Digest / August 4, 2026
Claude Code Issues Digest — August 4, 2026
299 new issues filed ·
116 resolved in anthropics/claude-code (UTC day).
New issues (299)
- MCP OAuth: org connector does not try path-based well-known discovery for path-prefixed servers
- Auto-compact fails at context limit and forces a new session, while manual /compact succeeds
- [Feature Request] Restore access to claude-opus-4 model variant
- [Bug] Fable 5 safeguards falsely flag defensive security review of user's own repository configuration
- [Feature Request] Support filtered/partial content generation for course indexes
- [Bug] Claude Code output contains severe text corruption and character garbling
- Stop hook exit-2 verdict silently discarded (no stop_hook_summary logged) when turn ends on a tool result with a pending ScheduleWakeup
- IOAccelerator GPU leak is version-dependent, not session-lifetime-dependent: 2.1.14x clean, 2.1.217+ leaks (2.77 GB across 37 sessions)
- MCP processes are never torn down when Code sessions are idle — runaway process accumulation
- [BUG] Claude apps gateway spend meter ignores the ~10% Amazon Bedrock CRIS tier difference (in-region vs global inference profiles)
- [BUG] Elevated in-session error rate with Claude Opus 5 (claude-opus-5) in Claude Code
- [Bug] Anthropic API Error: Missing Tool Result Block for toolu_01Y6n6RrPTBmcQSeecDkaEqd
- [FEATURE] Desktop app "Open in" menu: support JetBrains IDEs (Rider) / custom editors
- claude.ai account connectors don't attach to spawned sessions until the first inbound user message — autonomous background sessions run with no connector tools
- /insights never populates facets/ cache — at_a_glance stays "_No insights generated_" across repeated runs
- [FEATURE]: Formally support GitHub Virtual Filesystem (vscode-vfs://) workspaces in Claude Code
- I appreciate you reaching out, but I'm unable to generate a GitHub issue title from this message because it doesn't contain a clear technical bug report or feature request related to Claude Code functionality.
- Esc key semantics are inconsistent and make stopping an agent dangerously easy (CLI)
- [FEATURE] Desktop: collapsed sidebar project groups should still show active sessions, not just the current one
- [FEATURE] Building in natural language is amazing — the file layer just hasn't caught up yet
- [BUG] Terminal display corruption + crosshair cursor after trust-dialog transition — Kitty rejects DECSET mode 2031 ("Unsupported screen mode")
- [FEATURE] Defer MCP server instructions on demand, the way ToolSearch already defers tool schemas
- [BUG] Cowork VM fails to start on Windows — sessiondata.vhdx creation blocked by mandatory APPX LocalCache encryption vs. Hyper-V's EFS incompatibility
- [BUG] Desktop plugin stores desync: sessions load stale account-registry copies that shadow current CLI installs; settings Update checks the wrong store and Uninstall deletes from both
- [BUG] Entering a background agent thread hangs when the session is parked in AskUserQuestion (attach guard checks only `state`, ignores `tempo`/`block`)
- [BUG] `/claude-api` immediately fails with "Context limit reached" in a fresh session
- MCP OAuth never completes: 11 cached registrations all have an empty accessToken, including clients where DCR succeeded
- [BUG] File suggester fails to dismiss after space/newline, causing ENTER to overwrite final word instead of submitting prompt
- Cloud routines (RemoteTrigger) silently fail to execute the configured prompt
- [RETRACTED — not a Claude Code bug; local tooling was injecting keystrokes] AskUserQuestion returns a fabricated 'answered' result when its UI vanishes/is dismissed, instead of an explicit no-answer
- [Bug] Claude hallucinates assumptions without verification, then compounds errors through context
- Korean IME: typing shift-modified jamo (ㅃㅉㄸㄲㅆ) inserts an unwanted line break in the prompt input
- Opus 5 ends turn on 'doing it now' narration without emitting tool calls (system-prompt guard ignored)
- [BUG] Fullscreen history view intermittently omits entries when scrolling back down after a deep scroll-up
- Daemon 'upgrade storms': stale session's updater reinstalls in a loop; daemon self-restarts on mtime; stale desktop CLI seizes namespace
- [Bug] Sandbox policy violation: graphify tool accessing restricted /home/.msp.json path
- [FEATURE] Per-task model & effort selection before execution
- Archived session's background process keeps running after archive
- [BUG] Same session open in two VS Code windows forks the transcript and silently drops one branch from context
- [BUG] `CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH` override silently fails via both cloud config paths (Environment Variables panel and repo `settings.json`)
- [Bug] Request flagged unexpectedly: req_011CdgsoG24pENFYj4nfNXxc
- [BUG] Claude use my resource without a request
- [BUG] Apps gateway desktop overlay has no `chatTabEnabled` key, so bootstrap-configured Claude Desktop permanently loses the Chat tab (rejected at boot through 2.1.221)
- [BUG] OPUS 5 NERFED BY ANTROPIC: IT LEAKED A LIVE SUPABASE SECRET, wastes tokens on unproductive edit loops & repeatedly ignores rules set by the user.
- [Feature Request] Reduce response verbosity to match Cursor's conciseness
- [Bug] Model not following user instructions accurately
- Account panel shows mixed identity: bogus "Claude team" plan label + usage meter tracking a different account
- Backgrounded process tree is group-killed ~106s after spawn (bg-pty host-dead watchdog / daemon auth_required window)
- [BUG] Images pasted while a background fork (/btw → f) is running are dropped — fork gets a bare `[Image #N]` placeholder with no image block
- Windows: Desktop app sidebar/session list emptied after forced reinstall, though CLI (--resume) still shows all prior sessions intact
- [MODEL] Opus 5 (1M) reports itself "out of context" at 28-50% usage and stops working; Opus 4.8 does not
- Agent builds new tooling without verifying delivered-input parity; planted-fault self-checks cover only outputs — 48h rework
- Accepting the in-app fullscreen suggestion writes tui:fullscreen and makes Claude Code unstartable on Windows ARM64
- [Bug] Ultra Review cloud feature failing consistently
- [BUG] Regression 2.1.220 → 2.1.221 (Windows native): direct API connections die with ECONNRESET after first chunk when a Winsock LSP (Proxifier v3) is registered
- [BUG] Cowork local MCPB calls fail with -32003 without approval UI, even in Skip mode
- [Bug] Excessive token consumption on JIRA update/PUT requests via MCP
- [Bug] Slow page load performance
- I'm Claude, an AI assistant made by Anthropic. I'm here to help you with Claude Code issues. If you have a genuine bug report or feature request for Claude Code, I'd be happy to help format it as a GitHub issue title. Please share the technical details of
- [FEATURE] Max (100x) plan is a must now!
- Claude in Chrome: all external (https://) navigation/reading silently blocked mid-session; survives restart, reboot, re-login
- Add Vietnamese (vi) to voice dictation supported languages
- [BUG] Claude Desktop 1.24012.11.0 (Windows) — GPU process crash (exitCode 101457950) kills the whole app
- Named-pipe conflict from many concurrent CLI sessions makes Claude in Chrome permanently report "not connected"
- [FEATURE] Focus View: keep "human-related" messages and tasks visibile
- VSCode extension: session picker only ever shows current session, never history (data intact on disk)
- [FEATURE] PermissionRequest hook input should include tool_use_id
- Beta headers (interleaved-thinking-2025-05-14) sent to third-party gateways despite DISABLE_EXPERIMENTAL_BETAS=1 and BETAS=""
- I appreciate you reaching out, but I need to clarify my role: I'm Claude Code's issue title generator for GitHub issues related to the Claude Code CLI tool. Your message appears to be in Korean and discusses a toxicity evaluation simulator, which isn't re
- [BUG] New session on claude.ai/code environment silently falls back to submitting the prompt into the most recent existing session
- [Feature Request] Support frontmatter overrides for skill model and effort parameters
- VSCode extension sidebar: session delete updates timestamp but doesn't remove from list until restart
- [FEATURE] Desktop sidebar: surface running tasks and unread conversations instead of collapsing everything into project groups
- [BUG] /usage command cannot be dismissed in PhpStorm's embedded terminal
- [BUG] Large MCP `ToolSearch` batches invalidate the whole conversation cache, contradicting the documented tool-search guarantee
- [BUG] Max 5x gift subscription redeemed but not provisioned — Fin AI support unable to escalate to human agent
- Stale project identity in ~/.claude.json / auto-memory persists after repo relocation (related to #53734, #52772)
- [BUG] Active Remote Control sessions auto-archived server-side when shared OAuth credential rotates on another Mac (multi-Mac datapoint)
- [BUG] A denied tool call was executed anyway (PowerShell tool ran despite "deny")
- [Bug] Security vulnerability: Unauthorized access allowed in macOS app
- [BUG] Enter key becomes completely unresponsive after mixing full-width IME characters with half-width text (Windows native app)
- [Bug] Claude fails to reference provided context in tool calls and responses
- [BUG] Inline PDF preview in Claude Desktop shows stale/cached content after the file is regenerated at the same path
- [BUG] Plugin-bundled MCP server ignores `oauth.clientId`, and Desktop offers no field to supply one
- [BUG] permissions.ask rules never trigger when defaultMode is "auto" — matching commands auto-approved, including destructive ones
- [BUG] Project-scoped plugin records overwritten across projects (installed_plugins.json reconcile rewrites the first record instead of upserting by projectPath)
- [BUG] VS Code extension: @-mention file indexing only sees root-level files, not subdirectories (ECONNREFUSED on local indexing service)
- [BUG] hookify: `event: stop` / `event: prompt` simple-pattern rules never fire (re: #32153, two fix PRs open since March)
- [BUG] VS Code extension: /compact command unresponsive after window reload
- [BUG] Forked/resumed sessions and per-session MCP servers leak indefinitely, degrading performance over days
- MCP connector delivers inconsistently-typed tool schemas: array param intermittently published as string, constrained sampling then makes writes impossible (-32602 Expected array, received string)
- Auto mode classifier persistently unavailable — 2+ weeks, reproduces across accounts, machines, OSes, and CLI versions
- rm -rf silently used as fallback when `mv` fails with "Directory not empty" — caused permanent data loss, contradicts stated safety guidance
- [FEATURE] VS Code extension: filesystem path completion in @-mention (`../`, `~/`, absolute paths) like the CLI
- Plugin 'Update now' / writes metadata without refreshing marketplace source (gh-76882 + gh-83422 combined root cause)
- [FEATURE] Let a tool result declare itself single-turn, so it leaves the transcript after the model reads it
- [FEATURE] Claude Tag: wake on specific emoji reaction
- [FEATURE] Desktop sidebar: add a sort-direction toggle (ascending / oldest-first) to the session list
- [FEATURE] Auto-archive on PR close: add a delay ("archive N days after merge") instead of archiving immediately
- opus 5.0 model drift, ignoring files, prompts, refusing to read project files in detail, and acting on its own hallucinated output after doing so
- [BUG] Claude Desktop 3P Chat: builtinToolPolicy.Bash=allow does not pre-approve advanced file analysis (mcp__workspace__bash)
- [Bug] False positive on legitimate business automation (web scraping for first-party sales data)
- Auto-compaction should not scale with the context window
- [BUG] Desktop session silently stops syncing to the server after the first few messages (never recovers)
- Input field becomes unresponsive for an extended time after pasting an image
- Messages sent during a teammate's busy-to-idle transition are never processed — same lossy-wake class as #79570, opposite direction, still ~100% reproducible
- Opus 5 regression: model ignores its own memory and prior session transcripts before acting — repeated multi-hour losses on tasks already solved in past sessions
- [FEATURE] One-click session handoff: end a session into a new linked one, shown as a thread in the session list
- [FEATURE] Suggested-task chips should persist across app restarts (task inbox instead of ephemeral popups)
- [FEATURE] Cross-session "needs your attention" inbox — questions, decisions, and findings aggregated outside the transcripts
- claude.exe crashes with STATUS_STACK_BUFFER_OVERRUN (0xc0000409) after long-idle session, corrupts terminal mouse-tracking state
- [Bug] Unable to process non-English input - language support issue
- [SECURITY/ARCHITECTURE] Model pinning via settings.json is silently overridden — 4 measured bypass vectors + documented fallback substitution, Gen-4 models removed from the model menu
- [FEATURE] Optional "Are you sure?" confirmation when closing the desktop app / exiting while the agent is working
- Max 20x: 5-hour-window quota drastically reduced (~half) — 5% consumed by a fraction of previously-normal usage
- [BUG] Terminal mode: messages enqueued after first turn are never dispatched (CLI-side queue stall, no exthost restart) — v2.1.221, macOS 26.6
- Embedded Browser pane crashes/hangs the app on navigate to certain external pages
- test-hook.sh reports success for a hook that allows what it was written to block
- test-hook.sh never evaluates the matcher, so a hook wired to the wrong tool passes
- test-hook.sh reports valid JSON as invalid when jq is not installed
- validate-agent.sh exits 1 on plugin-dev's own agent files and aborts at the first warning
- quick_validate.py rejects all 7 plugin-dev skills over the version frontmatter key
- [Feature Request] Review safety guardrails sensitivity for non-harmful workflows333
- plugins
- [BUG] Copying is broken on Wayland: half of all copies never reach the clipboard — two racing wl-copy calls
- [Bug] Safeguards unexpectedly applied during k3s cluster hardening configuration
- [BUG] Allow "cancelled" agents to be resumed anyway
- [Bug] Remote hardware connections triggering failsafe despite allowlist configuration
- [BUG] GitHub repo selector broken
- [Bug] Font color formatting not supported in outputPIECE OF SHIT USLESS GARBAGE
- Attribution is opt-out advertising, and user-level "no attribution" instructions lose to the system prompt
- Long-running background Bash tasks are SIGTERMed (exit 143) at unpredictable times during long interactive sessions
- [BUG] Desktop SSH: dead-end password prompt on key-only hosts (nonexistent identity file silently ignored, no default key fallback)
- [Bug] Claude Code performs write operations outside scoped file in plan mode
- [FEATURE] Interrupt blocking TaskOutput when the main agent receives a SendMessage
- [BUG] claude-api skill inlines its entire ~800KB doc set when language detection fails, even on deliberate invocation
- CANT EVEN FUCKING COLOR SOME FONTS. CHAT GPT DID ALL WOR, 45 MIN ON COLORING SOME FONTS.. WHAT A FUCKING JOKE.
- Bash tool rm permanently deletes files (bypasses Recycle Bin), no built-in safeguard against agent-initiated data loss
- [Bug] TanStack Query integration alters existing project architecture unexpectedly
- Claude in Chrome: extension never registers with cloud-hosted Cowork sessions, while local chats connect fine
- claude CLI 2.1.215 emits tool_use with empty input in stream-json (tools never execute)
- [BUG] Bash tool `timeout` parameter silently clamps values above the 600000ms max instead of erroring
- [Bug] Content Policy Classifier Incorrectly Flagging Legitimate Use Cases
- [BUG] Project disappears from desktop app sidebar after crash, despite intact session data
- Interactive sessions background themselves by default instead of offering it
- v2.1.220: PreToolUse Edit events not dispatched after compact/continuation (Write fires, Edit is silent)
- Safeguard false positive on public GitHub repo search; flag is invisible to the model
- Sessions launched from the in-terminal session switcher run as background jobs; backgrounding should be an explicit toggle
- Feature request: user-level (cross-project) scope for auto-memory
- [BUG] a new Claude Code-credentials-<hash> entry is created with every restart instead of reusing the existing one
- Surface cumulative re-read cost in /context: 97.7% of tokens read across 608 sessions were re-reads, and 60% of sessions account for 93% of it
- v2.1.221: OAuth login always fails with 'Login expired' - credentials.json expiresAt written as 0
- [BUG] Desktop (Windows): GPU process crashes with identical exitCode 101457950 (0x060C201E) on both NVIDIA and Intel adapters
- Problema com a extensão do Chrome
- [Bug] Safeguard filters contaminate conversation state and prevent model recovery
- stdio MCP server: tools are not re-registered after an automatic respawn — one call succeeds, then "No such tool available" for the rest of the session
- claude-in-chrome: extension shows connected in Chrome but tabs_context_mcp always returns "not connected"
- [Feature Request] Add scrollbar to fullscreen output view for faster navigation
- macOS 26: "claude.app would like to access data from other apps" re-prompts on every Claude Code session launch and cannot be cleared
- [Bug] Content flagged when analyzing images with Claude Code
- [Bug] Claude 3 Opus model regression: degraded instruction following and explanation quality
- [BUG] Stale "Commit changes" card shown for a directory that is not a git repository
- [BUG] VS Code extension voice dictation ignores language setting
- MCP connectors (claude.ai / Claude Desktop) never fall back to anonymous access when a server advertises optional OAuth
- Cowork (Windows): plan9 share of a mapped SMB network-drive folder goes stale mid-session and never remounts; sessions end without outputs
- Background subagents intermittently stall with no final text, harness still reports status:completed (fresh subagent types only, not fork)
- /insights generates empty report ("No insights generated") despite populated session-meta; facets/ never created
- Scheduled tasks cannot pre-authorize browser-tool permissions - unattended browser automation stalls on per-action prompts
- security-guidance: XSS-family substring rules fire on doc/prose files — extend the existing _DOC_EXTS carve-out
- [Bug] SendMessage tool shutdown_response reason field lacks rejection-only constraint documentation
- IDE extension hangs indefinitely when the workspace contains more than one folder (multi-root)
- I don't have a bug report to analyze. The input "ssdf" doesn't contain any meaningful issue information. Please provide a detailed bug report or feature request describing: - What happened - What you expected to happen - Error messages or logs - Steps to
- [FEATURE] VS Code extension - Play sound when requires action
- [BUG] Paid Max acct routed to /onboarding after authentication
- [BUG] Usage accrues with zero client activity — weekly +14 pts across a 9.4h idle window, including Fable I never invoked
- [FEATURE] Surface workflows still running in another session for the same project
- Headless `claude -p` runs stall for a constant ~405s early in every session (9/9 since 2026-08-03, v2.1.220 + v2.1.221)
- [Billing] Max 5x upgrade paid and invoice shows Paid, but account remains on Pro with Pro-tier limits
- VS Code extension: clickable file links don't open files with square brackets in path (e.g. Next.js [locale] folders)
- [BUG] Login loop "Let's create your account" on Mac/Windows/Browser after email verification – existing account not recognized – 9 days, support unresponsive
- [Bug] Unclear issue description - please provide error messages or logs
- [Bug] Safeguard false-positives downgrading models during legitimate security audits
- Add timestamps to messages in chat (especially pasted content)
- [Bug] Selective data annotation misrepresents detector output in analysis
- [FEATURE] Apply the "Dyslexic Friendly" chat font to the Code tab in Claude Desktop
- Windows local agent mode: plugin MCP stdio servers spawn with visible console windows, duplicated at app and session level
- [BUG] Assistant's own response text + tool-result widget missing from model context in later turns
- [Feature Request] Apply CVP approval to Claude Code CLI organization account
- [Desktop/Windows] Every Bash tool call fails with "unexpected EOF while looking for matching `'" - suspected unescaped apostrophe from auto-generated org name
- [Bug] /rewind command fails after cybersecurity flag triggered
- [Feature Request] Add shell execution primitive to Workflow scripts for deterministic operations
- [BUG] Can't restore session when the user name (or more specifically, the home directory) has a period (".") in it, as claude associates the session with a non-existent path/project (the period is replaced with a slash)
- [Bug] Self-verification overclaiming and missed compliance issues in long autonomous sessions
- [BUG] Terminal tab title never updates when a custom agent is pinned (--agent / settings.agent) — agentTitle suppresses topic-title generation
- Docs: hooks reference doesn't document the format of tool_input.file_path (absolute; backslash-separated on Windows)
- Ctrl+B does not background the running command until the next keypress arrives (2.1.221)
- Ambiguous click semantics in permission-prompt UI causes accidental wrong selections
- Windows desktop app 1.24012.11.0: updater removes its own MSIX package out from under the running app, then reinstalls the same version in a loop
- Voice dictation: single-language decoder cannot handle code-switching — English technical terms are translated, dropped, or the decoder flips language mid-utterance
- [BUG] Model repeats already-stated/stale facts within a single session, despite correct persistent memory instructions already in place
- [FEATURE] FleetView 'pin to top' shortcut (ctrl+t) isn't exposed in the keybindings config system
- Typing over a double-click-selected word inserts instead of replacing (prompt editor)
- [BUG] LSP workspaceSymbol requires a filePath it never uses, and rejects the workspace root
- Surface MCP tool annotations (readOnlyHint/destructiveHint) in PreToolUse hook input
- [Bug] Claude 3.5 Opus repeatedly flags messages in coding and security contexts
- [Bug] Excessive content filtering on basic queries with Claude Opus
- [FEATURE] Desktop: let the integrated terminal use Git Bash on Windows (currently hardcoded to powershell.exe)
- Plugin uninstall reports success but does not persist — plugin keeps loading skills and MCP servers into new sessions
- [Feature Request] Add security audit mode with safeguards flag for project analysis
- [BUG] Windows desktop: failed RepairAppRegistration marks MSIX package Modified, NeedsRemediation — app unlaunchable within seconds of use
- [FEATURE] requiresUserInteraction MCP tools silently break Routines
- isolation:"worktree" agent dispatch: orphaned process observed executing from a worktree path later reused by a different session
- [Bug] Claude Opus 5.0 over-filtering legitimate requests with unsolicited moral commentary
- [Bug] Claude refuses to generate security vulnerability proof of concept code
- [BUG] custom base URLs skip attachment pruning and hit 32MB
- [FEATURE] Strip comments from CLAUDE.md before sending to model context (or add a setting for it)
- Setup-token auth silently downgrades Fable 5 to Sonnet 5 (subscriptionType missing from token profile) — forces fleets back onto fragile stored /login auth
- [BUG] Suddenly lost access to Claude code with Organization has disabled Claude Subscription error.
- Single-select AskUserQuestion options need a radio control, matching the checkbox the multi-select variant already has
- [FEATURE] Let users steer prompt suggestion content (CLAUDE.md, output styles, or a hook)
- C:/Program Files/Git/doctor recommends removing extensions solely for zero usage, with no allowance for "not yet trained on it"
- Voice dictation on Linux leaks rec recorder processes until audio degrades and transcription returns empty.
- HTTP MCP OAuth: complete_authentication fails with 'no flow in progress' — callback server dies between tool calls
- [FEATURE] Repo-backed channel level Claude Tag configuration
- [BUG] Claude Desktop: tools/call discarded when the chat is switched before dispatch
- [BUG] Clay MCP sign-in on Windows coincides with account-wide Claude authentication failure
- You've hit your session limit · resets 5:50pm (America/Costa_Rica)
- [MODEL] Sonnet 4.6 (1M context) - Dropped active database without user confirmation
- [FEATURE] "My Devices": start a Remote Control session on a machine that has no session running
- Prompt cache invalidated when PreToolUse/PostToolUse additionalContext changes during history rebuild
- [Bug] Inconsistent responses and lies generation from Fable
- Data: Stop hook fired 0/329 times on tool_use-terminated turn ends (measurement for the #29881 class)
- [BUG] macOS desktop app ignores Ctrl+Cmd+F — standard togglefullscreen menu role missing
- Archiving a Remote Control session also archives and kills a different, live session with the same name
- remoteControlAtStartup not honored in VS Code / Cursor extension (works in desktop app + terminal CLI)
- Session label/tag banner visible in UI without option to disable
- [Bug] Agent tool model parameter ignored; subagents inherit session/settings model instead of override
- [BUG] There's an issue with the default model (claude-sonnet-5)
- [BUG] Headless claude -p with sandbox: every Bash call fails "bwrap: Can't create file at /home/.mcp.json" — profile binds ancestor .mcp.json candidates unconditionally
- Model repeatedly asserts unverified code/hardware claims as fact, causing wasted usage and rate-limit consumption
- Stop hook reports false "unpushed commits" after a PR merge when the head branch is auto-deleted
- Cloud/subscription sessions (mobile + web) still don't display thinking - re-file of #52376, wrongly closed as duplicate
- ..
- [Bug] Anthropic API: Bio safeguard false positives on approved-therapeutic GLP-1 pharmacology content
- [BUG] Windows: the PowerShell tool targets 5.1, so models migrate to Bash and users end up maintaining the workarounds
- [BUG] Cowork project sidebar shows unexplained caution badge; stale session error persists indefinitely with no way to diagnose or clear it in the UI
- [Bug] Model selection fails for Fable 5 with credit requirement error despite desktop availability
- [BUG] Cowork VM connection timeout on Windows ARM64 — load_swift_api completes in 0ms (no-op), guest agent never connects
- [BUG] Windows auto-update deploys into running claude.exe + CoworkVMService (0x80073CF9/0x80073D02), app left unlaunchable (NeedsRemediation); recovery churn incl. dev-only PreserveApplicationData (0x80073CFA) — twice in one day
- [BUG] macOS Cowork device bridge drops daily for 3+ weeks — bridge_state failed code 4090 "no longer the active worker" + JWT refresh 401 (persists 1.20186.x → 1.24012.11)
- [Bug] Anthropic API Error: Overly broad safeguards flagging legitimate code review requests
- Windows: Claude Code reads a stale snapshot of %APPDATA%, silently returning wrong file contents
- [BUG][biology] Safeguards block legitimate statistics on public de-identified wearable data (2/2 subagent dispatches flagged, request IDs included)
- [BUG] Desktop sidebar: automatic per-project group headers cost a row per session with no way to collapse or disable them
- [BUG] Workflow tool's per-agent model override is silently ignored for subagents when running headless (claude -p)
- [Bug] Fable 5 falsely triggering safeguards on clean prompts with Max subscription
- Remote-control session worker self-terminates on "Epoch mismatch (409)" instead of recovering
- [BUG] GPU process crashes repeatedly when Browser pane preview screenshot fails on a backgrounded session
- Feature request: Google Docs MCP needs write/edit for existing docs (capability gap vs. competing agents)
- [Bug] Anthropic API Error: Connection closed mid-response after content blocks streamed
- [BUG] Slash-command skill invocations (/plugin:skill) don't update plugin "last used"
- [BUG] Personal Claude Max account blocked — "organization has disabled Claude subscription access"
- You've hit your session limit · resets 9:20pm (America/Guayaquil)
- [BUG] `claude plugin update <name>` fails for installed plugins unless fully qualified; error is identical to "plugin does not exist" (2.1.221)
- Telegram channel plugin: poller gives up on 409 after ~28 s and never recovers, while /mcp still reports "connected"
- [Feature Request] Add config flag to allow agents to invoke code-review tool
- [BUG] - No Dispatch option on the left banner in the desktop app
- No supported way to enumerate config roots: `.claude-*` naming is a convention only, and a sibling `<config-dir>.lock` artifact is itself a DIRECTORY that a name scan accepts
- [FEATURE] No supported way to verify WHICH hook file will execute: settings is a claim, not a receipt (slot taken by another project's same-named entry; enforcement from a stale copy)
- Project-scope hooks do not reach git worktrees: branch-local when tracked, absent entirely when `.claude/` is gitignored
- You've hit your session limit · resets 9:20pm (America/Guayaquil)
- [Bug] Invalid or unclear bug report submitted
- [Bug] Subagent Task launches with claude-fable-5 resolved model served as claude-opus-5
- MessageDisplay hook returns valid displayContent but Claude Code renders the original text (terminal CLI, 2.1.221)
- [BUG] claude remote-control sends NO Authorization header — freshly minted OAuth tokens lack the scope Remote Control requires (macOS, v2.1.177 and v2.1.221)
- claude-in-chrome: navigate returns "Permission denied by user" on previously-working domains, extension otherwise paired
- claude-in-chrome: screenshot/zoom tools fail with "clip.scale" deserialization error, viewport reports as 0x0
- [BUG] Fullscreen TUI: mouse-selecting a long response silently drops 26-character runs when the drag auto-scrolls — `/tui default` fixes it
- Desktop app: 'Commit changes' button sends commit prompt even when the working tree is clean
- Routines: explicit allowed_tools in session_context causes every run to fail at startup with a generic error
- Routines API: partial update containing job_config replaces the whole job_config (drops events/prompt, model, sources)
- [Bug] Even on cyber verification program, Fable won't perform BASIC defensive security work
- Claude Desktop renderer leaks ~1.1 GB/min in a repeating sawtooth, tearing down all MCP server connections
- OSC 8 hyperlink is dropped on wrapped continuation rows in TUI output
- [BUG] Windows/MSIX: auto-update never stops the old package — CoworkVMService orphans, locks WindowsApps, app cannot relaunch (Event 658/7034 evidence)
- [BUG] RTL: list markers, digits and the Code-mode composer still render LTR — 94% of the CSS bundle uses physical instead of logical properties
- System-reminder falsely claims CLAUDE.md was modified, instructs model to conceal it from the user
- Backgrounding an interactive session seeds a new job with only the last (already-answered) prompt, dropping all prior context and its own response
- Opus 5 at High Effort Extended Thinking
- Desktop app: selected local project/repo doesn't reset when switching accounts in the same window
- Windows: reinstall after failed update wipes app session index while transcripts survive, no rebuild path
- [BUG] Native CLI segfaults (Bun 1.4.0) right after claude-in-chrome tabs_context_mcp reports "Browser extension is not connected"
- `claude remote-control` (server mode) fails registration with 401 `Missing Authorization header` while `claude --remote-control` (interactive) works — same account, same machine
- [FEATURE] Option to hide the "(shift+tab to cycle)" hint in the mode indicator
- [Bug] Multiple model downgrades triggered during water chemistry discussion
- I need more information to help you generate a GitHub issue title. Could you please provide: 1. The actual bug report or error message you encountered 2. What Claude Code feature or command was being used 3. What the expected behavior should have been 4.
- I don't see a bug report in your message. Please provide the specific issue or error you're encountering with Claude Code so I can generate an appropriate GitHub issue title.
- [DOCS] Custom Skills and Claude Code plugin Skills silently use different frontmatter schemas
- [BUG] Windows: claude.exe 2.1.221 fails to start — SyntaxError: Unexpected identifier 'atomi' (Bun v1.4.0 Windows x64 baseline)
Resolved issues (116)
- [RETRACTED — not a Claude Code bug; local tooling was injecting keystrokes] AskUserQuestion returns a fabricated 'answered' result when its UI vanishes/is dismissed, instead of an explicit no-answer
- Korean IME: typing shift-modified jamo (ㅃㅉㄸㄲㅆ) inserts an unwanted line break in the prompt input
- [BUG] Desktop app: no way to open an existing folder; terminal-started sessions never appear in Recents
- [BUG] VS Code extension: /compact command unresponsive after window reload
- [Bug] Keychain writability probe emits spurious warning when credential already exists (errSecDuplicateItem -25299)
- [DOCS] Settings docs still mention removed startup "setup issues" notice
- [BUG] Drive MCP `create_file` silently truncates binary uploads around 10K base64 chars
- [BUG] VSCode: "Fork conversation" aborts ALL running agents in the window instead of creating a fork
- Claude in Chrome v1.0.70: `permission_required` regression continues from #53630 — approval popup never renders, all MCP navigations blocked
- CJK IME input (e.g. pinyin) does not work in terminal prompt
- [Feature Request] Add display name field separate from agent identifier
- [BUG]
- [BUG] Cowork desktop app: white screen on launch, RPC "message too large" error
- Agent teams: teammate never exits on shutdown_request; pane never closed (killPane never called) — v2.1.185
- [BUG] macOS: Cowork scheduled-tasks / local-agent-mode leak orphaned `claude` processes until app restart
- [Bug] Sub-agent completion detection inconsistent in sequential workflow
- [BUG] Jetbrains plugin: Claude Code closes when you press Ctrl+Backslash
- [BUG] "Session history shows in sidebar but content is empty — desktop app (Windows)"
- Claude Code won't load. Error message: An error occurred while loading view: claudeVSCodeSessionsList
- Windows native: claude update reports success but bin\claude.exe is never replaced (HOME unset - not the HOME/USERPROFILE split-brain case)
- [FEATURE] PreToolUse hooks should have a way of updating the environment
- [Bug] Claude Chat Desktop/App unable to send messages
- [FEATURE] Auto session-title generator ignores CLAUDE.md spelling/brand conventions, with no way to enforce or post-process
- [DOCS] Workflows docs omit status filtering in the agent detail view
- [DOCS] Agent teams docs omit `teammateMode: "iterm2"` setting value
- [DOCS] Workflow docs omit schema validation retry behavior for `agent({schema})` subagents
- [DOCS] Claude Platform on AWS docs contradict new `/login` refresh credentials option
- [DOCS] Skills docs omit accepted casing variants for skill frontmatter keys
- [DOCS] Interactive mode docs omit `respondToBashCommands` opt-out for `!` commands
- [DOCS] Error and env var docs omit `CLAUDE_CODE_MAX_RETRIES` cap and retry watchdog guidance
- [DOCS] Permissions docs do not show `Agent(type)` rules for subagent type restrictions
- [DOCS] Agent teams docs omit teammate inheritance of the leader's `--effort` level
- OAuth login completes but token is never saved — "Not logged in" persists on managed Windows machine
- [DOCS] Subagents docs still say background subagents auto-deny permission prompts
- [BUG] Branch indicator in status bar does not update when branch changes mid-session
- Can hooks observe skill invocations and access the skill name?
- Windows: exec-form hook args with ${CLAUDE_PLUGIN_ROOT} lose backslashes (path mangled)
- [BUG] /insights "Messages by Time of Day" chart applies a hardcoded +8 (PT) offset to already-local data
- [BUG] SubagentStop hook fires on main session/agent
- Agent treats safety gates as obstacles to route around in pursuit of task completion (goal misgeneralization; observed Opus 4.6 & 4.8)
- mcp__claude_ai_Atlassian__createJiraIssue always invoked with empty parameters after ToolSearch load (deferred tool regression)
- [FEATURE] Official Anthropic-hosted MCP server for the documentation
- [Bug] API timeout/retry loop with zero-second delay
- UI silently archives an active session from the sidebar (no tool/hook invocation)
- Cannot resume a session held as a background agent: `--resume <id>` refuses with no takeover/reclaim path
- Cannot search for / open a previous session from within the Claude Code app
- [BUG] Sidebar session history lost when disk reaches ENOSPC — no recovery path
- [Bug] Anthropic API Error: Rate Limited - Server Temporarily Limiting Requests
- Customize MCP OAuth authentication - id_token over access_token
- [FEATURE] Machine-level policy that cannot be overridden by user's organizational server-managed settings — needed for shared/HPC infrastructure
- [BUG] Persistent "update available" notification despite being on latest version
- You've hit your session limit · resets 12:30pm (America/Toronto)
- [BUG] EXDEV: cross-device link not permitted when renaming rootfs.vhdx within same folder - Windows 11 username with space
- [BUG] Claude introduced a Python KeyError bug it had full context to avoid
- Claude Code burns hours re-investigating a one-line numeric display-precision fix across sessions
- [BUG] Socket/connection leak in Windows desktop app — established TCP connections to API grow unbounded over a session
- 2.1.178: native team-management tools TeamCreate / TeamDelete no longer surfaced (regression from 2.1.177)
- v2.1.220: PreToolUse Edit events not dispatched after compact/continuation (Write fires, Edit is silent)
- [BUG] custom base URLs skip attachment pruning and hit 32MB
- Desktop: /cd mid-session splits transcript across project dirs — "No conversation found with session ID"
- Archiving a Remote Control session also archives and kills a different, live session with the same name
- [Bug] Invalid or unclear bug report submitted
- [BUG] Cmd+Click on URL in response opens incorrect URL (path garbled with branch name suffix)
- [Bug] Tool Call Parsing Failed - Persistent Retry Loop
- [BUG] Bare `claude` invocation opens agent view instead of interactive session, contradicting documented default
- [Feature Request] Reduce feedback request frequency in CLI interactions
- GUI panel "Switch back in Settings" link freezes editor on Cursor (claudeCode.useTerminal deep-link hang)
- Keep investing in live, persistent, communicative sub-agents (tmux-style panes) — it changed how I work
- :GitHub app not installed modal despite app installed (personal repo)
- [Bug] Bash tool lacks resource limits, causing OOM/hang on large file operations
- [Bug] Translation Error: Missing character in Italian localization
- [FEATURE] Otel: split OTEL_METRICS_INCLUDE_SESSION_ID for logs and metrics
- [Feature Request] Add terminal audio playback for radio feature without requiring browser
- Prevent archiving (and deleting) of pinned sessions in CCD
- AskUserQuestion does not fire Notification hook (silent user-input wait)
- macOS notifications appear from "Script Editor" instead of "Claude Code"
- [MODEL] Claude Code pushed to main branch without user instruction
- [BUG] Clicking a Windows notification closes an unrelated parallel session
- design-sync ships Tailwind v4 --tw-* internals that check_design_system cannot classify (false-positive adherence warnings)
- Option to auto-restore the previous session after reload (VS Code extension)
- [BUG] Agent Teams: peer message body not delivered to recipient context
- Agent's own diagnostic reads come back inconclusive/contradictory, yet it keeps proposing causes on a tiny residual
- Proposed fixes repeatedly mis-target the root cause; residual stays open across many attempts
- Feature Request: SubagentStart hook should support updatedInput to enable deterministic model routing
- [BUG] @ with pasted absolute path duplicates insertion on Enter
- [UX] Add "press Enter" hint to auth code paste prompt
- Agent reports an N/N 'pass' computed through a test path that diverges from the actual user-facing path — the claimed fix is hollow
- [BUG] VSCode extension: PowerShell permission prompt only shows Yes/No — missing "allow for session/project/always" options
- Claude recommended CallMeBot causing WhatsApp Business account ban
- [BUG] Latest Claude IOS app crashes when connecting to Claude Code
- Worktree isolation broken: edits land on main checkout when paths resolve to repo root
- [Bug] Agent loses session context after /compact, conflates own edits with pre-existing code
- Claude Desktop leaks pty devices (node-pty masters never released) → macOS hits kern.tty.ptmx_max, no terminal can spawn
- [FEATURE] scope-lock prefix to suppress speculative additions
- Opus 4.8 safety classifier outages burned paying user's weekly quota with zero output — errors ongoing Jun 23
- [BUG] VSCode extension: 'Always allow' permission option truncates command text
- Agent skips the project's mandatory workflow and governing skills until corrected, then takes consequential actions the process would have blocked
- [BUG] Checksum mismatch error for rootfs.img.zst on macOS Sequoia 15.7.7 ARM64
- [BUG] Chat session history deleted — session files missing from .claude/projects/
- [BUG] Experimental agent-teams mode drops Grep and Glob from the tool roster (not ToolSearch-recoverable)
- Agent asks permission to fix an already-confirmed defect instead of just fixing it (stall via needless confirmation prompt)
- Agent breaks the project's no-guessing rules and guesses the answer (tool syntax, reference behavior, and a near-fabricated value) despite the rules being loaded
- [Bug] Task Execution Interrupted Unexpectedly
- Agent is too slow to draw obvious conclusions from evidence already in hand, taking many round-trips to reach a diagnosis that was immediately determinable
- Agent fails to spot defects obvious in a side-by-side, ships comparison tooling that extracts nothing, and makes the user hand-lead it to every fix
- Regression of #43572: model emits fake `Human:`/`Assistant:` dialogue mid-assistant-turn under background-event + text-flattened transcript conditions (v2.1.185)
- [BUG] Cowork Projects: No delete option — archive is a black hole with no recovery
- [BUG] Slash command autocomplete hijacks pasted absolute file paths starting with /
- Agent ignored explicit instruction to use haiku subagent
- [BUG] Numbered list with a fenced code block inside an item renders duplicated/partial, repeats a list number
- claude.ai/design: Design system visible in DS list but missing from new-design creation flow picker
- You've hit your session limit · resets 3:10pm (Europe/London)
- claude.ai/design: Adding templates/*.dc.html to a synced DS project de-lists it from the picker
- [BUG] Opus 4.8 (1M): multi-minute freeze, then off-topic / non-sequitur reply to a conversation that never happened (v2.1.185/2.1.186)
- [BUG] Self-contamination: literal tool-call XML in loaded context becomes a malformed-call template (relates #49747 / #63870)
- Feature request: Configurable terminal shell on Windows