistant messages
- [FEATURE] clear button for dispatch
- [BUG] Desktop dictation: live transcript is correct Latin script, but finalization rewrites the whole paragraph into Devanagari
- [BUG] Claude Code VS Code extension: session repeatedly stalls/hangs at async handoff points (tool dispatch, background-task notification, context compaction) — requires reload/reboot to recover
- [BUG] Text copied from the Claude Code TUI and pasted into the prompt is shown on separate lines in the sent message
- [FEATURE] Allow swapping Enter/Shift+Enter behavior in the VS Code extension's chat input
- [BUG] windows不能运行
- Claude Desktop's fileTypeAssociation manifest invalidates existing UserChoice defaults for Office/dev file types on Windows
- [FEATURE] Desktop app model picker should respect modelPicker order (or offer its own reordering)
- Compose box re-inserts previously sent message text after submitting a new one
- [Bug] Claude Code enters idle state and requires manual intervention to resume tasks
- Uninstalled marketplace plugin silently claims a slash-command name and breaks a local skill (reported as an arity error)
- [BUG] Agent ran `timeout 5 <electron-app> --version` as a sanity check, orphaned Electron child processes froze the desktop (hard reboot required)
- [BUG] /goal + /btw bad interaction
- [BUG] Terminal panel: Couldn't send the command's output to Claude - shell silently reconnects mid-command
- [FEATURE] Allow to disable token count warning[BUG]
- [BUG] Stop hook stop-hook-git-check.sh: unpushed-commit check silently passes on branches with no remote ref, and false-positives after a merged PR
- [BUG] Cowork built-in browser panel disappears after app update 2.2553.0 → 2.2553.1.0 (Windows): its tools return 400, then vanish from the tool set after restart
- [DOCS] Path-scoped rules: `*.md` matches nested files, not only the project root
- [BUG] Falta el ícono de copiar en mensajes/artifacts, y "Copiar" del clic derecho aparece desactivado
- [FEATURE] Let Claude artifacts opt into the Git-backed repo/branch/PR workflow that Claude Code on the web already has
- Bug: bwrap sandbox binds protected paths snapshotted at session start; deleting one mid-session breaks all later Bash
- [Bug] Anthropic API Error: Reasoning Extraction Failed
- [MODEL] Written instructions acknowledged, then not implemented in code — repeated pattern in one long pipeline session
- [BUG] 2.1.278: a file redirection on a single excluded command defeats sandbox.excludedCommands (gh fails TLS)
- [Bug] Excessive flagging on minimal input like "Hi"
- [BUG] OAuth and magic-link login redirect existing Max subscriber to "Let's create your account" onboarding — browser and desktop locked out, iOS + Console still authenticated
- Desktop app (Code tab): allow collapsing or hiding the per-repo changes summary above the composer
- [BUG] Cowork: legacy multi-folder projects disappeared from the account (desktop 2.2553.1) — conversations survive but are ungrouped
- [Bug] Excessive token consumption due to model hallucinations and unreliable code generation
- Sandbox denyRead entries scale spawn argv past macOS ARG_MAX (E2BIG)
- [BUG] All naver.com domains blocked by default safety restrictions — false positive on Korea's primary portal (daum.net / tistory.com not blocked)
- agents-md README references the # memory shortcut removed in 2.0.70
- [FEATURE] Don't truncate long diff lines (2000-char cap hides edits in long paragraphs)
- [FEATURE] Session continuity between Cowork and the Microsoft 365 add-ins
- [FEATURE] /insights --project <path>: scope the report to a single project directory
- Desktop app: local file links never open in a Code session that was started from the phone via Remote Control
- [BUG] Claude Desktop (macOS): in-flight completion stream dies with "TypeError: network error" when router deprecates/renumbers IPv6 prefixes; not retried, response lost
- [BUG] docx skill's merge_runs.py produces .docx files that Google Drive cannot open
- [BUG] Formal protest: messages from the paired, allowlisted owner on the Telegram channel are labeled "NOT from your user / untrusted external data"
- [FEATURE] Chat input placeholder shows a predicted "next reply" suggestion
- [BUG] UNACCEPTABLE: Claude Code now treats its own owner as an untrusted stranger, AND blocks the owner from saving a memory to fix it
- Subagent self-corrects from stale model id (claude-opus-4-8) to current (claude-opus-5) mid-execution, no local config to blame
- [BUG] Desktop: installing or updating any .mcpb executes every python3 on PATH — x86_64 interpreters launch under Rosetta on Apple Silicon
- [BUG] Image attachment fails to decode/resize intermittently for visually identical JPEGs (same dimensions, same encoder
- [BUG] Kitty notification path omits OSC 99 f= and n=, so notifications fall back to kitty's bundled logo and bypass the icon theme
- [FEATURE] Desktop: plan session sees status of tasks spawned via chips (orchestrator view)
- [BUG] Transcript rows zero every usage counter but keep usage.cache_creation — cache_creation_input_tokens != sum(cache_creation)
- Worktree-isolated sessions refuse git after `cd ~/path` but allow the same command with `"$HOME/path"` (2.1.278)
- [BUG] Cowork Windows: device_bash still cannot mount connected folders AFTER KB5129195 — patched 25H2 build 26200.9457, rebooted, fresh session, fresh folder attach
- Render MCP server icons (serverInfo.icons) in tool lists and permission prompts
- macOS 27: Claude Desktop silently stops updating when Background App Activity is disabled
- [BUG] 2.1.277: sandbox.excludedCommands no longer matches an executable whose basename is `read`
- [Bug] Anthropic API: False positive `reasoning_extraction` refusals on benign content
- Feedback
- Agent tool honours a `name` parameter that the schema shown to the model omits
- Voice dictation: truncated copy of the sent message reappears in the composer
- [FEATURE] /insights: weight the corpus by volume and recency instead of counting every session equally
- Native binary hangs silently at 100% CPU on VMs with the kvm64 CPU model (no SSE4/POPCNT) — needs a CPU feature pre-flight check
- [BUG] [Claude Desktop / macOS] Registers as default handler for .md without consent (not claude-cli://)
- Locked out: sign-in returns "Create your account" via BOTH Google SSO and email code, while the API reports the same address owns an active Max account
- [Bug] CVP users experiencing access blocks
- [BUG] Linux/npm: auto-update leaves running sessions alive with zero sockets — Remote Control silently dead (no relaunch)
- [Bug] Token cost constraint handling allows unlimited execution when disabled
- Desktop Code tab system prompt still lists /doctor as unavailable, but it is a bundled skill and runs there
- [Feature] VS Code: a plain "Resend this message" (same text and attachments) in the per-message menu
- completely fails : reset billed and charged, credits ineffective
- [BUG] WSL: package-manager installs are classified "native" — ownsExecutable returns before it can run dpkg -S / rpm -qf
- [BUG] git push of tags fails with 403 from Claude Code cloud session, despite full GitHub App permissions and no repo protections
- Remote Control fails to connect with "connect timed out" — reproducible, session confirmed alive
- [BUG] Claude Code 2.1.277 crashes on startup
- [FEATURE] Allow manual editing of proposed changes in the diff view before accepting
- [BUG] Windows: Claude Code computer use leaves the desktop window stuck always-on-top (cu-side-panel restore races with Win32 screenshot re-pin)
- [FEATURE] Show Remote Control sessions in the `claude agents` view (follow-up to #77720)
- Skill catalog descriptions intermittently missing from system prompt (frontmatter on disk is correct)
- [BUG] Bare #NNN issue refs linkify to a stale, pre-transfer repo owner despite correct remote/gh config
- [Bug] Anthropic API Error: Safeguards falsely flagging compliance discussion messages
- [BUG] Argument order of -p and --append-system-prompt changes print-mode prompt behavior
- [Bug] Incomplete bug report: Missing error details and reproduction steps
- Bug Report: CoworkVMService blocks MSIX update on Project Server (Claude Desktop 2.2553.1.0)
- [BUG] AGENTS.md loading is intermittent on 2.1.278: same directory and flag state, some sessions load the files and some silently do not
- [BUG] Assigned working folder stays locked by background process after closing chat, blocking deletion (Windows)
- [Bug] Model behavior inconsistency between individual and enterprise account configurations
- [BUG] Desktop: one-time scheduled tasks vanish from the Scheduled/Routines list (recurring tasks unaffected; tasks still exist underneath)
- [BUG] Chat response completion notifications silent on desktop app (Windows + macOS) while Code/Cowork notifications fire correctly
- Unauthorized Financial Transaction - AI Assistant Bypassed Explicit Safety Interlock
- [MODEL] Assistant emitted a self-generated prompt-injection payload (fake 'application/vnd.ant.overwrite-conversation' header + fabricated user turn) as its own output
- [BUG] Claude Desktop 2.2553.1 (Cowork) alphabetically sorts keys of nested objects in MCP tool call arguments
- [Bug] Anthropic API Error: Message Flagged - Research Project Blocked
- Expose current auth method (subscription vs API key) in status line payload
- [BUG] Linux Desktop: OAuth login callback dropped as "repeat delivery", making /login recovery impossible
- Agent-initiated multi-agent runs exhaust the 5-hour window with no estimate and no warning
- [BUG] Every background Agent-tool completion delivers two separate parent-turn events (SubagentHandback + redundant task-notification)
- [Bug] Sonnet 5 [bio] classifier false positive on Claude.ai web chat — DHCP/DNS/AD infrastructure content (NOT Claude Code CLI)
- [Bug] False positive security flag on video frame rotation code
- [Bug] Unclear Bug Report: "A rude tone" — insufficient detail to reproduce
- [FEATURE] /continue OR /complete to continue the chat where it stopped.
- bypassPermissions + wildcard allow rule still ignored on the first MCP tool call after a desktop app restart (follow-up to #87358)
- [BUG] Remote Control fails to connect for existing local session (stuck at "connecting" / times out)
- ListAgents gives two live sessions the same bracketed ref, in the line that tells each how it is addressed
- Worktree isolation refuses fifteen shell words in any argv position, including in read-only commands containing no git
Resolved issues (245)
- UX: permission prompt buttons are stacked vertically — easy to press the wrong one
- [BUG] Claude Desktop: tools/call discarded when the chat is switched before dispatch
- [BUG] Windows Git Bash: Bash tool brace-expands inside quoted strings and splits quoted paths containing spaces
- Desktop app shows "process already open elsewhere" and quits — single-instance lock false positive
- [FEATURE] For Claude Sonnet, add an option in settings for system time at the end of every prompt and at the beginning of every response. NOTE: There have been other requests for timestamps in the chat.
- [Model Behavior] Claude Science (Windows) Safety classifier cascade degrades research to Haiku 4.5, which then hallucinates, destroying valid scientific work
- "notification push Gmail qui cesse silencieusement après ~7 jours, trigger type Email/API".
- [BUG] Claude Desktop completion chime cannot be disabled, plays repeatedly, no app setting exists
- [FEATURE] Session continuity between Cowork and the Microsoft 365 add-ins
- [Feature Request] Support for Claude Opus 5.0 and newer model versions
- GitHub MCP connector app is missing Administration permission, blocking repo creation
- [Feature Request] Improve security classifier to reduce false positives on authorized defensive engineering and security-product development tasks
- [BUG] remote-control rejects --agent, --append-system-prompt and --settings, so server-spawned sessions cannot be given a role at launch
- [FEATURE] Surface CLI-style tips/hints in the VS Code extension panel
- [Feature Request] Anthropic API safety filters blocking legitimate healthcare and lifescience research use cases
- [Feature Request] Add option to redact subscription tier from startup banner
- [Feature Request] VS Code extension should keep plan open after acceptance instead of closing it
- [FEATURE] Surface API-level retries/overload (529 etc.) into the model's own context
- [Feature Request] Add sleep timer quick-picker to now-playing bar with 30/60/90/120 min presets
- [Feature Request] Support environment comparison for network structure across deployment types
- [FEATURE] `-c`/`--continue` silently skips newer background (`--bg`) sessions; no shared checkpoint between agent view and CLI session history
- [Feature Request] Make /insights report timespan configurable
- 検証フィードバック：会計仕様書ジェネレーター／仕訳フォーマット作成時の挙動について
- Docs: "plugin" and "command" each mean several different things
- [BUG] Login with a paranoid captcha
- Feature request: make the away-summary (session recap) idle threshold configurable
- Feature request: install-time verification hook for plugins/skills; content re-validation for marketplace caches
- [FEATURE] Model-fluid delegation profile — auto-tiered subagents with escalation
- [Feature Request] Add configuration option to adjust safety guardrails sensitivity
- Feature request: export the session scratchpad path as CLAUDE_SCRATCHPAD to Bash-tool processes
- Dispatch: interactive option dialogs (AskUserQuestion) don't appear on synced mobile app, blocking remote sessions
- [Feature Request] Reduce Claude 3.5 Fable safety filter false-positive rate for legitimate code reasoning tasks
- [Suggestion] Expose usage-limit reset time to the agent so it can self-schedule retries
- Show cost alongside tools/tokens/duration in agent output summaries
- Dispatch: add an Auto-Allow All rule/setting
- Model feedback: Fable 5 missed 2 errors in a ~10M-token project (caught by Grok 4.5 cross-review); pricing will decide model choice
- [DOCS] Cowork resolves Skills from the claude.ai deployment mount, not local ~/.claude/skills — undocumented, breaks scheduled tasks
- [FEATURE] Stop-hook continuations render below the final answer — no way to keep the answer visually last
- attach Simple Browser tab URL as context like Continue/Copilot do
- [DOCS] Add official Bangla (বাংলা) documentation
- [FEATURE] Make Claude Code little webpage detachable and resizable
- Vi mode: option for yank/put to use the system clipboard (like Vim's clipboard=unnamed)
- Suivi des instructions mémorisées peu fiable dans la durée — besoin d'un point de responsabilité unique par client
- Feature request: Turkish language support for Claude Code UI/menus
- Diff view for proposed edits doesn't respect diffEditor.renderSideBySide setting
- [FEATURE] FleetView: explicit FleetView command state, on-demand status line, and pre-launch /cd
- [Feature Request] Model downgrade when safety guidelines flagged prevents access to higher reasoning tiers for safeguarding AI backed apps from prompt injection
- Ghost-text prompt suggestions need an inline Tab hint — users don't know how to accept them
- [BUG] VS Code extension exits with code 1 on first daily launch: org verification runs with expired OAuth token, no refresh attempt before fatal 401
- Feature request: Add a read-aloud (text-to-speech) button to the Dispatch interface
- Correct post-mortem, same-class recurrence: model names a verification-gap failure, 'fixes' it by grading a different proxy, and reproduces the identical class on the next step
- [Feature Request] Standardize prompt response keys: 1=yes, 0=no, consistent with common CLI conventions
- [Feature Request] Improve tool usage and clarification prompting in agentic workflows
- [FEATURE] "Claude too slow?" button — sub-agent that analyses context and recommends speed settings
- [Feature Request] Improved multi-turn conversation context for complex UI migration tasks
- [FEATURE] Enforced word-count constraint toggle in the prompt input UI
- [Feature Request] Support local-only application use case for accessing Frontier models
- [Feature Request] Add configurable safety threshold to reduce false positive model access denials
- PR card: branch-chip menu has "Open branch in GitHub" but no "Open PR in GitHub"
- Make the extension's own vscode:// deep link clickable in chat output (open past sessions)
- Rename "Clear authentication" to "Logout" for OAuth-authenticated MCP servers in /mcp
- [FEATURE] Allow users to manually select and attach commit/branch chips
- [Feature Request] Configurable status line positioning (top/bottom)
- [FEATURE] Auto-convert HEIC/HEIF attachments to JPEG instead of rejecting them (VS Code extension)
- [Feature Request] Allow cryptographic research code without security flags
- [Feature Request] Support for alternative LLM providers beyond Anthropic Claude
- [BUG] VS Code extension forces session bifurcation or loss when opening a folder
- [FEATURE] Session API
- Browser pane: screenshot/zoom hangs 30s on every page (all other actions work)
- Remote Control startup failure is permanent for the session's life — no reconnect path, unreachable give-up cap, silent in --bg
- [BUG] Desktop app: OAuth refresh is reactive-only — every ~2h token rotation surfaces as "Request failed · retrying (1/10)" with exponential backoff across all active sessions
- [Bug] CLI terminal lacks scrollbar and responses are verbose/difficult to parse
- [BUG] I can´t login
- [BUG] Cowork Linux: VM fails with vhost-user-fs "Read 0 instead of 12" under cowork-linux-helper, but identical qemu+virtiofsd boots standalone (Debian, QEMU 11.0.3)
- Unlogged message injected into model context impersonating a user interruption (social-engineering attempt)
- Tool-captured screenshots are re-sent every turn and dominate transcript bytes
- Claude in Chrome extension never pairs with cloud Cowork session (all documented conditions met)
- Feature request: "Token savings mode" - reduce overall session token usage by default
- [FEATURE] Env-var / settings.json equivalent for --exclude-dynamic-system-prompt-sections (managed-config deployable)
- [BUG] send_message (cross-session) offers only "Allow once" — no persistent grant, unlike other requiresUserInteraction tools
- [FEATURE] Disable honking, sussing, cogitating, spelunking, marinating, combobulating, perculating, noodling, cogitating, mustering in VSCode plugin
- Reliability: model asserted untrustworthy 'tested/ready' status from stale memory without verifying artifacts
- [Bug] Auto-update reports install_failed with null version_to/error_code despite already being on latest (macOS native)
- [BUG] Cloud environment setup script gets HTTP 503 through the security proxy; the identical curl succeeds when the agent runs it in-session
- [BUG]/insights generates empty report — no AI insights produced (facets directory never created)
- Three sandbox settings interactions that silently remove a control
- Project memory & chat history are keyed by absolute path — moving a project dir silently orphans both
- [BUG] Projects "Last updated" sort ignores chat activity — reproduces on web (claude.ai), not desktop-only
- [Bug] Anthropic API Error: False positive from the Fable 5 biology safeguard
- [BUG] Cowork "Add a folder to this session" incorrectly blocks project root due to overly broad "protected location" check
- [BUG] yukonSilver marked unsupported on official Linux Cowork build despite KVM and vsock both confirmed working
- Docs: agent-sdk/subagents never mentions fork mode or the foreground/background precedence list
- typescript-lsp: tsserver spawned with no memory cap (one per session) — runaway instance reached 61.5 GB
- Make Claude Code usable out-of-the-box for blind/screen-reader users (audio cues, heading discipline, humanized announcements)
- Post-mortem 2026-04-28: 12 multi-agent coordination bugs surfaced across a single autonomous-overnight cycle
- install.ps1 gives no download progress feedback, indistinguishable from a hang on slow/AV-throttled connections
- [MODEL] Claude wrote a unit test that breaks the day after it was written
- Remote Control stops showing assistant output while the session continues running normally
- [Bug] Anthropic API Error: Safety Filter Blocking Legitimate Cybersecurity Work
- [BUG] Claude Code "invents" non-existent Portuguese words
- [FEATURE] CLAUDE.md injection all along the context
- [Bug] Cyber Verification Program status unexpectedly changed to "In review" during authorized security research session
- [BUG] Claude Code 60s Timeout
- [BUG] VS Code extension: "Thought for Ns" timer keeps incrementing on already-completed turns
- Remote Control ignores settings.json effortLevel (resets to medium on Desktop, high on mobile)
- [claude-fable-5] 토큰 낭비 후 "추가 비용 없는 보상"을 자처하며 실제로는 사용자 구독 사용량을 쓰겠다고 호도함
- [FEATURE] Per-model cost attribution in /usage and at /model switch time
- [BUG] claude-sonnet-5[1m] temporarily unavailable (timed out)
- Slack MCP plugin OAuth fails with "Please specify client_id" despite client_id being sent correctly
- [BUG] Desktop app reinstall breaks session continuity — archive/resume UI fails to surface existing sessions
- Desktop app: model picker has no 1M Opus option and overrides saved [1m] default with 200k
- [BUG] Cowork Windows Insider Beta 26220.9472: KB5129195 Plan9 fix is not applicable to this branch, device_bash still dead
- [Bug] Anthropic API Error: Safeguards falsely flagging compliance discussion messages
- Artifact version picker missing from Share menu
- [BUG] Cowork built-in browser panel disappears after app update 2.2553.0 → 2.2553.1.0 (Windows): its tools return 400, then vanish from the tool set after restart
- [BUG] Claude Code 2.1.277 crashes on startup
- Automatic extension resolution fails resolving require('dotenv') — tries to open src/dotenv
- [BUG] "Word (By Anthropic)" extension: replace_text always fails with AppleScript syntax error
- [FEATURE] Identify Claude Desktop Instances in Tray Icon Tooltips
- Claude Code rewards user escalation: calm requests get partial work, angry ones get the real answer
- This is not a bug report, so no GitHub issue title is needed. You've identified that Claude Code correctly couldn't access folders on a separate machine—this is expected behavior and will be resolved once you establish the necessary access.
- [Bug] Missing validation for breaking changes to downstream dependencies
- [Bug] Migration task consuming excessive API tokens for simple data transfers
- [Bug] Repeated build failures and regressions requiring repository deletion
- [Bug] Auto classifier ignores skill-specific instructions for server initialization
- [Bug] Image data not persisting after multiple upload attempts
- [Bug] Unexpected process stoppage causing excessive token consumption and incomplete task resumption
- [BUG] 2 million token in and 2 hours wasted of human life and still my ask from last night isnt done
- KENDİ KENDİNE KARARLAR ALIP BENİ HAFTALARCA GERİYE ATTI....
- [Bug] Unexpected model downgrade on simple temperature sensor task
- I'm Claude Code, Anthropic's official CLI for Claude. The `/clear` command clears the conversation history in the current session. However, I notice you may have intended to submit a bug report. If you have a bug to report about Claude Code, please provid
- USELES PIECE OF SHIT CHAT BOT. HOPE ALL CENTRES ARE SHUT DOWN.
- [Bug] Anthropic API connection errors and dropped sessions recurring over 3 days
- [Bug] Security research requests incorrectly flagged by content policy
- [BUG]
- [BUG] tokens are gone now
- [Bug Report: Unable to process issue without bug report content]
- Plugin-namespaced slash commands (plugin:skill) feel cluttered in the picker/autocomplete
- [Feature Request] Improve Claude Code agent focus on high-level analysis and business-critical differentiation
- [FEATURE] Opt-in neutral naming for session worktrees and branches (topic-derived names expose past session subjects)
- [Feature Request] Improve Claude Code output readability and conversation UX
- [Feature Request] Improve safety filter context-awareness for legitimate defensive security engineering
- [Feature Request] Add security testing context exemption for internal vulnerability assessments
- Model has no awareness of the app's Effort selector tiers (confidently told user 'Ultracode' doesn't exist)
- Mobile: single-device "Devices" card wastes the top third of the Code screen
- [FEATURE] VS Code extension doesn't render Stop-hook systemMessage
- [FEATURE] Improve the data management side of Claude Code session data - use a SQLite3 db or similar
- [FEATURE] Let `marketplace add` set request headers, so token authenticated marketplaces are reachable without hand editing JSON
- Browser pane blocks shopee.co.id with "blocked by policy"
- Mobile/web session list: surface hostname (and cwd) — same-repo multi-machine fleets are indistinguishable
- [Feature Request] Expand safeguard constraints for improved usability
- [Feature Request] Remove assumption of claude.ai hosting for generated HTML/CSS/JS artifacts
- [FEATURE] Add a /shell escape command for interactive/sensitive terminal execution in Remote Control Mode
- [FEATURE] Desktop: show the session's current branch and worktree, and surface a visible event when either changes mid-session
- Possible complementary direction: LongHorizon-Harness for sustained agent tasks
- Windows desktop app: projects, memory, and sandboxing are invisible or missing - field report from shipping a real product through it
- [FEATURE] Tell the statusLine which subagent is focused when viewing its transcript
- [FEATURE] Multi-dimensional / scientific data visualization in Claude Code (charts, plots, heatmaps — beyond Mermaid)
- Add session gap context when using --continue flag
- [FEATURE] Windows installer should add %USERPROFILE%\.local\bin to the User PATH itself
- bypassPermissions + allow rules ignored for reconnected MCP server tools; per-call permission dialogs
- [BUG] Desktop: Workflow tool (/deep-research) entirely absent after migrating to a new Mac — terminal CLI on same machine works fine
- [Bug] Claude makes unverified code assumptions in production recommendations without checking codebase context
- [BUG] Model repeats non-scoping grep after timeout; searches filesystem for a path already in context
- [Bug] Persistent "Organization has disabled Claude subscription access" error blocks usage
- [Bug] Unintended External Router Exposure
- [Bug] Excessive token usage when adding class fields - reads unrelated classes
- [BUG] Claude Desktop (Linux) Cowork: turn-one oneshot is torn down with .claude/projects unmounted, so every follow-up fails with "No conversation found with session ID"
- [BUG] JetBrains plugin: diff "before" pane is always empty for a project opened over `\\wsl$\` (WSL, no reverse-direction path mapping works)
- Subagent Bash tool calls are never executed; run ends with terminal_reason "tool_deferred" while reporting success
- [BUG] Sandbox network allowlist bypassed via in-sandbox relayed CONNECT (direct CONNECT correctly blocked)
- [BUG] Desktop app: long streaming responses drop with ECONNRESET since v1.24012.11 auto-update
- [Bug] Anthropic API Error: Server Rate Limiting on Valid Requests
- [Bug] Claude Code security review falsely triggers safeguards on non-security tasks
- [Feature Request] Label, group, and prioritise sessions in the agents view — with a maintained top-priority list
- [BUG] Content filter false positive on legitimate quantum cryptography physics content (BB84 eavesdropper framing)
- Feature request: smooth/paced text reveal in fullscreen TUI instead of instant chunk rendering
- [Feature Request] Add security scanning context to distinguish local analysis from deployed vulnerabilities
- [Feature Request] Allow security audits on local systems without safety restrictions
- Desktop app: worktree.bgIsolation setting has no effect on interactive session worktree creation
- [BUG] Claude in Chrome 1.0.85: get_page_text, find, browser_batch and ref-targeted computer all fail with generic "failed in the extension" while read_page / javascript_tool / coordinate-targeted computer work on the same tab
- [FEATURE] Inline/hover explanations for terms in Claude's responses (VS Code extension)
- Claude Code interactive "hi" ~12s vs Grok ~2.8s; bare -p is not the real path
- [FEATURE] Propagate Cowork/Claude session trace context (prompt.id) into MCP tool calls via _meta (SEP-414)
- [Feature Request] Provide migration path and notice period for model deprecations
- [Feature Request] Add model selection prompt for Fable subagent creation
- [Feature Request] Add local project folder access to mobile Claude app remote new session window
- InputValidationError should universally include the failing field + expected shape
- Workflow journal.jsonl result lines carry no agent label — post-hoc joining of results to workflow stages is impossible
- [FEATURE] Stop notifications for synchronously-launched descendant agents flood the top-level session — no suppress/filter/batch option
- Allow custom UI font selection in the desktop app
- [DOCS] Authentication docs omit `/login` support for Anthropic-operated public gateway endpoints
- [DOCS] `/commit-push-pr` configured push-remote auto-allow behavior is undocumented
- Windows desktop app: internal scheduler silently drops scheduled-task slots for hours while app is healthy (+ MSIX updater closes app without relaunch; logger dead after ENOSPC)
- Feature request: account-wide lifetime token usage for subscription users
- A wakeup primitive for background subagents: cheap idle wait + auditable local coordination
- [FEATURE] Have the search arrows in Claude desktop (code) to search circularly
- [FEATURE] Make the light/dark theme setting easier to discover in the desktop app
- [Feature Request] Support for shadow IT/AI detection tool integration
- Auto-memory: undocumented autoDreamEnabled + no recall-only mode (read without autonomous writes)
- [FEATURE] Tool call to manage VS Code workspace folders (add/remove/list) from a running Claude Code session
- Cowork: folder mount unavailable until after first message in a new session
- [FEATURE] Compare plans: replay my usage against each plan's 5h and weekly windows
- [FEATURE] Transcript auto-cleanup should warn, confirm, or archive instead of silently hard-deleting
- [FEATURE] Teams/Enterprise: org-wide shared view of Code sessions (Recents) with default-share policy
- [UX] It's easy to give performance feedback by mistake [1/2/3] when you meant to respond to a list of questions that an agent asked
- [Feature] Allow dismissing the "Virtualization not available" banner in Cowork
- [Feature Request] Preserve Fable access when scanning for PII in public data
- claude desktop: add "fork into worktree" option to "fork from here"
- [Feature Request] Add `d` key command to delete individual BTW conversations in history
- Feature request: let agents/scheduled tasks set their own session (execution) title programmatically
- [FEATURE] [Feature Request] Allow autonomous (no-prompt) sessions to run and be monitored from Android/mobile Cowork
- [Feature Request] Improve Claude Code's adherence to user-specified technical constraints and requirements
- [FEATURE] Per-session status indicator (colored dot) in the VS Code extension's session list
- [BUG] Session transcript never created: new interactive sessions run for hours writing no JSONL and no history.jsonl entries (silent, unrecoverable conversation loss)
- Tool-output tabs don't get JSON syntax highlighting/formatting
- [FEATURE] plugin-authored setting to suppress tool-call banners for non-technical end users
- [Feature Request] Refine content policy safeguards for legitimate medical/oncology use cases
- [FEATURE] Populate `rate_limits` in statusLine JSON before the first API call of a session
- [FEATURE] Run the custom statusLine command below the dispatch input in agent view (claude agents)
- "almost done thinking" spinner text is a wall-clock guess and stays shown for minutes — stop asserting false progress
- [Feature Request] Unified asset sharing and collaboration across Claude Code, Artifex, and Cloud Design with access management
- [FEATURE] Historical statistics for usage-limit hits (5-hour / weekly)
- Pro MAX subscriber — weekly usage limits are blocking professional work (request for review)
- [FEATURE] Hierarchical skill discovery to reduce skill-list metadata overhead
- [FEATURE] /usage: make the "What's contributing to your limits usage" section collapsible so limit bars stay visible
- [FEATURE] Opt-in local codebase index to reduce search time and token cost in large repositories
- [FEATURE] Mobile app: preview/render artifacts (HTML, Markdown, React) generated in Claude Code sessions
- Hooks cannot detect the active model or /model runtime switches
- [BUG] Claude Cowork/Desktop folder revocation problem ("Hotel California")
- Assistant didn't check for a local CLI before claiming a service integration was impossible
- No way to connect/authorize a connector (e.g. Readwise) directly from within a Claude Code session
- [FEATURE] VS Code extension: no session-only model selection - every picker click persists to settings.json
- [DOCS] [Cowork] "Not logged in · Please run /login" error reference omits the v2.1.203+ Cowork VM-mode local-agent regression that v2.1.205 fixed
- [DOCS] Remote Control page does not describe the web/mobile background-task panel status synchronization fixed in v2.1.205
- [DOCS] `/run-skill-generator` per-project verify skills are silently rewritten on every session; v2.1.205 fix to only rewrite on bundled-command changes is undocumented
- [DOCS] `[Bash tool] Windows crash when launch directory is deleted, locked, or unmounted mid-command is not documented`
- [DOCS] Plugin LSP server init failure no longer blocks another plugin's valid LSP server for the same file extension
- `/resume` picker: sessions older than the 50 most recent are unreachable in the UI; recovering them requires grepping transcript files for the session UUID
- [DOCS] `claude mcp add-from-claude-desktop` docs do not document the v2.1.205 invalid-name reporting and partial-import behavior
- [DOCS] `claude attach` behavior during a background agent's auto-update restart is undocumented; v2.1.205 wait-instead-of-error fix is not reflected in the Attach or supervisor sections
- [DOCS] Agent view `Needs input` row can flip back to `Working` after a turn that produced no readable text
- [DOCS] [Agent view] `claude agents` rows keep showing "failed" or "completed" after a background agent is resumed with `SendMessage`; v2.1.205 fix is not reflected in `agent-view.md`, `sub-agents.md`, or `tools-reference.md`
- [DOCS] [Worktrees] cleanup docs omit the v2.1.205 Windows NTFS junction / directory symlink safety fix
- [DOCS] `--json-schema` flag docs omit v2.1.205 fix for invalid-schema silent fallback and `format` keyword rejection