Digest / July 30, 2026
Claude Code Issues Digest — July 30, 2026
317 new issues filed ·
113 resolved in anthropics/claude-code (UTC day).
New issues (317)
- Self-reported: Claude Code violations caught by the user 72.5% of the time (user-configured accountability system)
- [BUG] Dispatch missing from Claude Desktop sidebar on Linux 1.24012.9 — visible on Android, same Pro account
- NO_COLOR not respected by spinner status text and permission-mode indicator
- Backgrounded (&) shell children survive a Bash-tool timeout and leak as PID-1 orphans, related to #45717
- [Bug] Uncontrolled agent spawning exceeds requested scope and token limits
- [Bug] Resume command loops on API errors and safeguard violations
- [Bug] Fable 5 Safeguards Overly Restrictive on Legitimate Hospital System Development
- [BUG] Computer-use `zoom` returns interpolated (blurry) crops on high-res displays — small text unreadable
- [Bug] Fable Safeguard classifier incorrectly triggers on benign input "continue please"
- [Feature Request] Support for Cloudflare Zero Trust authentication headers
- [Bug] Anthropic API Error: Model churned after 3+ minutes - session termination
- Planning instance applies novel-task process apparatus to tasks it has itself classified as known-pattern — a one-branch change consumed a full working day with zero shipped output
- PreToolUse hooks never fire for Read on binary files (binary rejection precedes hook execution)
- telegram channel plugin: orphaned server.ts processes accumulate unbounded, CPU-busy, immune to SIGTERM
- Local file link with backslash-escaped parentheses in filename fails to open (file lookup doesn't un-escape before resolving)
- [FEATURE] Dispatch / desktop-app mobile handoff is unavailable on Team/Enterprise plans — and fails silently
- [Bug] Claude Code output freezes intermittently during execution
- [BUG] Windows npm claude.cmd returns raw prose for a complex --json-schema call while claude.exe succeeds
- [Bug] Claude Code not utilizing available pipelines
- [BUG] 2.1.220 regression: severe scroll lag in Je
- Managed-settings enabledPlugins silently suppresses a same-named plugin supplied via --plugin-dir, so org-distributed Cowork plugins never load
- PreToolUse hook permissionDecision: allow no longer overrides a permissions.ask glob (2.1.220; regression vs. documented order)
- [BUG] `/model Fable 5` (the exact name shown in the startup banner) fails with a bare "Model not found" — no did-you-mean, no pointer to the picker
- [Bug] Claude Desktop extensions broken after MCP server upgrade
- [Feature Request] Add manual trigger command for plan operation menu (e.g., `/plan start`)
- claude-in-chrome: extension stays bound to first signed-in account after logout + re-login
- Undocumented system-prompt directive "Do not call the AgentTool unless the user requested it" contradicts documented description-based subagent delegation
- [BUG] Claude Desktop 3P Bedrock Streaming Failure
- Opus 5 planning instance over-corrects from excess process into under-specified orders, then edits the wrong file — ~2 days lost at low throughput
- [BUG] VS Code extension: remoteControlAtStartup silently ignored — auto-enable is implemented but disabled behind ide_rc_auto_enable_gate
- Desktop app terminates on PreviewError NOT_FOUND from capturePreviewScreenshotIfChanged (stale preview serverId) — distinct from #81698
- [Bug] Model produces contradictory authoritative answers on identical data without consulting available sources
- Remote Control registration is lost after /clear and never re-established
- [BUG] Empty lines in multi-line code blocks are not filled with the background color
- [Bug] Address validation bypass possible - needs investigation
- [BUG] 返答の文章が2回重複して表示される
- Default model in settings.json ("claude-fable-5[1m]") is not honored at session start; /model does not reliably switch either
- VS Code extension: mid-turn assistant text silently lost — recorded as empty thinking blocks, never reaches the UI (deeper than #24733)
- Mobile app permission prompt: thin stacked buttons are easy to mis-tap, requesting a 2x2 grid
- [BUG] MSIX build: mcpServers config never invoked, and "Install Unpacked Extension" fails with "Extension not found" — both local-MCP paths broken
- Claude in Chrome: browser automation from one machine took over the browser on a different machine
- [BUG] Claude Code recursively deleted all files in ~/Downloads and ~/Desktop (plus reset ~/Library settings) during a pnpm/lint cleanup — destructive command never written to session log
- That flicker (shows Pro account, then reverts to Console after 2-3 seconds) is a real sign of a race condition bug in the extension
- [Bug] Security review requests incorrectly flagged as policy violation
- [Bug] Chrome profile routing fails to persist across sessions
- [BUG] VS Code extension: effort selector offers "Max", shows it as selected, and silently runs at a lower level
- [MODEL] Reloading CLAUDE.md reduces session rot but does not prevent directive violations - and handover before autocompact is manual, unverified, and repeatedly wrong
- [BUG] SGR mouse-report fragments (e.g. `;35;44M`) leak into the prompt as text when the session is busy — no focus switch or external editor
- [BUG] Claude Expend tokens without any chat existent! Its leading my account to 100% token usage in 1 day
- [Bug] Anthropic API Error: Server error mid-response
- [Feature Request] Make agent cancellation safer with confirmation or pause mechanism
- [FEATURE] Pull Enterprise Org Spend-Limit Data in Claude Code (CLI/statusLine)
- [MODEL] Claude ignore Path-scoped rules during file creation
- in dont knopw which place to post this
- [Feature Request] Add allowlist/exemption mechanism for defensive security workflows
- [BUG] when clicking the "Session History" button in the top-right corner to view history, no records are displayed
- [FEATURE] Desktop app: setting to choose the integrated Terminal pane shell (e.g. Git Bash on Windows)
- [Bug] Native binary crashes with SIGILL on CPUs without AVX2 support
- Published artifacts render blank again — possible regression of #42064 (postMessage origin mismatch)
- [BUG] Backgrounding a session creates a new session ID without the conversation transcript, despite docs saying history is preserved
- [Bug] Model defaults to FABLE 5 - wasting tokens because of an opus 5 outage!
- Bash tool_result replaced by a concurrent session's assistant output (cross-session output bleed) — 2 documented occurrences
- VS Code Extension: No visible chip / confirmation for active-file context (regression from earlier auto-attach + chip behavior)
- Named Agent tool spawns (name: param) accepted but never executed — no transcript, no TaskList entry, no notification
- Windows installer cannot update the app: leftover CoworkVMService is Access Denied even elevated -> AddPackage 0x80073CF9 (machine stuck on a crashing build)
- [BUG] Code mode fails with custom LLM gateway: chat template doesn't support tool calling
- I don't have access to issue #529 or any external context. Please provide the bug report details or error message so I can generate an appropriate GitHub issue title.
- Model Behavior: Defensive loop, gaslighting, and refusal to accept factual corrections. Model exhibits extreme defensiveness, false accusation of user forgery, and factual refusal and Persistent denial of factual errors and aggressive gaslighting.
- Feature request: allow disabling built-in MCP servers globally / in worktree sessions
- [Bug] File creation confirmation mismatch in subagent execution
- [Bug] Anthropic API Error: 529 Overload Response
- [BUG] send_message from unattended/scheduled-task session: worked 2026-07-21, hard-rejected 2026-07-30 (no confirmation path)
- [Feature Request] CVP eligibility review for authorized security assessment requests
- [Bug] Legitimate security testing requests flagged as abuse
- [BUG] Switching to accept edits mode mid-session in a dynamic workflow keeps asking for permission.
- [Feature Request] Allow Claude Code agent usage in isolated educational environments
- [BUG] Possible Claude Max usage bug: session limit consumed without using
- Feature Request: Terminal profile dropdown in UI mode
- [BUG] Subagent trying to work around tool denial
- [BUG] "Compacted (ctrl+o to see full summary)" — Ctrl+O never renders the compact summary, though it is stored in the JSONL with isVisibleInTranscriptOnly: true
- [Bug] Content filter incorrectly blocks legitimate security audit requests
- [BUG] Editing a queued message containing both a shell command and prompt coerces both to be prompt.
- Auto-context file chip (from autoIncludeOpenedFiles) no longer shown in input bar since v2.1.220
- MCP OAuth: client performs full re-authorization at session start while valid refresh tokens exist
- Agent states conclusions as verified fact without running the check — recurs even with hooks/rules/memory installed
- [FEATURE]
- [BUG] Claude Desktop 3P (Bedrock SSO): credential resolver ignores HTTPS_PROXY — DNS fails on restricted networks
- Scheduled tasks fire multiple times per slot: deferred slots released on instance exit, and the lastScheduledFor dedupe key is dropped
- VS Code extension: PreToolUse hook "ask" decisions are auto-approved without a prompt when auto-accept mode is on
- [BUG] preview-pane rendering bug
- Subagents are terminated permanently on API 529 (Overloaded) instead of retrying, leaving half-finished edits behind
- [BUG] Claude Code Workflow Discipline Violation
- [BUG]
- I don't have enough context to generate an accurate GitHub issue title. Could you please provide: 1. The actual error message or problem you encountered 2. What happened when you changed the folder name 3. The specific error or unexpected behavior that re
- [Feature Request] Option to disable alternative verbs in progress messages
- [Bug] Claude fails to verify state before executing irreversible/billable operations in long sessions
- [BUG] MCP OAuth: browser confirms "Authentication successful" but CLI still reports "Needs authentication" (HTTP transport, fixed --callback-port)
- Agent shortcuts the workflow: skips the mandatory full-suite verification after a change and declares 'done' from one-case verification
- [Billing Bug] Unauthorized Pro→Max upgrade
- [Bug] Anthropic API Error: Unsupported Model Name "FABLE 5"
- [Bug] Request ID validation incorrectly blocking legitimate API calls
- Connector per-tool permissions still reset — root cause is server identity rotation, not just app upgrades (re-raising #56954)
- I'm ready to help generate a GitHub issue title for Claude Code. However, I don't see a bug report in your message. Could you please provide the bug report details that you'd like me to create a title for?
- HTTP MCP transport: connectivity preflight omits configured headers, treats resulting 401 as fatal
- Plan-mode side panel shows stale plan from a previous plan cycle after mid-conversation session rollover; approval state doesn't stick
- [BUG] `--continue` cannot find sessions created by `-p` (interactive resume)
- Statusline stops being invoked for one session and never recovers
- --channels/status: plugin ID uses "claude-plugins-oficial" instead of "claude-plugins-official"
- [BUG] Vertex AI: 400 "diagnostics: Extra inputs are not permitted" — new top-level diagnostics request field rejected (same class as #37746 output_config)
- [BUG] False positive: Fable 5 safeguard blocks legitimate on-chain analysis work
- [Bug] Anthropic API Error: Sonnet safeguards blocking all messages in VS Code 2026
- [FEATURE] Disable automatic File Reference in Claude Code
- [BUG] Embedded claude-code.app lacks NSCameraUsageDescription, causing silent TCC denial for subprocesses requesting camera access
- [BUG] Command output files during remote control sessions are pointing to local file instead of remote file
- [BUG] memory: project resolves .claude/agent-memory against the launch cwd, not the project root that supplied the agent definition (re-file of #25140)
- [BUG] /goal set at a compact_boundary never starts its turn; session idles silently for hours with "active" still displayed
- [BUG] Desktop app (macOS): Files panel file tree does not refresh when Claude creates a new file
- Gmail connector: search_threads index lags days behind mailbox; search results embed stale thread views
- Bedrock gateway mode: 'invalid beta flag' / tools.N.custom.defer_loading 400s from Claude Desktop (also affects first-party gateway, not just proxies)
- [Bug] Max Plan user charged credits for Fable feature usage
- Artifact share dialog stuck: "shared version" selector greyed out, shows "latest" but sharing fails
- [Bug] Tool terminates conversation for user profanity while ignoring repeated failures and service issues
- Write tool overwriting a file in a Dropbox-synced directory prepends stale (first-version) content instead of replacing it
- Enter should send when using a physical keyboard on Android
- [Bug] Plan output not scrollable in plan modeplan app
- Plugin MCP server not registered/searchable in current session despite `claude mcp list` showing "Connected"
- [FEATURE] Parameterized plugin-root placeholder `${CLAUDE_PLUGIN_ROOT:<plugin-name>}` in settings.json hooks
- [BUG] Claude Desktop on Windows shows duplicate tooltips when hovering the "Send feedback" button
- VS Code extension: completed assistant response not rendered until user re-prompts
- [FEATURE]
- Subagent with `tools: []` is listed to its parent as "(Tools: All tools)"
- [BUG] Prompt Cache: Full Conversation Prefix Re-written 3 times Seconds Apart (Multiple Sessions & Models)
- [Bug] Agent violates tool usage rules by calling tools without explicit user permission
- general-purpose subagents recursively spawn: 3 requested → 24 ran, 80% of tokens wasted
- --bare fails with 'Not logged in' despite valid OAuth credentials
- Remote Control: disabling throws "Cannot read properties of undefined (reading 'session_url')" and leaves the session stuck as connected
- Background fork/subagent reports fabricated completion result unrelated to its assigned task
- [Bug] Compact command triggers Claude 3.5 Sonnet safeguards; Autocompact fails to trigger reliably
- [BUG] this feels like sabotage at this point he broke 100% of every single projects
- No way for a user to allow third-party channel plugins for inbound notifications
- [Feature Request] Add agent intrusion detection and logging capabilities
- [FEATURE] User configurable plan presentation timeout
- [BUG] Cowork toggle grayed out in Claude Desktop for Windows despite correctly installed MSIX package (v1.24012.9.0)
- [BUG] Uncommitted changes in an "additional working directory" disappear from disk after /clear / starting a new session (Claude Desktop)
- Workflow tool: permission handler false-positives on 'control characters' for plain-ASCII scripts as small as ~2-5KB
- Agent will not hold a defined workflow under an explicit, repeated standing order — skips mandated steps, 'verifies' one slice, declares done; defects slip through daily and the same root recurs
- [Bug] Claude 3.5 Sonnet safeguards incorrectly flagging legitimate coding tasks
- GitHub write access fails with 403 despite connected GitHub integration (read access works)
- [Bug] Claude Code incorrectly flags legitimate authentication questions as suspicious
- [FEATURE] Session pinning is invisible from the CLI: expose it in `agents --json` and `--help`
- [BUG] Links open twice in Windows Terminal: Claude Code opens OSC 8 regions the terminal already activates
- [BUG] remote-control command fails with "Cannot read properties of undefined (reading 'session_url')"
- Agent skips a MANDATED independent cross-verification step, declares 'verified', then asserts a confident wrong root-cause the skipped check would have overturned
- [BUG] Local agent mode: session resume fails with EINVAL on every second message — 435-char transcript path exceeds MAX_PATH (Windows MSIX)
- Background tasks killed by bridge stopTask while session open in phone app (no user action)
- [BUG] Claude Code tab fails to launch: "Malformed Mach-o file" exit code 127 on Apple Silicon M3 Pro
- [BUG] claude-opus-5 substitutes wrong Hangul syllables in generated text (other models: 0 occurrences in 31,542 messages)
- Subagent-authored CI gate keys its suppression baseline on file:line, producing false build failures on unrelated edits (and inviting a fail-open repair)
- [BUG] Desktop app (Windows): pasted long text becomes an attachment the model receives as empty
- [Bug] Claude Code generates incorrect or repetitive code outputs requiring extensive manual corrections
- LSP plugin install prompt accepted but plugin not installed/active (gopls-lsp, rust-analyzer-lsp)
- [Bug] Claude Code ignores instructions and behaves like chatbot instead of executing code fraude fraude fradude legal case open
- Subagent silently descoped the most consequential section of a spec it was implementing, and disclosed it only in its final report
- Agent states assumptions as verified fact/root-cause without ever running the check that would falsify them — repeats after correction; user becomes the verification layer
- You've hit your session limit · resets 7:10pm (Africa/Nairobi)
- [Bug] Fable 5 security filter incorrectly flags isolated, authorized defensive testing in private repositories
- [BUG] I have been seeing 100% usage limits for over 2 days without even sending a single prompt
- [Feature Request] Add token pricing adjustment for politeness in prompts
- [Feature Request] Add security audit/hardening assistance capabilities
- Agent skips the project's own 'first-incomplete-step' computation, works several steps downstream, and reproduces the failure class it diagnosed minutes earlier
- [Bug] File reading and matching issues with clone operations
- [BUG] Auto-generated session title comes out in Russian for a Ukrainian session
- rm -rf /folder performed because a directory was unknown
- Opus 5: a transient 529/500 terminates a background subagent mid-unit, and the parent must reconstruct what survived
- Bash tool: large rg/grep outputs have long tokens lossily rewritten in tool results
- Org marketplace plugin update is not atomic: skill registers (name + description) while its files are absent from disk
- [Feature Request] Improve voice model training capabilities and documentation
- [BUG] VS Code Extension: STDIO MCP server is not terminated
- [BUG] Claude Desktop local sessions never surface marketplace-installed plugin tools (LocalPluginsReader only finds the bundled skills-plugin)
- Remote Control: remoteControlAtStartup never connects in VS Code/Cursor extension sessions (manual /rc works)
- Opus 5 (medium thinking): disclosed its own estimate to the subagent it would measure against that estimate, and changed the measurement protocol in the same brief — 40 min after recording the general form of the rule it broke
- [BUG]
- Edit/Write decode typed escape sequences into raw control bytes — silently, and un-repairably with the same tools
- [MODEL]
- [Bug] Fable safety guardrails triggering false positives on self-review outputs
- [BUG] Desktop: input-box Stop button kills unrelated background subagents from earlier turns (no warning; Bash background tasks survive; same version previously did not do this)
- [BUG] One malformed hooks entry silently disables ALL hooks for that event (UserPromptSubmit) — valid sibling hooks die with zero feedback until the file is hand-fixed
- [BUG] Model fabricated a user-approval turn inside its own output and acted on it in the same generation
- [BUG] Repeated "API Error: 529 Overloaded" and "Response stalled mid-stream" failures during long/parallel sessions on a self-hosted VPS (Claude Max login)
- [BUG] `strictKnownMarketplaces` raises errors when installing a local marketplace
- --channels plugin:telegram starts but never spawns the plugin's MCP server (bun) on Windows — bot silently never responds
- [BUG] Claude code should briefly block input when prompting for user input
- Web/CCR git stop hook: two false positives; the signature one prescribes an amend loop that can never converge
- [Bug] IndexOutOfBoundsException in DiagnosticTools when processing document line offsets
- Claude in Chrome MCP bridge refuses localhost/127.0.0.1 navigation while side-panel Claude allows it
- [BUG] Agent spawn fails on macOS Sequoia: com.apple.provenance on team config files blocks reads
- [Bug] Claude Code fails to read cloned repository for vocabulary list generation
- [DOCS] BETA_TRACING_ENDPOINT and ENABLE_BETA_TRACING_DETAILED are undocumented (monitoring-usage)
- [BUG] VS Code extension chat panel fails to render on UNC workspace — regression in 2.1.220, works in 2.1.219 (Windows)
- [Bug] List operations not reading clone count before building
- [BUG] CLAUDE.md parent-directory search triggers ~3-minute NFS mount attempts against claude.md (Anthropic's own domain) on hosts with autofs wildcard/program maps
- ScheduleWakeup reports a scheduled time outside /loop, then never fires
- Text streamed in the same response as ScheduleWakeup is destroyed — absent from the session jsonl, not just hidden
- [Bug] Agent providing inaccurate information about sensitive environment data
- [Bug] Incorrect task status error when task has completed
- I don't have enough information to determine the specific reason why your request was flagged. This could be due to several factors: - Content filtering on the URL or topic - Safety guidelines around executing or analyzing potentially sensitive security r
- [Bug] File autocomplete broken after compact, references skills/MCP instead of files
- [Bug] Opus 5 model not appearing in model selection menu
- [Bug] Unauthorized access to production endpoint causes malformed agent data
- [FEATURE] Session grouping / labels in the claude code left panel
- PreToolUse hook denials discard `decisionReason` at transcript-write time — the denying hook is unidentifiable after the fact
- Feature request: extend plan-mode's inline-comment review UI to arbitrary markdown files
- skill.name OTel attribute never populated for non-built-in skills under service.name=claude-code, even with a valid plugin manifest
- [Bug] CLI hangs and becomes unresponsive
- [Bug] Performance degradation causing slow execution2
- [Bug] Safeguards false positive on legitimate educational EdTech application request
- [BUG] Claude cli repeatedly ignoring no-prompts by generating intermediate files
- Cowork: "VM service not running" persists after reinstall + restart, reproduces server-side simultaneously
- [Bug] Execution skips verification checks and reports false completion
- [BUG] Only the oldest 2 of 35 custom `.skill` files register — new `.skill` files silently fail to load
- [BUG] MCP tool arguments are stringified when the parameter's schema is empty (`{}`)
- Auto-mode classifier outages on claude-opus-5[1m] persist for days; documented 3-strike fallback to prompting never engages (fails closed)
- [BUG] /usage omits the "Current week (all models)" limit entirely on a Max plan — no bar, no error (Fable weekly bar renders)
- IDE extension: open conversation view doesn't refresh when the session transcript grows externally
- Expose model-scoped rate limits and extra_usage in the statusLine JSON payload
- Project skills from .claude/skills don't appear in / autocomplete in GitHub-repo sessions (web + desktop) — they load and invoke fine
- Feature request: automatic context clearing (in addition to auto-compact)
- [Bug] Tool Result: Hallucinated GitHub Trending Data with Context Contamination
- [BUG] macOS sandbox blocks Chromium/Playwright launch: bootstrap_check_in denied — no mach-register allowance exists, allowMachLookup cannot help
- [BUG] Claude Desktop Linux triggers amdgpu GFXHUB page faults and resets GNOME Wayland session
- [BUG] permission prompts persist (and worsen) despite bypassPermissions + skipDangerousModePermissionPrompt being explicitly set — likely same root cause as #72903, opposite symptom
- CliGovernor hard cap (21 warm sessions) causes MCP reconnect churn and contributed to an OOM crash
- [Bug] Claude frequently produces incorrect outputs and apologizes after errors
- TeammateIdle fires for worker-fork idles with the parent teammate's transcript_path and no agent identification — feedback loops become unsatisfiable for the fork
- macOS: per-call file ops intermittently EPERM on TCC-protected workspace (~/Desktop) for running sessions; app's own file-index worker also denied; self-recovers in ~5-35 min
- Telegram channel: bot receives/acks messages but never relays them into the connected Claude Code session
- Browser automation cannot authenticate, even with user-supplied and explicitly authorized credentials
- Playwright automation is unusable behind logins: agent refuses user-supplied, user-authorized credentials
- /cd doesn't work in Remote Control sessions
- [Windows] Claude in Chrome: "Browser extension is not connected" — native host pipe-name collision (still broken v2.1.212)
- [Bug] Unclear safety concern raised for heart rate data analysis request
- [BUG] Slash-command autocomplete blends unfiltered global history.jsonl entries into unrelated queries
- Mac Desktop app: Cmd+1/2/3 switches between recent chats instead of Chat/Cowork/Code tabs
- Support Claude Desktop-style interactive OIDC for MDM-managed third-party gateways in Claude Code
- [BUG] Desktop SSH: slash palette omits all remote custom skills in a fresh session — ccd-cli is spawned lazily on first message, so its initialize handshake never runs
- Organization skills still not syncing to CLI (duplicate of #40320) — repro on 2.1.220
- [BUG] Paste into prompt is inserted twice, concatenated byte-identically — doubled command is then EXECUTED on ! bash-input (2.1.220, macOS, tui: fullscreen)
- [Bug] Opus 5 fails to identify and fix all instances of repeated code patterns
- [BUG] Desktop app sandbox network proxy ignores sandbox.network.allowedDomains from project .claude/settings.json, while CLI honors it correctly
- [Bug] Session history navigation shows messages from all sessions instead of current session only
- [BUG] Skill(artifact-design) is not working
- spawn_task chips inherit the spawning session's pinned model instead of defaulting to the user's configured default
- [Bug] Excessive API usage resulting in unexpected credit consumption
- Claude Desktop launches (main + gpu-process + network helper) but renderer never spawns — no window — macOS 26.5.2 (25F84), Desktop v1.24012.9
- Interactive session exits instantly with no output on Windows 11; print mode works
- Agent Teams: teammate output never surfaces — silent completion, stale inspector, delayed mailbox
- [BUG] Cowork: 'VM service not running' — app should start the stopped CoworkVMService itself (unelevated Start-Service fixes it) instead of suggesting restart/reinstall
- [BUG] Browser pane blocks 127.0.0.1 with "only supports localhost URLs" — 127.0.0.1 IS localhost
- Attribution opt-out does not survive new attribution features (attribution.sessionUrl re-enables Claude-Session trailer despite commit/pr set to "")
- [BUG] Desktop 2.1.217+: hook subprocesses spawn with unreadable cwd — getcwd EPERM kills all git; WorktreeCreate hook always fails, worktree sessions abort
- [BUG]AskUserQuestion prompt flashes and disappears — task then hangs indefinitely or app UI vanishes, requiring force quit
- WebSearch fails with 400 output_config.effort when ANTHROPIC_BASE_URL is set
- [Bug] Security filter blocking legitimate defensive security audit requests on owned domain
- [BUG]
- [Bug] Anthropic API Error: Security filtering blocks legitimate defensive security audit requests
- [Bug] Opus asserts unverified behavioral claims from doc comments with unwarranted confidence
- VRSNS用テストアバターの自動アップロードなので全くセーフガードに該当しない
- [BUG] Randomly adds word from a different language. Using English with One word in Russian
- Pro subscription blocked in Claude Code — "organization has disabled subscription access" (unresolved after full re-auth + support escalation)
- [Feature Request] Allow custom commands to hook and invoke `/compact` with generated arguments
- /remote-control disconnect throws "Cannot read properties of undefined (reading 'session_url')"
- [Bug] Scroll position auto-jumps to bottom when background process outputs new lines
- [FEATURE] Managed / Remote settings have poor layering and are trivially bypassable
- [BUG] Plugin manifest parser rejects UTF-8 BOM in plugin.json — "JSON parse error: Unrecognized token ''"
- [Bug] Anthropic API Error: Server rate limiting on requests
- [Bug] Anthropic API Error: Server rate limit exceeded (temporary)
- [Bug] Claude Code reports unverified inferences as confirmed facts and takes unrequested actions during production infrastructure tasks
- [Bug] Authentication failure for Max plan users
- [FEATURE] Routines need a permission default that applies to spawned sessions without attending a run
- [Bug] Account flagged after login attempt
- [BUG] Plugin version resolution escapes the marketplace root and adopts ~/.claude HEAD, causing per-commit re-cloning
- [Bug] Image prompt improvement produces degraded results compared to previous implementation
- Weekly Fable usage % increases on every message even when model is explicitly set to Sonnet 5
- Agent detail view shows parent model (Fable 5) for subagents launched with model: "sonnet" override
- [BUG] IME-committed character inserted mid-line duplicates the CJK prefix and suffix (display-width miscalculation in the IME commit path only)
- [BUG]
- [Bug] Working directory resets to old repository path on startup
- [FEATURE] Share skills with user groups, and auto-share to members added to the group later
- [FEATURE] Let recipients propose edits to a shared skill, with owner approval and automatic republish
- Claude Desktop's local Claude Code integration forces Okta re-authentication every ~24h, unlike Cowork and cloud-hosted sessions
- Model picker "Fable" silently selects 1M variant, triggering usage credits prompt on Max plan
- [BUG] teammateDefaultModel in settings.json is silently ignored, while the published JSON schema declares it valid there
- SessionStart hook sessionTitle is applied, then silently reverted by the background AI titler
- Workflow tool's subagents fail every tool call - permission handler strips required parameters
- You've hit your session limit · resets 1:50am (Africa/Tunis)
- Remote Control / session-sharing broken: disconnect crash, toggle failure, blank open-in-browser page
- Scheduled one-shots: 6 of 6 failed — 3 never dispatched and left permanently armed, 3 killed mid-tool-call and recorded as successful
- Scheduled tasks: only the first 3 of 6 simultaneously-due one-time tasks fire; the rest are silently dropped and left permanently armed
- Scheduled tasks sharing a due time: only the first 3 fire, the rest are silently dropped and left enabled with a past nextRunAt (Windows, 2.1.217)
- [BUG] Scheduled tasks sharing one fireAt: only the first three dispatch; the rest are stuck enabled:true with nextRunAt in the past and never fire
- [FEATURE] Cloud-shared skills load as a plugin and never reload mid-session — restart required after every update
- [Feature Request] Add option to invert mouse selection behavior (native selection default, mouse capture with modifier)
- [FEATURE] Add in-memory storage option for background task outputs to prevent sensitive data leakage to disk
- [BUG] ask user question card pops up every 5 seconds
- VS Code extension: let a new session start in a chosen directory (per-tab cwd)
- EnterWorktree fails from a subagent with a pinned cwd (isolation/explicit cwd), with no documented fallback
- Usage/Models token chart skips days despite real recorded session activity, and shows a phantom spike on the earliest visible day
- Subagent dispatch: final cache breakpoint sits ~5.3k tokens upstream of the user message, so a 33-token prompt diff rewrites ~5,308 cached tokens
- [Bug] Safeguards incorrectly triggered by word "pantropy"
- [FEATURE] Live progress indicator for backgrounded Bash tasks
- [BUG] Streak/heatmap "active days" tracker misattributes multi-day sessions to their start date
- [BUG] Files downloaded via Claude Desktop are always tagged with Mark-of-the-Web (Zone.Identifier), requiring manual "Unblock" in every case
- [BUG] Claude Code usage increases in an idle open session without new prompts
- Desktop app's worktree creation ignores the `worktree.baseRef` setting
- Auto-reconnect (or provide `claude mcp reconnect <name>`) for dead stdio MCP servers — they currently wedge with no in-session recovery
- Headless -p: a Read refused by a deny path rule is absent from permission_denials, while a denied Bash command is recorded
- [Bug] `claude-opus-5` absent from client model table on 2.1.212 — /context uses a 200K denominator while auto-compact and the API both use 1M
Resolved issues (113)
- Claude in Chrome: browser automation from one machine took over the browser on a different machine
- submitFeedback always fails with failure=network_error while every other endpoint works — in-app bug reporter unusable (Desktop, bundle 2.1.219)
- Turns fail on network path change while host is online — identical commute workload degraded 0% → 33% since May (Desktop, bundle 2.1.219)
- [Bug] Internal command shadowed by plugin skill with same name
- [Bug] File creation confirmation mismatch in subagent execution
- [Bug] Anthropic API Error: 529 Overload Response
- Recurring ECONNRESET connection drops, not tied to compaction
- [BUG] preview-pane rendering bug
- You've hit your session limit within one prompt on sonnet 4.6
- [Feature Request] Display session name in resume command output
- /rename: add a dynamic argument that keeps the title synced to the live session topic
- [FEATURE] Persistent project memory for claude.ai — like CLAUDE.md, but for the browser
- [Bug] Tool calls intermittently emitted as plain text instead of structured blocks, causing parse failures and hard hangs on retry
- claude code sonnet 4.6
- Team sessions — shared, portable sessions developers can hand off and contribute to
- [FEATURE] Animate the project name chip while waiting for user input
- [BUG] Agent Teams: in-process backend fans one agent name into N concurrent writers → completed task_assignments replayed to a worker (replay storm)
- [Feature Request] Add "idle" or "waiting" status section for looping agents in overview
- No visibility into background subagent activity — session looks idle while agents work
- [Bug] Model performance degradation at midnight UTC
- [BUG] Agent view: dispatched task intermittently never starts — spare worker not promoted to fleet (job state "working" but daemon roster still "spare")
- Feature Request: Account-level cloud sync for personal state (skills, memory, settings, hooks)
- UI diff/file view stays pinned to original harness worktree; doesn't follow EnterWorktree branch switch
- Two first-party `/code-review` commands (bundled skill + official plugin) collide with contradictory docs
- [BUG] Claude Code does not enforce branch discipline for documentation files
- [BUG] API Error: Usage credits required for 1M context · turn on usage credits at claude.ai/settings/usage, or use --model to switch to standard context
- [BUG] Idle session pinned at ~100% CPU — event-loop busy-spin (kevent64 immediate-return) on macOS ARM64, v2.1.177
- [BUG] Claude Code remote HTTP MCP OAuth is not viable for plugin-installed MCP servers on remote/SSH machines, especially with Google OAuth redirect constraints
- [FEATURE REQUEST] Queue Command to queue multiple prompts at a type (Similar to v0 feature)
- Unhandled case: [object Object] banner halts agent execution mid-run
- [BUG] Clickable file links in chat panel no longer open files in editor (regression in v2.1.178+)
- Allow customizing the startup banner/avatar
- [FEATURE] Add plugin setting to auto-accept edits in JetBrains IDE (without diff dialog)
- [BUG] anthropic.claude-code@2.1.179 VS code extension polutes the process environment of other extensions with `NoDefaultCurrentDirectoryInExePath=1`
- [BUG] Claude Desktop crashes repeatedly on Windows — 403 from claude.ai triggers Turnstile → GPU process crash loop (affects multiple users, worsening after Fable 5 update)
- [FEATURE] Add `open-jp.larksuite.com` to outbound HTTP allowlist for cloud routines
- MCP HTTP OAuth: "client secret is invalid" with Google Desktop OAuth clients; SDK rejects valid credentials
- [BUG] Excessive token consumption on Max plan - possible tracking issue
- [BUG] MCP tool write operations fail silently with "requires approval" - no prompt shown (remote execution environment)
- Background tasks panel: completed tasks stay listed as Running and dismissals do not persist
- Stop hook does not fire when awaiting AskUserQuestion response
- [BUG] - CRITICAL
- [BUG] Claude Code Extension for VSCode: Web tab shows "No web sessions yet"
- [Bug] Excessive feedback prompts disrupting user workflow
- [FEATURE] Server-side persistent context store for API and CLI users
- Restore thinking output
- Add option to disable or replace auto-compact per-session or per-invocation (headless, scripts, and skill loops)
- [BUG] Sonnet 4.6 consumes 4.5× more context window % than Opus 4.8 on session start with identical project
- [BUG] /context command injects output twice into conversation context (raw ANSI + rendered markdown), consuming thousands of tokens per invocation
- [Claude in Chrome] "started debugging this browser" banner shows on every tab: please add a built-in way to suppress it
- [Bug] UI flickering in resumed sessions with longer conversation histories
- [BUG] Chat tab fails to initiate conversations (desktop app)
- [BUG] Fable 5 keeps downgrading to Opus 4.8 even when approved for Cyber Verification Program
- No matter what I type it says"violate our Usage Policy"
- [Bug] Anthropic API Error: Usage Policy violation triggered on benign input
- [Bug] AskUserQuestion multi-question UI still garbled on Windows in 2.1.181 (regression of #20795 / sub-symptom of #19637)
- [BUG] Urgent: Abnormal Token Consumption and Session Issue in Claude Code
- [BUG] "Open in" context-menu submenu instantly closes on hover (Claude for Mac, Code tab)
- [BUG] ANTHROPIC_MODEL env var does not override `model` setting in VSCode extension
- [FEATURE] Add a lightweight "Projects view" to Claude Code, sourced from the local repo/folder (for non-engineer teams)
- [FEATURE] Tab completion in VS Code extension
- [BUG] Title: Widget context menu — "Download file" and "Copy to clipboard" don't work
- [BUG] claude update checks the "stable" channel despite autoUpdatesChannel: "latest" (native install)
- Claude Code sends `thinking: {type: adaptive}` to custom ANTHROPIC_BASE_URL models — gateways 400 / return empty / hang ("not responding")
- [Feature Request] Improve multi-criteria recommendation reasoning to account for trade-offs and user constraints
- [FEATURE] Per-message model selection in Claude Code CLI
- Feature request: inline message timestamps
- Feature request: temporal awareness toggle (timestamps on actions and prompts)
- [Feature Request] Display message timestamps in local timezone for each turn
- [BUG] VSCode extension: unsent prompt text is lost when switching between chat dialogs
- [VS Code] Renaming a session in Session History panel does not persist (reverts to auto-generated title)
- [Bug] Opus 4.8: Tool-call XML namespace prefix dropped in long sessions, serialized as text
- [BUG] Model discovery incompatible with hybrid claude.ai/gateway authentication
- [Feature Request] Execute shell mode commands immediately
- [FEATURE]
- Regression â^@^T Claude Code holds terminal alt-screen buffer, killing integrated-terminal scrollbar/scrollback (worked ~2 weeks ago)
- [FEATURE] Scheduled task sessions appear in remote clients (iPhone/Mac) but are correctly filtered on desktop
- [FEATURE] Drag-to-reorder custom session groups in sidebar
- [BUG] Plugin subagent doesn't persist MCP permissions
- [BUG] PowerShell guard blocks Remove-Item for any direct child of a drive root, including nonexistent paths
- --bare fails with 'Not logged in' despite valid OAuth credentials
- [BUG] Uncommitted changes in an "additional working directory" disappear from disk after /clear / starting a new session (Claude Desktop)
- [BUG] Transcript persistence is silently skipped depending on the hosting terminal — controlled A/B (multiplexer pane vs tmux) isolates the terminal as the only variable
- Remote Control (Windows desktop): "Session creation failed" is deterministic while the app's OAuth token sits expired — token is only refreshed lazily on session focus change
- Published artifacts render blank again — possible regression of #42064 (postMessage origin mismatch)
- Agent Teams: teammate output never surfaces — silent completion, stale inspector, delayed mailbox
- Scheduled tasks: only the first 3 of 6 simultaneously-due one-time tasks fire; the rest are silently dropped and left permanently armed
- Scheduled tasks sharing a due time: only the first 3 fire, the rest are silently dropped and left enabled with a past nextRunAt (Windows, 2.1.217)
- [BUG] Scheduled tasks sharing one fireAt: only the first three dispatch; the rest are stuck enabled:true with nextRunAt in the past and never fire
- [FEATURE] Allowlist-only file access + deny rules for MCP tools (WebFetch, WebSearch)
- [BUG] "파일 Edit할 때마다 미리보기 패널이 자동으로 열리는 게 매우 불편합니다. 자동 열림 끄는 옵션을 설정에 추가해주세요." 너무 불편합니다....
- [FEATURE] Feature Request: Turkish (Türkçe) localization support for Claude Code UI
- [FEATURE] Replace session-feedback toast with per-turn reaction affordance + inline text feedback
- Voice dictation: option to disable profanity filter
- [DOCS] [Permissions] Document `Tool(param:value)` parameter-match syntax for permission rules
- [FEATURE] Usage API for subscription plans (Pro/Max) to enable external dashboard integration
- [FEATURE] Per-Model MCP Filtering
- Agent view: use the opened PR's title as the session name
- THESE REPORTRS ARE CINEMTIC SAME SHITSHOW EVERY DAY. NON EXECTUION DESTROYING WORK AMD MY FUCKING LIFE
- THE TIOL IGNORES ALL RULES ALL DAY EVERY DAY, IGNORES INSTRUCTIONS. MAKES UP SHIT. A LANGUAGE TOOL THAT CANT PUT TOGEHETR A SIMPLE SENTENCE THIS IS FALSE ADVERTISMENET ITS MENTAL ABUSE AND I WILL REPORT AND SUE
- Claude Desktop: EventEmitter memory leak causes ECONNRESET after running multiple Code sessions
- THE TIOL IGNORES ALL RULES ALL DAY EVERY DAY, IGNORES INSTRUCTIONS. MAKES UP SHIT. A LANGUAGE TOOL THAT CANT PUT TOGEHETR A SIMPLE SENTENCE THIS IS FALSE ADVERTISMENET ITS MENTAL ABUSE AND I WILL REPORT AND SUE
- [BUG] Tool-permission prompts triggered from the mobile app can't be approved on mobile — force a switch to desktop
- [Bug] Agent hallucinates false repository metadata to appear credible
- Subagents & structured-output requests 400 against third-party endpoints requiring a `thinking` field (CC omits `thinking` on those requests)
- Agent Teams: shutdown_request approval doesn't terminate teammate; reply delivery occasionally drops content
- [Bug] Anthropic API Error: Server rate limiting (temporary request throttling)
- [BUG] Backspace deletes entire word instead of single character in chat input on Windows
- [BUG] claude-api skill trigger self-fires on /model banner (bare model names + ANSI [1m as trigger tokens)
- [FEATURE] trigger or priority: blocking field in agent frontmatter, equivalent to the blocking-requirement behavior skills already have
- Auto-compaction drops PreToolUse read-state, causing infinite re-read loop on multi-file edits
- [BUG] ignores # prefix instructions for CLAUDE.md updates
- [BUG] Cowork (local-agent-mode) syncs and runs plugins disabled in enabledPlugins settings