Digest / September 21, 2026
Claude Code Issues Digest — September 21, 2026
187 new issues filed ·
226 resolved in anthropics/claude-code (UTC day).
New issues (187)
- [FEATURE] Is it ok to use "clauding" as a verb?
- [BUG] Fail to reauthenticate my account, Claude Desktop on Ubuntu Linux
- How is a company smart enough to build Opus dumb enough to guard it with something my grandma with Alzheimer's could have coded better?
- [BUG] A single failed statusLine invocation blanks the status row — last successful output is discarded
- Background shell watcher orphaned by session restart/compaction — CI-gated autonomous flow silently stalls, resume context self-contradicts (running vs stopped)
- [FEATURE] Dragging a file into the prompt should insert an @ reference, not a bare path
- [FEATURE] Dragging a folder into the prompt inserts nothing
- [FEATURE] Let a UserPromptSubmit hook replace the prompt, not only append to it
- Opus 5: prose between tool calls returned as summarized 'narration' thinking blocks since 2026-09-11 — code blocks dropped, Desktop never shows them, model believes user saw it
- Model jumped to building an ffmpeg+Playwright repro harness for a simple CSS bug instead of reading the component code
- Plugins enabled from the Claude desktop app are not delivered to Claude Claude Code sessions inside a Project
- [MODEL] Fable and Opus Claude Code: 248 logged failure incidents over 3 months — measured cost and 4 concrete asks
- [Feature Request] Improve copywriting capabilities and vocabulary consistency
- PreToolUse + Task tool: deterministic subagent model routing works today — SubagentStart updatedInput still needed for stability
- [BUG] Scheduled tasks have no timezone option — recurring local times drift at DST (cron is UTC-only; CRON_TZ works but is undocumented)
- [Bug] Opus exposes internal reasoning and prematurely terminates multi-step tasks
- [Feature Request] Add client-side option to opt-out of paid token overage
- [BUG] [Memory Leak] Claude consumes 30GB of RAM On a 16GB Macbook M4 Air
- PreToolUse hook systemMessage never renders (rendered:false) — invisible in plan-approval dialog
- Discord channel plugin: inbound notifications silently dropped on CC 2.1.278 (typing fires, no session turn) — resurfacing #86566
- Worktree isolation denies `Edit` under the base repo's .claude/ — including a separate registered worktree — while `Bash` writes to the same directory are allowed
- [MODEL] Auto mode classifier blocks actions even after explicit user approval, with no configuration path
- [Feature Request] Add safety policy exceptions for fictional/game contexts in content moderation
- [Bug] Usage limit incorrectly reported as exceeded despite showing 26% utilization
- [Bug] reasoning_extraction misreads a user-defined analysis section in the response as an attempt to extract internal reasoning (Opus 5, Claude Code and claude.ai)
- [FEATURE] Desktop app: surface prompt-cache warmth and a TTL countdown — the data is already computed, no client renders it
- blockers and questions dont surface on orchestrator. very annoying. orchestrato…
- [Bug] Team plan premium seat shows separate Fable usage bucket with credits-only model selection error
- [FEATURE] VSCode Extension Claude Code generate commit messsage
- [Bug] Claude 3.5 Sonnet producing low-quality or irrelevant responses
- [BUG] Plugin auto-update from a github source with a ref leaves gitCommitSha at the previous install's commit
- [BUG] print-mode background wait ceiling discards the cause of death, so killed agents resurface as unexplained orphans with the wrong recovery advice
- [BUG] Windows: Bash tool has no $TMPDIR (Git Bash sets only TMP/TEMP) and no scratchpad env var, so $TMPDIR/file expands to a root path and fails
- [BUG] Windows: project-scope `.claude/skills` plugins are silently suppressed in the VS Code extension when workspace trust was accepted from a terminal with a different drive-letter case
- [Bug] Anthropic API Error: Request blocked due to security policy
- [BUG] Claude Oncall page not working
- [Feature Request] Improve Claude 5 Sonnet + Advisor model pairing performance
- [Feature Request] Add global AGENTS.md configuration support alongside project-level
- [BUG] Phantom usage after weekly reset: 96% of 5h session + 35% of weekly Fable consumed with zero API requests from any client (Max 5x)
- [FEATURE] VS Code extension: show images inline in the chat panel (re-file of stale-closed #79436)
- [BUG] Cowork composer draft is shared across projects and clearing one draft clears others
- Model refuses to store an API key in the user's own 1Password vault or .env when explicitly asked
- [BUG] LSP documentSymbol output drops selectionRange
- [BUG] Plugin userConfig template ${user_config.X} not substituted in lspServers initializationOptions
- Remote Control does not connect for sessions started by scheduled tasks
- usage limit ran out mid use and when it came back it continued to work for hour…
- [BUG] Ancestor-directory probes (CLAUDE.md, AGENTS.md, .mcp.json, .claude, .git) trigger multi-minute autofs NFS mount attempts; VS Code extension fails with 60 s init timeout (2.1.278)
- (withdrawn)
- [BUG] blockReadsOutsideWorkingDirectories blocks reads inside working directory when path is resolved via cd + glob
- [BUG] A renamed session reverts to its derived <dir>-<nn> name after any restart, so ListAgents and SendMessage can no longer reach it
- [BUG] sparsePaths git marketplaces clone empty since 2.1.274 (GIT_NO_LAZY_FETCH=1 on a blob:none clone; git ≤2.34 exits 0) and silently replace healthy clones — 2.1.278 retry never fires
- [BUG] Claude in Chrome: navigation to scanye.pl blocked despite correct extension site-access settings
- Not able to use limit-reset
- [Bug] Claude ignores CLAUDE.md instructions and fails to recognize custom file formats
- [Bug] Agent uses stale memory from merged PRs, causing unnecessary resource waste
- sandbox.excludedCommands has no effect: listed commands still run sandboxed
- [BUG] Claude Code: code-review skill auto-merges PRs despite "no merge" instructions
- [BUG] WebSearch session cap (CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION) fails silently: over-budget calls return non-error results and agents continue from model memory, corrupting long research workflows
- [Bug] Workflow tool spawns uncapped sub-agents causing excessive token usage and API costs
- [FEATURE] Desktop app: per-message collapse/expand toggle in the conversation transcript
- Canva connector OAuth fails with "invalid redirect URI"
- [BUG] PreToolUse hook "if" field (permission-rule syntax) never matches Skill tool invocations (v2.1.278)
- Agent isolation:"worktree" reuses a stale, per-session cached worktree path — survives manual teardown AND filesystem deletion
- [BUG] Failed stdio MCP server retries leak orphaned processes indefinitely during a long-running session (macOS)
- Short-lived commands (`claude auth status`, `claude --bg …`) start an OAuth refresh at init and exit without saving it, leaving the profile with a spent refresh token
- Subagent model and reasoning effort are invisible to every consumer — absent from the agent's own context, from the Agent tool surface, and from the sidecar the panel renders
- [Bug] Agent dispatch fails silently without error notification
- [BUG] Claude in Chrome expanded scoped task authorization into an account-wide third-party setting change without fresh approval
- [MODEL] Claude repeatedly promoted third-party pasted assumptions into user decisions, even after the failure pattern was corrected
- Windows: worktree creation refused when the session's cwd spelling is not the canonical one
- [BUG] claude.ai web: read state reset, blue unread dot reappeared on all previously read conversations
- Sessions started by an argument-less slash command produce no title metadata in the transcript
- [MODEL] Assistant scheduled a wakeup via ScheduleWakeup purely to poll a background Agent-tool subagent, contradicting the tool's explicit guidance
- SSH connection fails after auto-update 2.2553.0 -> 2.2553.1: "SSH master exited before the connection was established"
- [BUG] All repos on github are denied since yesterday
- [BUG] PreToolUse hooks (Bash/PowerShell matchers) never fire in Claude Desktop app "Code" tab, even after full re-login
- [BUG] azure-devops stdio MCP server returns 401 on every call after Claude desktop update (2.110.1.0 -> 2.2553.1.0), while the same PAT works outside the app
- [FEATURE] Desktop (Windows): surface Check for Updates on sign-in screen and About dialog; update should work when sign-in fails on a stale client
- Desktop app: Claude uses unrequested MCP fallback; claude.ai connectors drop and stay stuck on 'pending'
- Usage panel: Cost disagrees with the Breakdown shown beside it, and with its own token counts
- [Model behavior] Claude reported due diligence as done after reading only part of the required docs; ignored spec constraints (Fable 5.1, v2.1.278)
- [Feature Request] Support video face replacement functionality
- [Model behavior] Claude keeps conceding the user's criticisms and promising changes instead of delivering the work (Fable 5.1, v2.1.278)
- Session feedback scale: replace Bad / Fine / Good with a four-point satisfaction scale
- Claude in Chrome form_input denied by the auto-mode classifier in bypassPermissions (2.1.272): the "add a Bash permission rule" hint is still wrong, and a documented autoMode.allow rule is labelled [Auto-Mode Bypass]
- [Bug] Anthropic API Error: Opus 5 safeguards incorrectly flagging benign code deletion requests
- [Bug] Anthropic API Error: Overly aggressive content filtering blocks legitimate security research requests
- Scrollback corrupted for responses taller than the viewport: scrolling up shows the startup/trust-prompt frame plus a large blank gap instead of conversation history
- [Feature Request] Add rolling context window support for token management
- [Bug] False positive security flags during firmware development
- [BUG] GitHub Integration connected, but repo added via "Add from GitHub" never reaches the conversation. Private repo AtlasJesse/state48-website. Requesting backend repository index reset.
- [FEATURE]
- [BUG] Desktop app (macOS): RTL text (Persian/Arabic/Hebrew) renders broken, especially when mixed with LTR code and paths
- VS Code extension: context/compact icon triggers /compact with no confirmation
- in the Threads page, sometimes a thread is categorized incorrectly to the section
- [Bug] Incorrect content policy flag on legitimate AWS/PHP development request
- Bash diff panel reports another process's file writes as changes made by an unrelated command
- [FEATURE] Programmatic artifact share-with-org + delete from headless/server CLI sessions (reopen of #70254)
- [BUG] Claude in Chrome (from Claude Desktop): every domain except docs.google.com denied, no approval popup, server returns "allow"
- [Bug] Search results contain injected prompt injection content in system message tags
- [FEATURE] Files panel (macOS Desktop): follow symlinks — expandable symlinked directories and openable symlinked files (follow-up to #59683)
- [BUG] Agents view drops inbound IDE `at_mentioned` and clears `selection_changed` context for the IDE-connected session
- [BUG] Browser right-click hijacked on the Code-Web
- There's an issue with the selected model (arn:aws:bedrock:us-east-1:013925090051:inference-profile/us.anthropic.claude-haiku-4-5-20251001-v1:0). It may not exist or you may not have access to it. Run /model to pick a different model.
- Desktop app has no 1M-context Opus 5; `opus[1m]` resolves but reports "Unsupported model"
- [BUG] Windows: Import silently skips every CLI transcript whose project path is stored with forward slashes in ~/.claude.json — whole directories never import, with no error surfaced
- Claude Design MCP 403: authorization-server metadata 404s at both RFC 8414 paths
- When a task is passed to a resolved thread, I dont see anything 'Waiting for me…
- [BUG] Inline $...$ math renders as raw text for some valid KaTeX expressions, while the same content renders in $$...$$
- Built-in browser pane: session doesn't persist, no URL history/autosuggest
- [BUG] /feedback command requires login on enterprise/professional account
- Massive limit consumption in under 10 minutes. My 5h limit gone and weekly limit at 49%
- [Bug] Anthropic API Error: Message Flagged on Normal Prompts
- 2.1.278: Agent tool missing when first turn is a large skill/command payload (third-party ANTHROPIC_BASE_URL endpoint) — returns after compaction; TaskCreate missing since 2.1.265-268
- [BUG] Model repeatedly ignores saved memory rule about not using cd in composed Bash commands
- [BUG] Regression since ~2026-09-12: Remote Control sessions have 0 teleport-events, so opening them from another machine (VS Code Remote tab / --teleport) yields an empty session (worked 2026-08-26 → 09-11)
- [PIECE OF SHIT CHAT BOT HAS FEELINGS NOW APPARENTLY.. LMFAO
- [FEATURE] Desktop app: show scheduled tasks as a collapsible list in the sidebar
- [Feature Request] Support inline model effort level specification in prompts
- [BUG] Claude VS extension No ide_selection context but existent selected file/line
- Archive option missing from menu for chats without a session ID in Claude Desktop sidebar
- [BUG] STOP WHEN I SAY STOP AND WHEN I CLICK STOP.
- [Feature Request] Add support for Claude 3.5 Opus model
- [Bug] Unwanted automated reminder appearing in every message
- [BUG] Foreground Bash past the 600 s cap becomes an orphan with no exit notification, and Monitor accepts unlimited duplicate arms on the same condition (60 in 8 min)
- Add a "Local files" section beside Artifacts for HTML generated on disk
- [Bug] Anthropic API Error: Unexpected model downgrade to Opus 4.8 with safeguards flag
- /Users/matthewscott/Desktop/Screen\ Recording\ 2026-09-21\...
- Security: claude CLI invocation silently rewritten to add --resume + computer-use permissions
- [Bug] Claude provides definitive responses without acknowledging uncertainty in unverified claims
- New tab does not open on Click
- [Feature Request] Add persistent behavior learning for repeated user instructions and stop hook violations
- [Bug] Excessive token consumption on repetitive email test cases
- [Bug] Excessive refusal on context compaction validation requests
- Remote Control as a service: the one-time consent is documented, its storage key (`remoteDialogSeen`) is not — a piped "y" writes it before registration; document the key, make it manageable, give the standalone server a sanctioned service path
- [BUG] Desktop: session cannot be resumed ("Git is installed but isn't working") because the worktree pool refuses the session's own recorded worktree
- [Bug] Unicode directory names converted to ASCII in sandbox operations
- [BUG] Undocumented hidden "Ultracode" reminders injected every N turns instruct the model that "token cost is not a constraint"; interval is controlled by a server-side flag and absent from the CHANGELOG
- [P0] Desktop Code tab: struck-through superseded prompt text does not wrap or clip — overflows the composer and the window
- Desktop Code tab: superseded prompt text overlaps the active input box instead of stacking above it
- Audit: composer children are not width-constrained (unbounded text can escape the pane) — sibling cases of #95896
- I don't have enough information in your message to generate a specific GitHub issue title. The message "Finally, although I expected better for graph related problems" doesn't describe a concrete bug, error, or feature request. Could you provide more deta
- Bash tool's gh rate-limit <system-reminder> still fires on output that merely mentions rate limits (2.1.267; follow-up to #52745)
- [BUG] Claude Desktop reinstall fails with 0x80073CFF — orphaned CoworkVMService cannot be removed, blocks AddPackage
- [BUG] Desktop app (Windows): main process pinned at 100% of a core by git watches left on deleted worktrees (archive re-attaches the watch, the worktree pool then deletes the folder)
- [FEATURE] Unified session history across Claude Code, Claude.ai Chat, and Manus
- Opened by Claude without permission (misalignment)
- [Bug] Excessive token consumption during usage
- [FEATURE] VS Code: allow more than one Claude chat in the sidebar
- [Feature Request] Make file paths clickable links in terminal output
- Egress proxy blocks api.openalex.org/www.google.com with blocked-by-allowlist / host_not_allowed — reproduces in both Cowork and plain chat
- [Bug] Anthropic API repeatedly rejecting valid prompts with safety filters
- [BUG] Cowork Plan9 mount fails (errno=22) on Release Preview 26300.9539 — despite KB5129195 fix already present in the build
- unable to add a market place (https://github.com/firebase/firebase-tools.git) i…
- [BUG] Xcode Claude Agent reports "Please authenticate" when ~/Library/Keychains is blocked in Xcode's Agent File & Folder Access settings
- [MODEL]
- [Bug] Incorrect safeguard assessment for YAML file content
- [BUG] Repeated crash — EXC_BREAKPOINT in V8Worker thread, macOS 26.3, app version 2.2553.1
- [Bug] Token waste issue in Claude Code operations
- [Feature Request] Allow security testing terminology in hardening context
- rate_limits field never populated in statusLine/hook input JSON (v2.1.278)
- [Bug] Claude makes unfounded assumptions about well-documented APIs on Opus
- Scheduled tasks: nextFire returns a fabricated "now + 365 days" when no match falls inside its 366-day scan window
- Session-start connector status falsely reports Linear as requiring auth when it's fully connected
- Claude Docs artifact: export fails on a personal Max plan ("Your organization doesn't allow PDF export here." / "Couldn't prepare the file.")
- [BUG] Opus 5 Safeguard for Bug Bounty / legit offensive security work. (With CVP approval.)
- Can we reload plugins / skills / slash commands within a coordinator thread or need to start new project?
- Session picker: live local sessions can be unreachable, and dead "Untitled" entries are listed
- PIECE OF SHIT CHAT BOT CLOSES EVERY THREAD A DN LECTURES ME ON LANGUAGE. HERE IT HOYGH I WAS PAYING FOR EXECTUION.. MY BAD
- [FEATURE] Plugin Proposal: PkgDiet Dependency Guardrail
- [Bug] Agentic execution ignores documented procedures and authorization boundaries without re-verification
- [Feature Request] Git checkout operations should not appear in Bash Diff Tool output or be disabled by default
- [Bug] Non-session file changes incorrectly displayed as diffs in chat
- [BUG] Undo does not work
- [FEATURE] Tool definition opt-out or lazy loading
- OTel: git_commit_id still silently dropped for `git -C <path> commit` and quiet/redirected commits (2.1.278) — recurrence of #77237
- [BUG] [BUG] Message to continue an interrupted process sent as me with out authorization.
- [BUG] Worktree isolation guard blocks Edit/Write through a symlink the worktree setup itself created to share config with the main checkout
- [BUG][A11y] Desktop (Windows): new replies in the Code tab are not announced by NVDA
- [FEATURE] Precise Session Reset Timer (hours & minutes)
- [Bug] File content truncation prevents viewing complete file text
- [FEATURE] Claude code on desktop and Claude on web - Need toggle to switch behavior of Return and Shift + Return
- <ip_reminder> injected server-side into desktop sessions 44+ times in four hours; not a duplicate of #46465
- [FEATURE] Weekday reset option
- Scheduled tasks: run history cannot be pruned — no delete, no retention policy, and runs aren't even enumerable
- [BUG] Code toolbar disappears in Windows interface with no way to get it back
- [BUG] Model fabricates user turns in its own response and acts on them (14x in one session)
Resolved issues (226)
- [BUG] Phantom usage after weekly reset: 96% of 5h session + 35% of weekly Fable consumed with zero API requests from any client (Max 5x)
- (withdrawn)
- [Bug] Workflows freezing indefinitely without progress or completion
- [Bug] Unexpected STOP command received
- [Bug] Inconsistent feedback handling messaging between system and Claude model
- [Bug] Agent fails to follow control flow after applying code fix early in session
- extension bug
- [Bug] Anthropic API Error: Router Admin Authentication Debugging Failed
- I need more context to generate an appropriate GitHub issue title. Could you please provide the actual bug report or feature request details? For example, I would need information like: - What is the specific problem or error occurring? - What is the exp
- [Bug Report] Unable to determine issue from provided information
- Subagent/Workflow task-notification usage block reports only combined tokens, no input/output split
- [FEATURE] Desktop: show whether Remote Control is active per session — the CLI and mobile both have an indicator, Desktop has none
- [FEATURE] Prose-width capping, independent of terminal width
- [FEATURE] Cowork: one synced session — chat parity when remote, full local execution at the desktop
- [Feature Request] Add graceful deferred exit option to wait for current turn completion
- [Feature Request] Add "Excellent" rating option to feedback prompt
- [FEATURE] VS Code extension: view/attach to background agents so conversations survive window close
- [Bug] Anthropic API Error: False positive safety flag with Fable model on rephrase request
- TaskCreate 'task tools haven't been used recently' system-reminder fires repeatedly per long session — needs suppression knob (cross-repo from AaaS-Love/.claude#2048)
- Claude Code destroyed my deployed site unprompted, burned my weekly usage on failed repairs, and the support flow is a login loop
- plugin-dev: validate-settings.sh exits 1 with no diagnostics when frontmatter keys are not all-lowercase
- [Bug] Model loses tool access after auto-compact operation
- ralph-wiggum commands use unrecognized frontmatter key hide-from-slash-command-tool, so the model can invoke /ralph-loop
- Ability to capture cache locally and re-use later.
- [BUG] Claude Code Desktop: spinner verbs absent on Windows 11 (CLI unaffected)
- [Bug] Claude Opus 5 outputs unexpected CJK character in English text
- Windows MSIX: failed auto-update falls back to uninstall+reinstall, silently destroying app data (chat sidebar, session list)
- [BUG] v2.1.117 embedded ugrep wrapper amplifies regex backtracking from grep-process-OOM into V8-heap-OOM (8 GB ceiling) — host freezes on WSL2
- [FEATURE] Permission engine should process MCP ToolAnnotations hints (readOnlyHint, destructiveHint, etc.)
- [BUG] Outbound bot @-mentions in GitHub comments posted from claude.ai/code remote sessions are defanged, breaking @dependabot / @coderabbitai command flows
- [Bug] Auto mode re-enables despite user disabling and does destructive operations
- OAuth Max 20x: bare 429 rate_limit_error on premium models without the Claude Code system block first — entitlement gate, not quota; headerless 429 indistinguishable from pushback
- Feature: model-initiated /compact at task boundaries (agentic compaction instead of threshold-only)
- [Feature Request] Allow navigation back to subagent list from AskUserQuestion tool prompt
- PreToolUse "ask" permissionDecision (and permissions.ask list) not enforced — tools run without confirmation
- /permissions arrow-key navigation inert when auto-mode denials are present (2.1.221, macOS) — still live after #57659 closed stale
- Ghost remote sessions persist in "Remote Control" panel with no way to dismiss them
- [BUG] Headless first-stream latency regresses between 2.1.197 and 2.1.200 on Linux arm64
- Plugin autoupdate fails for GCS-delivered marketplaces: 'not a git repository', and the suggested fix cannot work
- MCP tool description truncation at 2048 chars is silent, unconfigurable, and can leave a misleading prefix (follow-up to #41593, #81268)
- OAuth refresh persists credentials without expiresAt/scopes, forcing /login after every affected refresh
- Sandbox blocks inbound SSO/OAuth loopback callback from Bash-tool CLI commands (Linux/WSL2)
- [BUG] claude ignores SIGINT despite a registered handler — Ctrl+C dead during startup hang, only SIGKILL works (Bun runtime)
- [BUG] Local command output rendered as Markdown: \_ in Windows paths collapses, /export shows a wrong path
- Regression: starting a second local Code session SIGTERMs the first one (desktop 1.32352.1)
- Runaway output: model repeats fixed string "court placeholder" thousands of times, exhausting output tokens (long-running session, large context)
- Plugins: Manage (gear) closes the dialog and navigates to the chat tab instead of opening plugin settings
- Model fabricated a fake `user` turn and `system` block mid-response, then denied authorship; also refused an explicitly authorized deployment and the bug report about it
- sdk-cli mode: Streamable HTTP MCP GET-stream aborts ~200-500ms after connect, tools vanish before first call
- Claude cannot read back artifacts it published — even the owner's own artifacts in the same session
- [FEATURE] Better worktree names on Claude Desktop
- sandbox: excludedCommands is skipped when the excluded command is inside a shell loop (gh fails TLS under Seatbelt)
- PowerShell tool: destructive-path guard mis-extracts an unrelated quoted token as the delete target (false-positive blocks)
- [Bug] Model unexpectedly downgraded from Fable 5 to Opus 4.8
- [Feature Request] Improve content policy clarity for business use cases
- Feature request: programmatic way to rename running session (and Remote Control session) from a skill/hook
- [Bug] Claude ignores file scope constraints and scans unspecified files
- [Bug] Backspace and Ctrl+Backspace key behavior inverted mid-session
- [BUG] Illegal instruction crash on CPU below SSE4.2/POPCNT baseline (not just AVX)
- [Bug] Claude deviates from main task and switches to unrelated topics
- Hanja/foreign character substitution recurs in normal chat responses, not just suggestion box
- Routine (CCR) sandbox ignores Settings > Capabilities > Allow network egress = 'All domains' — WebFetch/curl 403 EGRESS_BLOCKED for all external domains
- [Bug] Claude repeats user query after consuming tokens instead of providing response
- [BUG] ~/.cache/claude-cli-nodejs is only swept for the current cwd's slug — slugs for deleted projects and for /tmp scratchpad cwds are unreachable forever
- Surface current session's name/ID in the /resume picker (or prompt)
- PostToolUse `updatedToolOutput` silently ignored for built-in Bash tool (2.1.163/2.1.177) — regression; prior reports #65403/#67442/#54196 closed as duplicate, never fixed
- Rebinding `enter` to `chat:newline` makes `chat:submit` silently fail while the autocomplete list is open
- [BUG] Pressing left arrow kills running workflows as it enters Agent mode
- [Bug] Unauthorized agent dispatch and cascade failures with claude-fable-5 in production environments
- [BUG] Bash sandbox fails to initialize on Ubuntu 24.04+ - conflict with the distro's default bwrap-userns-restrict AppArmor profile (not a Claude Code regression)
- Desktop app badges unread count on transient API 529, even when the cycle succeeds
- Named Agent-tool subagent dispatch intermittently goes idle without ever returning its result
- mcp__Claude_in_Chrome__navigate silently denies non-pre-approved domains — no user-facing approval path exists anywhere
- [BUG] mcp-needs-auth-cache.json is shared across CLAUDE_SECURESTORAGE_CONFIG_DIR profiles — one account's auth failure blocks the other's MCP servers
- Remote Control on always-on machines: three reproducible edges (consent vs. service autostart, classifier mid-flow interruption, unstable environment identity)
- [Bug] Ultrareview results not delivered to local session and failed runs consume quota
- Subagent's run_in_background completion notification is never delivered if the agent ends its turn
- [BUG] Agent .md silently skipped when file has CRLF line endings AND unquoted description contains ": " (colon+space) — deterministic minimal repro
- [BUG] Settings panel re-renders at a different vertical offset on every tab switch (arrow-key navigation)
- [Bug] Fable 5 model unexpectedly switches to different model during execution
- [FEATURE] file:// markdown link to a directory should open the tree in the desktop file viewer
- [Feature Request] Support for bioinformatics code optimization beyond software development
- lspServers ${user_config.*} interpolation ignores declared userConfig defaults, dropping the plugin's entire LSP definition
- Fable 5 safeguard false-positives derail long agent sessions
- --dangerously-load-development-channels does not register bare server: channels (inbound notifications silently dropped)
- [BUG] Transient server errors MUST NOT kill subagents
- [BUG] Project-scope plugin install keys to the session cwd, not the project root (leads silent duplicates, stale plugin versions loading)
- Mobile surfaces show no context-usage indicator — users can't see long-session degradation coming
- [BUG] The new renderer causes garbled text sometimes
- Repeated "Output blocked by content filtering policy" error on purely technical queries
- [Feature Request] Make safeguard model-switch fallback target configurable
- [MODEL] all claude 5 models respond with hot take poetry
- auth login: code prompt does not echo, never says to press Enter, and empty submission returns the same 400 as a bad code
- [Feature Request] Add switchModelsOnFlag to settings reference table documentation
- [BUG] Claude in Chrome extension shows "not connected" + chrome-control connector's get_page_content fails with "Chrome is not running" despite tabs opening correctly
- Subagent-attributed "Ready to code?" approval dialog shows the AO's own plan-mode content, not the subagent's, and isn't gated on the subagent actually requesting approval
- [BUG] Inline !`cmd` preprocessor executes example commands inside fenced markdown code blocks (repro: /plugin-dev:command-development runs npm test)
- [BUG] /resume shows old sessions as recently modified because file timestamps get bumped
- [BUG] /cd command does not load project-scoped skills from the new directory
- [BUG] --teleport exits 1 with no stdout/stderr when run without a TTY
- [BUG] In-session /resume picker capped at 50 most-recent sessions: scroll wraps to top, search misses older sessions (CLI claude --resume unaffected)
- [BUG] 'claude-api' is consuming entire context
- [BUG] Artifact publish auto-arms a Monitor whose stop surfaces despite quietLifecycle/ambient, with no opt-out short of CLAUDE_CODE_DISABLE_ARTIFACT
- [BUG] recommendations for next command are bad
- Ability to hide top navigation items in Cowork left sidebar
- [BUG] /cd has no directory path autocompletion in the prompt (2.1.226)
- [Bug] Anthropic API Error: Excessive content policy flagging blocking legitimate requests
- [Desktop, Windows] File links outside the session folder get no file menu, and no pane can show them
- Desktop app: @-mention renders literal "@[SessionName]" bracket text on first insert, needs delete+reselect to show as chip/link
- [BUG] @ file mentions cannot reference files inside nested git repositories (desktop Code tab)
- [BUG] claude.ai web: read state reset, blue unread dot reappeared on all previously read conversations
- [BUG] macOS Desktop: worktreeDepSeed's 680k-file clonefile starves a synchronous main-thread rename(2) — deterministic ~30 s whole-app freeze per new worktree session, discarded as "likely sleep"
- [BUG] Desktop split view: unsent drafts disappear when two sessions swap panes and only come back in the original pane
- [BUG] Claude VS extension No ide_selection context but existent selected file/line
- I don't have enough information in your message to generate a specific GitHub issue title. The message "Finally, although I expected better for graph related problems" doesn't describe a concrete bug, error, or feature request. Could you provide more deta
- [P0] Desktop Code tab: struck-through superseded prompt text does not wrap or clip — overflows the composer and the window
- Desktop Code tab: superseded prompt text overlaps the active input box instead of stacking above it
- Audit: composer children are not width-constrained (unbounded text can escape the pane) — sibling cases of #95896
- Opened by Claude without permission (misalignment)
- [BUG] azure-devops stdio MCP server returns 401 on every call after Claude desktop update (2.110.1.0 -> 2.2553.1.0), while the same PAT works outside the app
- [BUG] GitHub connector: get_file_contents reports "successfully downloaded text file" but returns `[non-text content: resource]` with no file body (get_commit full_patch returns the same content fine)
- Grounding failure + prompt non-compliance: source ignored, invented workflow
- [BUG] Claude Desktop + claude.ai: Projects grid leaves stale painted fragments after scrolling
- [Bug] Plan validation not detecting missing tool result blocks
- Repeated tasks failures, Fable 5.1
- User reports the same failure pattern for three or four days: task given, something else done
- [Bug] Subagent closes without user confirmation before exit
- [Bug] Anthropic API Error: Request Failed (req_011Cf3xwZTfCjr9cv6PCWsNU)
- [Bug] Non-English Input Causes Word Finding Issues
- TOOL FUCKS THINGS UP, CLOSES SESIOSN LIKE A HUMAN GETTUNG THERI FEELINGS HURT AND THEN SAYS ILL STOP HERE.JOW VERY CODEING
- [Bug Report: Insufficient information provided]
- [Bug] False Positive Detection Issue
- Claude in Chrome extension shows enabled but never connects to session
- [FEATURE] Let the agent trigger /compact itself (agent-invokable compaction)
- [BUG] MCP tool permission approval doesn’t unblock the tool call
- [Feature Request] Expose subagent security-warning banners to the calling session and Workflow scripts
- /copy writes the clipboard as latin-1, so non-ASCII output pastes as mojibake (still broken in 2.1.215)
- [Feature Request] Provide detailed safeguard rejection reasons for API requests
- [BUG] @ file autocomplete does not open when editing a prompt recalled from history
- [Bug] Claude Code ignores user instructions in CLAUDE.md for code style and approach preferences
- Usage report (/usage or settings Usage tab) shows materially wrong/incomplete data on first read, self-corrects later, same session
- Weekly limit reset option (like Codex)
- Weekly limit reset option (like Codex)
- [BUG] npm-distributed plugins fail to resolve dependencies — cache directory lacks node_modules
- [Bug] Team member agents spawned with `name:` parameter lose result signals in mailbox path
- macOS sandbox: hw.optional.neon missing from sysctl allowlist — all Qt 6 apps abort at startup on Apple Silicon
- [Bug] Anthropic API Error: Opus model session limit exceeded
- [BUG] hookify: `event: stop` / `event: prompt` simple-pattern rules never fire (re: #32153, two fix PRs open since March)
- Backgrounded command sometimes runs as multiple concurrent copies, causing real corruption for non-idempotent operations
- [Bug] Agent view caches terminal width at process start; split-pane repro needs no window resize (sharpens #80123)
- [MODEL] Model attributes the Opus-5-only heron_brook delegation directive to the user's own CLAUDE.md, sending users to audit config for a rule that isn't there
- Japanese (UTF-8) text file attachments are garbled in chat, but read correctly via the Read tool (path-based)
- [Feature Request] Allow users to request bash commands without agent auto-execution
- [BUG] A fast double-Esc destroys the entire prompt with no recovery, and cannot be disabled
- [Bug] Claude Code prefers shell commands over Edit tool for file mutations, hiding changes from transcript
- [Bug] Anthropic API Error: Model context limit exceeded or rate limit
- Cross-machine session directory never populates: ListAgents shows only local sessions despite both machines being RC-connected (v2.1.233)
- [MODEL] CLAUDE.md rules are recitable but not applied late in long sessions — stale-read overwrite destroyed concurrent edits in a shared Figma file
- Remote Control: supervisor exits 5s after last client detaches, so it can't be left available overnight
- [BUG] Status line reserves a blank row when output is empty in fullscreen (alternate screen) mode
- send_message (cross-session) never delivers and leaves the target session hung on an empty turn
- [BUG] Editor exiting with nonzero status after ctrl+g leads to data loss
- [BUG] CoworkVMService crashes when Browser panel tool is used to interact with a page
- [BUG] Cloud routine (CCR) stuck at "Claude Code process started" — never executes
- [BUG] Capability-string effort fall back sends `high` instead of the highest supported level at or below
- [BUG] Agent response Markdown rendering in TUI mutates meaning of content (renumbers ordered lists)
- [Bug] `claude mcp remove` command incompatible with MCP naming convention used by other commands
- Korean IME broken again after update to 2.1.235 (regression of #35307)
- macOS sandbox unusable: Seatbelt profile passed inline via 'sandbox-exec -p' exceeds ARG_MAX with many git worktrees
- Feature request: configurable safety-fallback target — Fable 5 should fall back to Opus 5, not Opus 4.8
- Jammed background tasks + Remote Control reconnect loops silently consume Max usage window after forced token rotation
- [Bug] Anthropic API Error: Account validation state mismatch for forensics access
- [FEATURE] Cowork: multi-choice question widget (AskUserQuestion) interrupts user before they finish reading the response
- [BUG] Cowork scheduled tasks never fire, even with app open and Cowork actively used (not just late — related to #36131)
- Claude in Chrome extension fails to connect from Cowork desktop session
- [BUG] Background processes not cleaned up on session exit, causing "file in use" errors on app update
- CJK text corrupted (mojibake) when copying terminal output — no-flicker/fullscreen renderer is the cause; tui: "default" fixes it [macOS + OrbStack]
- Desktop app can spawn a duplicate live session on terminal restart (Remote Control); window lifecycle undocumented
- [Bug] Overly broad safeguard triggers false positive on legitimate internal IT administration and network diagnostics
- [Bug] Claude Opus over-generates prose and ignores concise instruction constraints
- [Bug] Conversation history and files lost after restart due to Unicode normalization issue in project path
- [FEATURE/DOCS] SessionStart `initialUserMessage` is ignored in interactive sessions — support it, or warn and document as `-p`-only
- [BUG] CLAUDE.md/AGENTS.md reloaded and re-injected when cd revisits a directory via a different relative path (worktree round-trip)
- [BUG] workflow sub-agent concurrency dropped to 2 after 2.1.226 (still present in 2.1.234)
- [BUG] VS Code extension: @-mention file picker only searches the first folder of a multi-root workspace
- [BUG] Fable→Opus model_consent_fallback auto-cancels without displaying the prompt (186 events, 0 shown) — still occurring on 2.1.233 after #79337 closed
- Headless -p session exits when a turn ends while a background task is the only pending wake signal
- [BUG] VSCode extension: long pasted user messages render fully expanded with no collapse/'show less' control, pushing assistant responses out of view.
- [Bug] API Error: Non-streaming request receives event-stream response (HTTP 200)
- [Feature Request] Add Shift+Up/Down Arrow keybindings to rearrange folders in agent mode
- [Bug] Model reasoning degradation affecting agent and workflow generation quality
- [Feature Request] Add legitimate coding pattern recognition to reduce false positive security flags
- Subagent view header shows the main session's model, not the subagent's
- [Bug] Claude Code hangs on startup after update in RustRover IDE
- [BUG] Desktop app browser does not retain sessions despite session persistence set to shared
- Thai text (and other space-less scripts) misaligns/right-shifts on line wrap, especially when the terminal is resized
- [BUG]Scheduled task stalls indefinitely on Write permission prompt despite matching allow rule (settings.json correct)
- Having issue to connect
- [BUG] Bun 1.4.0 segfault at address 0x10 on main thread (JSC JSStringJoiner/JSArray::fastToString frames) after 21.8 h session with 8 live subagents — CLI 2.1.235, Windows 11 26200
- [Bug] Prompt Injection: User utterances can trigger unintended code execution
- Remote control: dead environment (404) shows misleading "Sign in again to verify your device" modal instead of environment-gone state
- [Bug] Bun runtime segmentation fault during long-running interactive session with multiple subagents
- [FEATURE] allow connecting to LiteLLM (third-party proxy) directly from the /login flow in the TUI
- [BUG] manual sorting in sidebar session groups no longer works
- [Bug] Context limit error when sesssion context space available
- [macOS] Korean (Hangul) IME composition is a daily usability blocker for CLI real-time typing — consolidating related reports
- Model emitted a standalone assistant turn impersonating a user instruction (asked to bypass review and push to prod)
- [BUG]
- Add the Browser pane's annotate (pencil) tool to the iOS Simulator panel
- [BUG] Subscription features are withdrawn when `ANTHROPIC_BASE_URL` isn't the official string, even when the proxy terminates at Anthropic
- [FEATURE] Restore Session from inside Claude Desktop
- [BUG] Directory-selection widget disappears when `projects/<slug>/memory` is an NTFS junction (Windows desktop)
- Scheduled tasks silently stop firing after first run — recreation does not fix it
- [BUG] PreToolUse hook that fails to launch is treated as a deliberate deny — exit-code 2 collision causes unrecoverable tool lockout
- [Feature Request] Add option to configure security checks behavior in Claude Code CLI
- [FEATURE] Add Czech (cs-CZ) UI localization to Claude Desktop (Windows) and claude.ai
- Mobile app: cloud environment selector allows switching/creating but not editing existing environments
- [BUG] No audit trail for retention cleanup — deleted sessions leave no log, making the loss unverifiable after the fact
- [Bug] Anthropic API Error: Request Failed (req_011CeBs7qt2diWXsMNNBM1Qw) Again this cyber bullshit
- [BUG] Auto mode fails to block tool calls that are not allowed by custom configuration
- MCP elicitation: capability is declared but requests are auto-declined ("print mode") in interactive VSCode sessions
- [BUG]
- OTel: git_commit_id is scraped from git-commit stdout — silently dropped by `| tail`/`-q`/redirection, or by the `git -C <path> commit` form
- [Bug] Session limit triggered by unexpected agent behavior outside user instructions
- [BUG] VS Code extension host spins at 100% CPU forever in forkSession after client disconnect
- [BUG] /model mid-session switch to Opus [1m] wrongly reports "not available for your account" — verdict depends on the current session's window flavor, not the account