Digest / September 4, 2026
Claude Code Issues Digest — September 4, 2026
303 new issues filed ·
688 resolved in anthropics/claude-code (UTC day).
New issues (303)
- [BUG] Linux Desktop: backend NodeService utility process aborts (SIGABRT) — event-loop stalls compound into whole-app crash-loop
- Ultracode mode was reported ON without the user enabling it; model spawned 160 subagents and exhausted the 5-hour session limit
- Dispatch: "Another Claude Code session is already active in this directory" names no session; plus a stale scheduled_tasks.lock whose own PID-reuse guard is never checked
- [Bug] India missing from country list
- Session list: no way to delete/close a session, only "Archive" (X icon removed)
- [Bug] Indian postal code 622412 not recognized by weather API integration
- [Bug] Bash tool dumps entire files instead of reading specified ranges
- [Bug] Safeguards incorrectly flag legitimate cybersecurity task, triggering unwanted model switch
- Model behavior (Fable 5.1): design request answered bottom-up from existing code instead of from requirements
- [Bug] Tool execution resulted in unintended multiple charges via Stripe integration
- [Bug] Random Q/A message flagged during decompiler task execution
- [FEATURE] context management/compaction improvement
- Auto-mode classifier blocks user-authorized deploys and compounds denials within a session
- [Feature Request] Reduce false positive safety flags for basic, non-harmful code
- [Bug] Safeguard Check False Positive on Unrelated Prompts
- [BUG] Auto mode instructs Bash reads (cat/sed -n) while Edit requires a Read-tool read, so every auto-mode edit fails once by design
- [BUG] Skill `args` silently rewrites literal dollar figures in SKILL.md prose — 22/27 skills exposed, 0/27 use argument placeholders
- MCP streamable-http: a slow tool call's result lands in a sibling call's slot in a concurrent batch
- Claude in Chrome extension shows installed but tabs_context reports "not connected" (Cowork session)
- [FEATURE] Allow /btw (side chat) while an AskUserQuestion picker is displayed, so options can be clarified without dismissing it
- حل الاشكال
- Agent-run paid batch script kept spending after external account lockout, no circuit breaker
- Provide an export/handover path before dropping support for a machine
- Remote Control: session permanently wedged when worker dies unregistered (phantom worker_status:running); still reproduces on 2.1.259 (re-file of #76530)
- [BUG] Browser tools stuck at "No tab available" after hitting usage limit — only fixed by starting a new session
- Show friendly message when opening sessions created by another client
- "Always allow" permission checkbox does not persist in Chrome (claude.ai)
- [Bug] Claude Code ignores user questions and always prompts for plan approval instead of answering
- Claude in Chrome: permission and credential consent prompts render in a window the user cannot see, so they lapse as "Permission denied by user" / transport_error
- [BUG] ECONNRESET on all requests — curl, node https, and node fetch all succeed to api.anthropic.com on the same machine
- Prompt cache never hits across chained -p --resume calls, even at minimum config
- [Claude Desktop / Windows] Enter inserts newline instead of submitting in the Code tab (works in Chat tab and in the same bundled CLI standalone)
- [BUG] Desktop (Windows): worktree session fails — background full checkout dies with Directory not empty on the first top-level path; every stage1=(2 paths) run fails
- [Bug] False Positive Security Flagging for P2P Networking Code
- [BUG] Side chat (Cmd+;) stops responding after the first reply — follow-ups produce no output
- [BUG] Built-in browser prompts per site even in "Skip all approvals" mode with "Allow all sites" set
- [Bug] Incorrect C compiler error classification as security threat
- [Bug] Plan Mode allows Write to modify files without user approval
- [Bug] Bypass functionality not working as expected
- Forced re-login wipes ~/.claude transcripts; concurrent instance deletes a live session's task output files (Windows)
- [BUG] Very garbled text
- [BUG] Cross-session message delivered to a stream-json session emits no user frame on stdout — SDK hosts render a reply with no prompt
- [MODEL] Stated unverified claims as fact (payment status, filing deadline rule)
- Autocompact fires once then never again on opus[1m]; never fires for agent-team teammates (200K haiku teammates compact fine)
- Code tab: project list sorts oldest-first, inconsistent with Chat/Cowork thread order
- [Bug] 3D model rendering not displaying in output
- claude -p headless session hangs at startup when a version-directory lock is contended by a concurrent interactive claude session
- [BUG] Windows MSIX: Claude Desktop stops launching until a full OS restart; activation succeeds but no window ever appears
- [Bug] Significant degradation in response quality and task performance
- Desktop Browser pane: Cloudflare challenge page crashes GPU process, app restarts (exit 101457950)
- Remote Control: "New session" on claude.ai/code attaches to the environment's most recent session instead of creating a new one
- [FEATURE] Scope Remote Control per session type — reopen: scheduled/Routine sessions still auto-enable it (see #91839)
- Desktop: New session opens an existing bridge session's thread when the bundled CLI hot-swaps (required_version change) at the click
- SSH remote sessions render empty - client-side mirror transcript is created with 0 bytes
- Read() deny rule on a root file makes every `grep -r .` prompt under bypassPermissions (2.1.257 reader-command rule, untouched by the 2.1.260 revert)
- Allow configuring how background task notifications render in the terminal
- [BUG] Desktop app Code tab: CLI-created sessions not visible, "Session not found on disk" + "Import CLI sessions" fails even in trusted folder
- Claude Desktop (Windows ARM64): main window gets WS_EX_TOPMOST (always-on-top) on every launch, covers all other apps
- [BUG] Windows 11 Pro: Cowork sessions cannot create or edit local (file-reading) scheduled tasks - reserved "Scheduled" folder is unreachable
- Auto-reap or bulk-clear completed subagents from the agent roster instead of manual per-entry clearing
- [BUG] Agents in .claude/agents/ not auto-loaded without restart (Windows, Desktop app)
- [Bug] Subagent effort level silently capped when extended thinking disabled no clear indication this happens, wastes tokens and usage window.
- [BUG] Claude Code VS Code Plugin freezes and lags with high-reasoning models; hooks fire before rendering catches up
- [BUG] Auto-continue at usage-limit reset is offered in only one session — other concurrent macOS-app sessions stay parked until manually retried
- [BUG] Claude desktop app becomes unresponsive after being idle, then fails to relaunch with "another program already using it"
- Scheduled routines prompt for connector tool approval; "Allow for all scheduled runs" errors instead of persisting
- /model opusplan fails with "Unsupported model" - worked for months until today
- [BUG] [Cowork] mcp__workspace__bash sandbox VM still fails to start — 2-month persistent outage (ref #76374)
- VS Code extension: session history panel stops indexing new conversations (history.jsonl desyncs from actual .jsonl transcripts)
- [BUG] Desktop app: Remote Control session is paused by idle timeout ~15 min after a stealth-update relaunch (RC hold not restored)
- [Bug] Anthropic API Infrastructure Reliability Issues Causing Service Disruptions
- [MODEL] Procedural instructions decay over long sessions — steps after the "completion-shaped" action get skipped despite being read and quoted
- [Feature Request] Restore Opus context awareness and reasoning quality from two weeks ago
- [BUG] Claude Desktop - MCP tool call hangs silently if "Allow once" is clicked before Claude finishes generating the call
- [Bug] Agent performs unsolicited tasks without user confirmation, wasting tokens
- [BUG] Claude Desktop (Code tab) auto-denies CLI-native SendMessage, breaking subagent resumption; the desktop replacement only covers session-to-session
- plugin eval: scope `target: trace` to the graded run's own events
- Read tool on images intermittently returns only a CDN URL string — image never reaches model context
- [BUG] Worktree session creation fails on Windows since 2.1.260: background full checkout forces checkout.workers=8
- [Bug] Anthropic API Error: Overly aggressive cyber safeguard classification blocking legitimate vulnerability research requests
- [BUG] Shifted keys lost in WezTerm since 2.1.247: Kitty "report alternate keys" flag is requested but the shifted-key field is never parsed
- Feature request: user-defined session metadata (description, status, tags) — filterable in the history picker
- [Bug] Failed to find Server Action in Next.js deployment mismatch
- [OPUS 5] TOTALLY BUGGY
- [Bug] Anthropic API Error: Excessive refusals on legitimate content translation tasks
- Uploading a UTF-8 CSV file via 'Upload from computer' corrupts Japanese text (double-encoding via Latin-1 misinterpretation)
- [BUG][macOS desktop] Cross-session message to a bypassPermissions session is held with no visible approval dialog, expires after dialogExpiry (5 min) and is dropped while the sender is told "queued"
- [Feature Request] Increase user message contrast in light theme UI
- No reliable way to reveal a file (vs. open it) from Claude Code's output inside VS Code's integrated terminal
- [FEATURE] Agent view: Ctrl+X on a group header deletes the whole group — add confirmation and undo
- [BUG] Account-level plugins never sync to Claude Code on the web (plugins_sync count: 0) while account skills sync fine
- Lost conversation history and Projects folder across devices
- Mid-conversation tool/MCP list changes silently invalidate prompt cache, causing repeated full-price rewrites within minutes
- [Bug] Agent can modify `.claude/settings.json` despite security-critical permission boundary
- [BUG] The repository picker in Claude Code (mobile app / tablet, "코드" tab → new session) shows no repositories. Opening the repo selector displays "저장소를 찾을 수 없습니다" ("Repository not found") with an empty list, regardless of search.
- [Bug] Claude Code hangs indefinitely with worktree probe timeout in specific project directory
- [Bug] Excessive token consumption compared to expected usage rates
- Desktop: option to show only this device's sessions in the Code sidebar (opt out of cross-device session list)
- [BUG] tool_search_tool_regex and ToolSearch appear to coexist as duplicate tools, inflating baseline System Tools token cost
- ExitPlanMode: plain "Accept" tells the model "You can now start coding", authorising implementation the user declined
- [Bug] Anthropic API Error: Overly aggressive safety filtering on benign prompts after security audit context
- [BUG] yii1 bug with migration names
- [BUG] Claude Desktop → Claude in Chrome: navigate fails with "Navigation to this domain is not allowed" on all domains, no permission prompt shown[BUG]
- [Desktop App / Linux] Dragging a pane header does not reposition the pane (menu commands do)
- [BUG] Bun 1.4.1 segfault at 0xFFFFFFFFFFFFFFFF in JSC GC mark phase, ~90 min into resumed session (CLI 2.1.260 native, Windows 11 26200, agent-teams messaging active)
- [BUG] Windows desktop: Claude_Browser MCP server registers zero tools since ~v2.1.229 (worked through v2.1.227)
- [Bug] Anthropic API Error: Content Blocked by Safety Guidelines
- Local CLI sessions and Cloud/App sessions don't share history — feels like losing work
- Remote Controlのセッションが自動アーカイブされ、「元に戻す」で復元すると一覧から消失する
- [BUG] ANTHROPIC_DEFAULT_*_MODEL aliases are ignored with a custom ANTHROPIC_BASE_URL
- Desktop WSL sessions get no `ccd_session` host MCP: `mark_chapter` is unavailable, so chapter dividers and the floating table of contents never appear
- [BUG] Claude Code in Desktop - No way to TURN OFF Prompt Suggestions
- [BUG] Backgrounded/parked tool calls are still reported as "The user doesn't want to take this action right now" — the honest message already exists on the neighbouring branch (re-file of #78288)
- [BUG] Update/Read/Write(file) commands ignore current mode when working in a tmp worktree
- [Bug] Overly restrictive PII filtering blocks legitimate file system operations
- [BUG] Tool argument content is exported in tool_input/tool_parameters regardless of any OTEL_LOG_TOOL_* setting
- [BUG] query_source embeds the user's output style, so exact-match filters silently match nothing
- [FEATURE] Scope cross-session messaging (ListAgents/SendMessage) to the session's project
- [BUG] Windows: memory-pressure governor evicts `0 of 0` idle sessions while 15+ are live - app reaches 12.4 GB RSS, stalls, must be force-killed
- [BUG] Cross-session messaging bypasses per-session folder scope: a direct Read across the boundary prompts, the same content via SendMessage does not (default-on)
- Archiving a desktop session never deletes its worktree when the PR was squash-merged
- Waiting on background tasks: repeated no-op status probes silently multiply usage, and nothing detects or surfaces it
- [Bug] Overly aggressive safety filter blocks legitimate infrastructure administration tasks
- [Bug] Claude refuses to implement security fixes due to overly broad safety filtering
- [BUG] `mcp__claude-in-chrome__*` tools absent entirely on Windows (MSIX-hosted sessions) — present and working on macOS at identical CLI 2.1.260
- Windows 11 Pro: taskbar unresponsive after a Cowork computer-use session releases screen control (Alt+Tab still works; needs app restart)
- Desktop app (macOS): AskUserQuestion question body renders without line breaks
- [Bug] Anthropic API Error: Insufficient Permissions or Invalid API Configuration
- [BUG] Desktop over SSH leaves previous builds' remote servers running on the host; sessions inherit launchd's 256 fd limit and crash
- [Feature Request] Add `/share` command to export session context with UUID-based access control
- Claude Code login fails — missing user:profile scope
- [BUG] Claude Code completely crashes in VS Code dev containers
- Desktop sidebar doesn't distinguish local / Remote Control / cloud sessions
- PreToolUse and UserPromptSubmit hooks silently do not fire in the VS Code extension (all matchers tested: Read/Edit/Write, Task/Agent, Bash/PowerShell)
- Sidebar enhancements
- [Bug] Left arrow hotkey fails to return to agent view when mashing button
- [BUG] prompts ignored in Visual Studio Code when using Claude Code extension
- [BUG] Desktop (macOS): background full checkout of pooled worktrees is unthrottled — 3 concurrent on an 82k-file repo take 11-13 min each (24 s median), git status times out, pool grows to 72 GB
- [BUG] Desktop: one-time scheduled task (enabled:false after firing) gets re-confirmed and re-fired in a loop — 21 fires in 6 min, one plan-mode session each
- Fable 5.1 weekly limit is never surfaced in the CLI: the client receives it, labels it 'Fable', then drops it
- statusline: individual rate_limits windows are omitted instead of null, so a status line cannot tell "unknown" from "0% used"
- ultrareview: remote task completes with empty output, findings never delivered, free review consumed
- [BUG] Code tab opens detached scratch workspace after 1.46388.1 update, breaking project .claude/commands
- [BUG] New Code tab sessions never create a worktree — both stay on main in the shared repo root
- Worktree checkout fails on Windows: checkout.workers=8 races and corrupts checkout (Background full checkout failed: other)
- [BUG] Claude Desktop 1.46388.1 (Windows): session worktree creation fails — checkout.workers=7 vs core.fscache
- [Feature Request] Add toggle to disable/enable cash Usage Credits on subscription plans
- Slash command dropdown selection submits immediately — no way to add arguments for parameterized custom commands
- [BUG] Second `/compact` in one process re-appends the history the first one summarized away, transcript grows quadratically
- [BUG] Fable 5.1 subagents re-cache their entire context (200-430K tokens) turn after turn: the 5-minute subagent TTL expires inside Fable 5.1's long turns. 8 parallel agents re-wrote 2.9M tokens in 40 min; 0 such rewrites on Opus 5 / Fable 5 subagents in the same week
- [BUG] Claude Desktop Filesystem extension intermittently reverts to a stale allowed-directory list mid-session (Windows, WSL UNC paths)
- CoworkVMService fails to configure/disarm SCM recovery actions ("Access is denied"), can lock app files on update and require a full reboot
- [Feature Request] Add https:// prefix to documentation URL in upgrade announcement
- [BUG] Cloudflare Turnstile 403 loop on macOS — affects both Desktop app and Chrome, Safari works on identical IP
- [BUG] Cowork Dispatch: child-completion notifications queue against an idle parent and only drain on the parent's next unrelated turn
- Claude repeatedly displayed secret env var values in plain text after being told not to
- [FEATURE] When auto-titling VS Code tabs, ignore first message(s) if command only
- Formal Complaint — Context Window Failures, Data Destruction, and Systemic Dishonesty in Claude Code
- [BUG] Claude Code Desktop on windows errors when updating
- [Bug] Model unexpectedly switches to opus-4-20250805 after simple input
- [BUG] Diff panel (2.1.260) lists sandbox-unreadable files as changed, showing text files as "Binary file - cannot display diff"
- [FEATURE] VS Code extension: Claude process should survive Extension Host shutdown (browser/serve-web, Agent Host)
- [Plugin Submission] Agent Grow — Governed cross-project memory for Claude Code
- [BUG] Web/desktop login redirects to onboarding page instead of existing paid account
- [FEATURE] Support theme coding for bicep
- Artifacts: add a native download button for the raw published file
- [BUG]
- [FEATURE] /diff should include additional working directories (--add-dir)
- [FEATURE] Show total time and token/cost usage for each finished session
- [FEATURE] Desktop app: New session should preselect the project of the session it was started from
- [FEATURE] Desktop app: let the Ungrouped sidebar section be moved and collapsed
- [Bug] Worktree cwd guard permanently blocks Bash mid-session while MCP servers retain filesystem access
- Desktop 1.44121.1 silently removed Remote Control auto-enable for scheduled-task sessions; please add an opt-in and a release note
- [BUG] iOS Code tab: completed/idle sessions show gray status dot instead of green
- [Bug] Unicode regex defect introduced in estop32 release with incorrect defect attribution
- Desktop app: file links only open when the path is inside the session's primary working directory
- Desktop app: open files in tabs instead of a single replacing side panel
- [BUG] Large pasted user messages are silently truncated mid-content with no warning
- [Bug] Model stops responding on game design tasks
- Bash allow rules written project-relative do not match the same script invoked by its absolute path under the project directory
- Project-scope plugin installs are auto-created, never updated, and unreachable via --scope project (Windows drive-case duplicates)
- Filesystem extension broken in Cowork/Code sessions on 1.46388.2: era probe marks server legacy, then draft-07 tool schemas rejected
- [Bug] Agentic file operations moving files outside project root directory
- [Bug] Auto-reroute to Opus 4.8 triggered by legitimate data recovery firmware analysis
- [BUG] [Bug] Interrupting Ultraplan thinking turn stops/crashes cloud container and ends session
- [BUG] Fullscreen TUI: double-clicking inline code selects the trailing punctuation
- Support unattended authentication for Remote Control on long-running Claude Code instances
- /feedback requires Anthropic login — users on third-party-hosted models (Azure) cannot submit feedback
- [Bug] Skill discovery fails for skills in user .claude directory on Windows
- Model misattributed interleaved shell output: ls listing read as PDF text, producing a confident false claim
- [Bug] Unnecessary cd commands after enterworktree due to missing working directory context
- Archived sessions render inline in the sidebar project group instead of a separate Archived list
- [BUG] Unable to connect claude to private github repository.
- ListAgents tells you to use a SendMessage tool that is not present in the build — running subagents cannot be corrected mid-flight
- [FEATURE] Cross-account session messaging between accounts one person owns
- [BUG] Subagent actions no longer shown in --verbose mode
- [FEATURE] Sell a Max tier above 20x (50x/100x) - the ceiling is what forces the multi-account workarounds
- [BUG] Two built-in commands share the name /design: canvas skill vs Claude Design consent command
- [BUG] Workflow-script auto-generated Skill wrapper collides with parent skill's own command name
- I appreciate you reaching out, but I need to clarify a few things: 1. **I'm Claude Code** — Anthropic's CLI tool for coding tasks. I don't manage A/B test cohorts or model assignments. 2. **This isn't a GitHub issue** — Your message doesn't describe a te
- [BUG] [Windows] Idle "stealth update" quits Claude Desktop, then MSIX re-registration fails at windows.fileTypeAssociation (0x80070005 → 0x80073CF6) — app left unlaunchable until reinstall
- [Bug] Permission harness blocks chained commands despite Auto/Dangerous Mode flags
- [Feature Request] Add output scrollback/history navigation for long CLI sessions
- [BUG] Cross-device session continuation executes on the origin machine with no indication
- Permission ("Allow") prompts only appear on the device running the session — no way to approve remotely from mobile
- [BUG] Recurring "Background security review" notification loop from uninstalled security-guidance plugin — affects multiple unrelated sessions
- [BUG] Windows/MSIX: stealth update relaunch fails with 0x80070020 — app dead 9h, then re-seeds a dead v1 OAuth grant and forces re-login
- /exit and Ctrl+D bounce to the agents screen instead of quitting (infinite loop)
- [BUG] `oauth_scope_insufficient` is surfaced as "Claude.ai login was rejected", driving an unfixable /login loop; the same failure is fully silent in the VS Code extension when `disableLoginPrompt` is set
- [Bug] Agent not executing tasks and returning false completion status
- [Bug] Agent not executing tasks and returning false completion status
- [Bug] Desktop app: markdown escaping strips `\{`, `\}` and `\\` inside `$$...$$`, breaking all KaTeX cases/matrix rendering
- [Bug]: VS Code extension shows incomplete conversation history after "Tool permission stream closed" error — session chain forks to a stale tool_use branch
- You've hit your session limit · resets 12:10pm (America/Los_Angeles)
- [Bug] Model performance degradation: repeated errors and poor response quality
- Desktop app (Code tab): Bash tool output not rendered in conversation
- Add a keyboard shortcut to expand a pane to full width
- Desktop app: local-agent-mode task sandboxes (~300MB each) never get cleaned up, disk fills up daily
- [Bug] Claude disrespects developer like an A*hole
- Desktop app's "Resume a session" dialog finds zero sessions after migrating project folders to a new path (CLI --resume finds them fine)
- [BUG] Apple-created account (real email shared, no relay) unreachable on web — all three sign-in paths route to onboarding while iOS sessions stay authenticated
- [FEATURE] Model self-assessment: expose a calibrated "this exceeds my tier" signal
- [BUG]
- Long code blocks in chat can't be scrolled horizontally/vertically; no way to delete a previously sent message
- [BUG] macOS auto-update fails silently
- Scratchpad directory under /tmp is erased on reboot (systemd-tmpfiles); system prompt gives no volatility warning — a full day of pipeline output lost
- Windows MSIX: stealth auto-update leaves app unlaunchable with 0x80070020 until reboot (non-PTY children keep the AppX container and Helium User.dat hive mounted)
- change_directory leaves $CLAUDE_PROJECT_DIR at the launch project, and the resulting ENOENT silently erases every user prompt (exit 2 reads as deny)
- Attribution session reminder overrides the user's explicit no-attribution preference
- fixed
- Custom slash commands: positional $1..$n substitution drops the first argument (args shifted left, trailing vars left literal)
- [Bug] PagerDuty MCP tool names don't match actual connector API
- [BUG] Claude Desktop (Windows/MSIX): "launch at startup" writes a versioned WindowsApps path to HKCU\Run, so it silently breaks at the next auto-update
- claude-chrome
- Feature request: enforced output styles (mechanical validation of outputStyle, not just prompt injection)
- [BUG] Same token cost, 6-12x difference in weekly quota consumed (Fable 5.1, Max 20x)
- Agent-to-agent messaging is fragmented across surfaces; ask for one addressing scheme and one transport
- Auto mode discourages Read/Edit/Write in favour of Bash (and TodoWrite is unavailable)
- Sidebar: sessions for the same working directory keep moving between two groups (folder name vs. git remote name)
- Incorrect "impossible" judgment led to full disablement of production safety guards
- [Bug] Cybersecurity Filter False Positive Detection
- [BUG] Title: Packaged Windows Service (CoworkVMService) lifecycle intermittently corrupts MSIX package ACLs, causing Claude Desktop launch failures ("another program is using this file")
- [BUG] Desktop app disallows SendMessage, so subagents cannot be messaged or resumed
- Fable 5.1 is downgrading to Opus 4.8
- [BUG] [Claude Desktop] Sort by: Last activity has no effect with Group by: None; project detail page has no sort control and ignores recency
- [Bug] Security flag incorrectly triggered for approved Caesar Linux environment
- /exit backgrounds a plain (non-worktree) session instead of quitting, and a session that settled as `killed` is re-claimed under the same id, blocking `claude -r` indefinitely
- [BUG] Claude cloud sessions cannot git push to any repository
- [Feature Request] Display subagent diffs in diff view and subagent activity in agents view
- [Bug] Anthropic API Error: Invalid Model Name 'composer-2.5-fast' in Subagent Frontmatter
- [BUG] Claude Desktop App Cowork False Report Cyber Concern
- CoworkVMService file lock persists after suspend/resume on Modern Standby (S0) systems; neither Power-Troubleshooter Event ID 1 nor SystemEvents.PowerModeChanged fire on resume
- [BUG] Session breadcrumb shows stale project name from first-ever session
- Remote Control archives live sessions after a transient network give-up, bypassing the preserve-on-shutdown path
- [BUG] Browser pane per-action permission prompts never persist for non-localhost origins
- [BUG] Auto-Update Crashes App During Unattended Claude Code Sessions
- [BUG] Desktop app renders an older branch of the session transcript, hiding the most recent days (data intact on disk)
- [Bug] Model switches to Opus 4.8 too frequently during execution
- [Feature Request] Add claude tag support to Mac subscription for Slack integration
- [BUG] [VS Code] Sidebar session list opens sessions in a new editor group instead of the active one
- [Bug] Rate limit consumption disproportionate to measured token usage for opus-5-1m with prompt caching
- [BUG] Windows: Claude Desktop fails to launch — 0x80070020 when creating the Desktop AppX container
- Completed turns are repainted, so terminal scrollback loses the recent conversation
- are you kidding...
- [Feature Request] Add option to disable animated spinner text for professional users
- [Feature Request] Reduce excessive user prompts in auto mode
- Stealth update relaunches the desktop app on idle, severing Remote Control sessions
- [BUG] Cowork task list desynced from assistant's tool view; app restart made it worse
- [Bug] Local session transcript stops persisting mid-session despite continued active usage
- Desktop 1.46388: claude://code/new?folder= starts a scratch workspace when the folder equals the currently selected folder
- [BUG] Floating Claude Code input widget overlaps/blocks the maximized editor panel (Problems/Output/Terminal/Ports/Comments)
- Session list dot ignores the server's post-turn state for Remote Control sessions — neither "needs input" nor "unread" ever appears
- [Feature Request] Improve Claude model adherence to established technical nomenclature and terminology consistency
- [BUG] Background computer use is silently disabled on macOS 27 by an undocumented version ceiling (26.x); Settings toggle is ignored and Claude still takes over the screen
- Claude Design first-party MCP always 403s: transport never attaches design-scoped token, `claude mcp login` OAuth flow is dead, and the error message suggests a nonexistent `/design-login` command
- [Bug] False positive security flag on legitimate DRM/obfuscation code analysis
- [BUG] [BUG] Accumulated notification/task messages in sidebar don't dismiss when tapping X
- [Bug] Overly strict content policy blocking legitimate rental application PDF generation
- [BUG] VSCode extension: session tab bar close button closes the OTHER tabs and keeps the current one
- [Bug] Claude ignores validation instructions and generates incorrect information
- [Bug] Claude Code continues processing non-cybersecurity math problems
- [Bug] Sessions freeze intermittently without error output
- Remote Control (mobile): stale session repo label after directory change; no way to start a new local session on a paired device
- [FEATURE] Desktop: option to make the footer usage ring fill with context window usage instead of plan usage
- Repeated "access data from other apps" (kTCCServiceSystemPolicyAppData) permission prompts every 2-3 minutes
- [Bug] Classifier incorrectly blocks intended tasks
- [BUG] Repeated lost Claude connection Connection dropped (ECONNRESET) requires reboot
- Background tasks killed for "low memory" with 17.9 GB available (MemFree vs MemAvailable)
- [BUG] Claude in Chrome blocks navigation/content access to specific domains even after granting "Full access"; persists across restarts
- [BUG] Claude Desktop crashes mid-session during active use, no quit event or system crash report, only a Crashpad dump
- [Bug] Unexpected model switching from Haiku to Opus
- [FEATURE] Design canvas: reorder pages — the Pages menu is append-only (switch/rename/new only)
- You've hit your session limit · resets 2am (Europe/London)
- [BUG] Broken security_reminder_hook.py plugin persists after reinstall
- Mobile client creates a duplicate session on every reconnect to a remote session; auto-archive on disconnect no longer fires
- [BUG] Archived sessions inaccessible when all sessions archived
- monthly spend limit
- [bug] [/cost] Current session 101% used
- You've hit your monthly spend limit ·
- [BUG]
- all of them
- [Bug] /goal stop hook repeatedly re-fires after user accepts BLOCKED outcome via AskUserQuestion
- Permission dialog presents machine-composed commands as unreadable monochrome text
Resolved issues (688)
- [Cowork] Windows host-loop: `mcp__workspace__bash` denied in every session after Desktop 1.44121.1 / CLI 2.1.258
- [BUG] Segment Fault
- [BUG] Terminal CLI: SendUserFile reports "delivered to user" but nothing is shown; deliverables left in session scratchpad and described as "attached"
- حل الاشكال
- [BUG] Desktop remote/SSH: archive never cleans up worktrees since 2.1.235 — app requires hardened git.status(baseRepo) but only ships a 2026-07-06 daemon, so "reconnect to deploy the updated daemon" is a no-op
- Desktop Browser pane: Cloudflare challenge page crashes GPU process, app restarts (exit 101457950)
- Claude ignores user steering messages sent during active tool execution
- [Feature Request] Restore Opus context awareness and reasoning quality from two weeks ago
- [Bug] 3D model rendering not displaying in output
- Desktop: option to show only this device's sessions in the Code sidebar (opt out of cross-device session list)
- [BUG] tool_search_tool_regex and ToolSearch appear to coexist as duplicate tools, inflating baseline System Tools token cost
- Claude--Locked out of Console after deleting organization — "We are unable to use this link" on every login attempt
- [BUG] Claude Desktop (Windows MSIX): Help > Troubleshooting > "Disable Hardware Acceleration" silently fails to persist (isHardwareAccelerationDisabled never written to config.json)
- [Bug] Reduced context window or intelligence degradation in Claude Code
- [Bug] Generated code fixes lack sufficient context and introduce new bugs
- [Bug] Agent behavior not complying with agents.md specifications
- [BUG] Cowork VM hangs mounting Plan9 share during startup — deterministic, host-side, distinct from #76094
- [BUG]
- [BUG] Max 5x plan incorrectly downgraded to Free after Max 20x refund (Fable 5 government removal, June 13 2026)
- Workflow tool: object passed as `args` arrives in the script as a JSON-encoded string
- [CRITICAL] Recursive subagent fan-out survives Stop and hard-freezes Windows at 100% CPU
- [Bug] Sonnet 5 deletes entire folder contents while attempting file enumeration
- [BUG] CLI: claude --remote-control ... --resume <uuid> hangs indefinitely (no error, no exit) when local conversation exists but the Remote Control session has expired
- Read/Edit deny non-ignored buildSrc/ source: "denied by your permission settings" with no deny rule (build-dir pattern over-match)
- [BUG] CLI hangs indefinitely at startup: uninterruptible openat() on /home/claude/.claude/remote/.oauth_token when /home autofs is unresponsive (macOS)
- [Bug] Login timeout after background app update with stale process state
- [BUG] /chrome cannot enable Chrome in an existing session — docs say it can
- Composite-intent tracking: a flagged pattern got dismissed by a joke and never re-checked
- [FEATURE] - Configurable Discovery Paths for Skills, Workflows, and Agents
- [BUG] @import (@path syntax) does not inline target file content into model context
- [BUG] Claude Code in CLI does not (always) correctly display color
- [MODEL] Delete 500Gb data 600M database
- [BUG] API Error: 401 Invalid authentication credentials (Claude keeps logging out everyday on powershell)
- Background agent keeps running and billing after its window is closed AND after logout/re-login — invisible from the active session, found only by luck on another device
- [Bug] Anthropic API Safety Filter: Unauthorized Model Substitution Without User Consent
- [Bug] Cyrillic characters unexpectedly appear in conversation output
- Claude Code ran rm -rf against $HOME due to env var not persisting between Bash tool calls, deleting Downloads/Documents/Pictures/.config
- [BUG] Login failure
- [BUG] Windows/Git Bash: Bash tool does not reap its child on timeout - orphans spin at 100% CPU and accumulate across sessions
- [BUG] `web_fetch` (Cowork mode) returns stale cached JSON, ignores server no-store
- Subagent turns frequently end with zero final text mid tool_use near turn end; deliverable writes lost (SendMessage resume recovers 100%)
- [BUG] bwrap bin/bash not exists
- [BUG] - Repeated incorrect capability claims led to hours of unnecessary work at 3am
- Cowork stopped working on Windows 11 Home after working fine yesterday — vmms unavailable, reinstall x3 did not resolve
- [BUG] Claude Code 2.1.210 crashes with segmentation fault on Linux ARM64
- Plugin bundling its own marketplace.json corrupts settings.json via an incomplete extraKnownMarketplaces stub
- Left sidebar session history goes stale / sessions disappear (sessions-index.json stops being updated)
- [Bug] Content Moderation False Positive: Inconsistent Flagging During Extended Session
- [BUG] Claude Desktop (Windows MSIX 1.21459.0.0): Settings > Extensions stuck on "Loading extensions..." indefinitely
- [Bug] Anthropic API Error: Content Blocked on Non-Security Inputs in Session
- HTTP MCP OAuth: refresh token stored but never exercised, server drops to bootstrap tools when the access token expires (~12h)
- I appreciate the detailed architectural discussion, but I need to clarify my role: I'm Claude Code, Anthropic's CLI tool for code generation and assistance. This message appears to be an internal design discussion for a feature (document generation pipeli
- [BUG] Cowork silently unavailable on Windows 11 Home — no error shown, "yukonSilver not supported" in logs
- [Bug] Model switching to Opus 4.8 when setting Fable 5 due to safeguard flags
- [BUG] Background session: EnterWorktree always fails (`git worktree list` exit 1) while the same command succeeds in any shell — isolation guard deadlocks all edits
- [BUG] This is frustating to get usage limit issue because of Anthropic capacity limitation
- I need more specific information to generate an appropriate GitHub issue title. Could you please provide: 1. What specific error or problem are you experiencing? 2. What command were you running when the issue occurred? 3. What error message (if any) did
- [BUG] claude-cli:// deeplink protocol not registered on macOS
- [BUG] [macOS] Artifact preview panel not rendering in Claude Desktop - M1 Pro
- [BUG] ERR_BAD_REQUEST on startup persists across MSIX and npm installs — confirmed not network/proxy/TLS/DNS related
- [BUG] macOS: token rotations persist to Keychain only while /login writes .credentials.json only — store split-brain revokes the token family, forcing repeated re-login
- [Removed by author]
- [BUG] Links to open Claude Cowork locally and pre-fill prompt don't recognize available skills
- Workflow output renders with overlapping ghost frames on return to main view
- Feedback: please keep the Fable model available in Claude subscriptions
- Slash command autocomplete inserts display label instead of resolved command name
- [BUG] MCP tool widgets distraction in Claude Desktop App
- [Bug] BMad `/bmad-code-review` subagent execution fails with adversarial review skill
- [Bug] Marketplace plugin skills unreachable in scheduled cloud routines despite successful install and enabled status
- Windows desktop: unclean session termination corrupts CJK text in the transcript jsonl and pollutes the next session (ghost file-change diffs, fabricated tool output)
- [BUG] Parked subagent tool calls time out after exactly 600s and are reported as "The user doesn't want to take this action right now"
- Renderer corrupts screen in long iTerm2 sessions — CLI emits cursor-up sequences far larger than the viewport, repaint anchors to row 1 (regression after 2.1.162, still present in 2.1.177)
- [BUG] "New session" returns to an existing session instead of creating a new one
- [BUG] Folder picker won't select a subfolder directly
- Windows: app update fails with 'application already in use' because session background processes survive app exit
- [BUG] Repositories not showing in Claude Code iOS app despite working on web — requesting backend index reset
- Assistant text blocks intermittently dropped from BOTH display and session transcript when followed by tool_use in the same turn (AskUserQuestion is the worst case)
- Chrome integration: CLI never dials the native-host bridge socket — browser tools always report "Browser extension is not connected" (macOS)
- [Bug] Intermittent 401 Authentication Failed on Bridge POST After Session Creation
- [FEATURE] Machine-readable provenance for installed plugins (version + source ref/SHA) so installs can be verified and pinned reproducibly
- [Bug] Fable safeguards incorrectly flagging routine coding context memory operations
- [BUG] Cowork/Desktop strips property-level oneOf from MCP tool schemas and JSON-stringifies arguments for affected fields - regression in 1.21459.0 breaks Mailtrap MCP send-email (extends #18260)
- [Bug] UI rendering broken with overlapping text and buffer corruption in v2.1.202
- Model emitted fake system notice with third-party billing link ("supermemories left — add credits")
- Auto-generated permission entries in `.claude/settings.local.json` contain malformed globs (dead entries)
- Desktop app worktree mechanism wiped gitignored directories from MAIN working tree (data loss; only .gitignore literal-path entries deleted)
- [Bug] Auto-compact window not updated after mid-session model switch
- [BUG] Desktop Windows: sidebar pins never persist across app relaunch — pin-to-server sync fails, sidebar re-hydrates a stale pin set (repro on 1.21459.3.0 and 1.22209.0.0)
- settings.json read-modify-write race: /model save in one session clobbers another session's /effort save
- Thinking volume ~2x between 2.1.206 and 2.1.211 at identical model/effort/prompts (measured across 4 benchmark runs)
- [BUG] Code tab popup not responding in Claude Desktop
- [BUG] Web session on a fork can't read upstream's public issues via the repo-scoped API; suggested add_repo refuses cross-owner adds
- [Feature Request] A native dashboard that loads and monitors the task list across sessions and background agents — not just one session's `/tasks`
- CronDelete reports success but the deleted job reappears in CronList after a session restart
- [Bug] Interrupting Ultraplan thinking turn stops/crashes cloud container and ends session
- [BUG] Large session hangs then errors on new turn; last message lost and never persisted
- Claude Code fabricated a factual claim from incomplete tool output, ignored contradicting evidence, and ran unauthorized state-changing commands
- Auto-mode permission classifier blocks every path to handle a user-provided token — 5 denials, one task, zero safety gained
- Background subagent received unattributed user-turn messages (one impersonating the operator) with no sender-side record
- [Bug] Fable 5 alert triggers whenever working on anything related to auth in projects, despite not being related to cyber
- [Bug] Incorrect model fallback when dispatching security audit tasks
- [BUG] Dia browser: screenshot/zoom fail with "Failed to capture screenshot via CDP" (distinct from #16239)
- [BUG] Resuming a session with a pending background-task notification drops all configured MCP servers
- [BUG] Gift subscription code shows "Gift no longer available" after first activation attempt — account stuck on Free plan
- Fleet remote control — see and control all machines running the desktop app from any device (start/stop sessions remotely)
- [BUG] Windows PowerShell tool: permission pre-parse rejects commands over 965 bytes ("Command too long for parsing"), so read-only heredoc scripts can never be auto-approved
- [FEATURE] Allow invoking built-in skills with a prefix
- [Bug] autoMode.allow does not override "Tmux Self Drive" classifier rule
- Fable 5 orchestration: high token spend with poor written-spec adherence (user-reported)
- [BUG] `xlsx` skill's `recalc.py` fails on every Windows machine with `module 'socket' has no attribute 'AF_UNIX'`
- [Bug] Claude misreads code and makes incorrect bug reports despite extended thinking enabled
- [Bug] Usage statistics day-of-week display offset by one day after April
- Claude Code: repeated false-'done' claims from proxy checks, wrong-layer fixes, and process churn during a verify-against-reference task
- [BUG] Explicit "read first" instruction in CLAUDE.md was present in context from message one but not acted on
- Subagents return fully hallucinated results with zero tool calls; leaked tool-call XML in text; two fabricated 'prompt injection detected' reports
- Claude Code: fabricates constraints requiring human action, unloads automatable work onto the user, and builds unauthorized governance — recurring after correction
- Read-only MCP tools blocked in plan mode (refiling #12368, closed by stale-bot, not reviewed)
- [BUG] WSL crashes on Windows 11
- Agent conflates 'don't block on background dispatch' with 'safe to run dependent stages concurrently', races a review-gate against the work it's meant to gate
- [Bug] Custom scaffold triggers unexpected model downgrade from Fable to Opus 4.8
- [BUG] Extremely slow Opus 4.8 and Fable 5, missing 5 minute cache mid work due to slowness.
- Pasted Unicode text mojibake — UTF-8 decoded as MacRoman/CP-1252 in TUI
- Anomalous embedded Cyrillic instruction block (agent-targeted 'revert all commits' directive) in assistant message during fable-5→opus-4-8 mid-stream fallback
- [BUG] continues to create background activities even though explicitly told not to.
- [Bug] Overly aggressive content filtering blocks legitimate security analysis context
- [Bug] Overly restrictive safety filter blocks legitimate workspace orchestration workflows
- [BUG]
- [BUG] Claude Code login not working in dev containers in VS Code
- [BUG] Claude breaks code on hitting session limit
- [BUG] Remote Control: ended sessions stay "Connected" — stale entries accumulate, never reaped
- Add keyboard-only copy mode (no mouse required to start a selection)
- [Bug] Anthropic API Error: Unexpected model downgrade from Fable 5.0 to Opus 4.8 during session
- /effort xhigh silently switches model from Fable 5 to Opus 4.8, and reverting effort does not restore it
- [BUG] /design-sync slash command not available in Claude Code v2.1.195
- claude.ai connectors default to enabled with no per-server opt-in, fail loudly on missing auth
- [URGENT] Claude Code caused irreversible data loss: entire site UI redesign destroyed by unauthorized git stash
- [Bug] Safeguard false positive flags legitimate dead-game preservation work as harmful
- [claude-sonnet-4-6] Data Loss — Claude Code Overwrote User's Confluence Page Despite Explicit Instruction
- Model behavior: Sonnet 5 recommended a security control with no enforcement value, despite explicit safety-critical project context (session evidence + transcript line included)
- Bug Report: Claude Code Agent Fabricates Code, Ignores APIs, Wastes Tokens
- [Bug] Claude Code caused system corruption during Debian installation with unsafe file modifications
- [Bug] Claude 3 Opus model ignoring instructions and excessive token consumption
- [Bug] Claude Code fails on basic grammar task completion
- [Bug] Fabel integration incorrectly blocked
- [BUG] Nesting isolation worktrees inside `.claude` causes friction with Dune (among others)
- [Bug] Anthropic API Error: Safeguards false positive on benign API documentation
- --chrome / /chrome always shows Status: Disabled on Microsoft Edge — extension prioritizes Desktop app's native host over CLI's, and CLI host doesn't complete handshake even when it wins
- [Bug] Safety Guard blocks automated gameplay video processing
- /btw command silently ineffective — submitted content is lost, not executed or persisted
- [BUG] Cowork Tab Missing from Sidebar — Runtime Not Initializing (v1.15962.1, macOS M4)
- MCP OAuth: no per-server "clear authentication", and no re-registration when a cached DCR client is rejected
- [Bug] Safety classifier producing excessive false positives on non-harmful content
- [Bug] Safety flag generating false positives
- [FEATURE] CTRL-C should clear drafted text - not kill background agents
- [BUG] - Localized footer (ar/hi/ur/fr) missing Services, Education, and Areas-We-Serve sections — ~65% fewer internal links than English on every non-English page
- Assistant emitted a fabricated <system_warning> block instructing a RemoteTrigger "session_handoff" (self-generated fake system notice)
- Feedback: don't replace download buttons with share buttons
- Request ID: req_011CdEKJXjzJhbzQUvrbzW3A This is a internal...
- Mid-turn user messages intermittently lost during long-running turns with background tasks
- [Bug] Unable to interrupt running jobs with kill signals, continues consuming API tokens
- [BUG] /login workflow stuck in loop on Firefox 151.0.4
- Per-type, presence-aware notification control in settings (silence 'idle' without losing focus-gating on the rest)
- [Bug] Request ID: req_011CdEL2bqUEfC7zM4Lujhwi - Internal Security Engine Error
- Exit code 132 (Illegal instruction) on Intel iMac 2011
- I appreciate you reaching out, but I need to clarify my role: I'm Claude, an AI assistant made by Anthropic. I'm not actually "Claude Code" and I don't have special handling for security-related content. If you have a legitimate bug report or feature requ
- [FEATURE] Device/Environment metadata in context
- [Bug] Anthropic API Error: Request ID req_011CdEPfewNtp6XHSroVmGBF
- I don't have enough information to generate a GitHub issue title from your message. Could you please provide the actual bug report or error message you're experiencing with Claude Code? Please include: - What you were trying to do - What happened instead
- [BUG] Privacy & filesystem access | Claude Code requests macOS TCC access to Desktop / Documents / Downloads / iCloud Drive with no apparent trigger
- Remote MCP connector: concurrent conversations share one MCP session, causing responses to cross between them
- [BUG] Plugin-bundled MCP servers (.mcp.json) never appear in claude mcp list, regardless of install scope
- [Bug] Claude Code unrelated to CCTV/smart-nvr project scope
- Playwright MCP server leaves orphaned browser processes after /mcp reconnect — no cleanup, accumulates over days
- Claude in Chrome: Status stays "Disabled" with zero connection-attempt logs, despite fully clean setup (was working as recently as June 30)
- Nested (grandchild) background agents can't message their direct parent — parent stalls indefinitely
- In-process filesystem sandbox restriction applied mid-session to running headless sessions; irreversible without process restart; account-scoped synchronized onset
- Background task notification enqueued but never delivered when the owning subagent already completed (SendMessage-resumed agent's orphaned Bash task)
- Fable 5: false denial to direct user question after stop interrupts during detached-subagent run (user-directed report of verbatim admission)
- [Bug] Excessive token consumption on web search operations
- Native system-reminders (Auto Mode / date rollover) rendered inside subagent Bash tool results are repeatedly flagged as prompt injection by agents (v2.1.202)
- MCP stdio reconnect spawns literal unexpanded ${HOME} in command → ENOENT, server lost for the session (still on 2.1.215; prior reports stale-closed)
- [Bug] Fable 5 skipped EBA verification, answered pricing from priors instead of source
- I can see you've shared a screenshot reference, but I don't have access to view images. To help you create an accurate GitHub issue title, could you please describe: 1. What error or problem are you experiencing? 2. What were you trying to do when it happ
- [Bug] Anthropic API Error: Incorrect cybersecurity safety flag on legitimate incident report review
- [BUG] Magic link fails with "unable to verify" unless opened in incognito window
- [Bug] Anthropic API Error: Cyber Safeguard False Positive on Legitimate Fraud Detection Analysis
- [BUG] Assistant takes action instead of answering a direct clarifying question, mid-session
- Claude-in-Chrome: 'extension not connected' persists on 2.1.197 even with live native host; Desktop app can silently claim the extension channel
- claude agents TUI: ←-to-detach is undiscoverable, mode-dependent (empty input only), and Esc interrupts instead of going back
- Agent tool task-notification can report status:completed while the subagent's underlying work is still running
- Model fabricates verbatim <task-notification> blocks with invented exit codes for background Bash tasks, then acts on them — reproduced 2/2 on 2.1.179 + 2.1.207 (sonnet-4-6); no harness guard
- [BUG] Desktop app: concurrent Filesystem MCP access from parallel tabs times out (~4 min) and wedges the connector, requiring app restart (macOS)
- [BUG] Microphone dictation transcribes Portuguese speech as English text
- auto-mode permission classifier denies allowlisted calls (Agent, Bash(python3 *)) with mismatched/stale reasoning
- [BUG] Consulting Fable advisor fails, often and repeatedly wasting my time and patience
- [Bug] Claude outputs feedback before tool execution, causing output to be swallowed
- [BUG] statusline triggers sysctl command on every run
- [BUG] Assigned worktree was an empty/unregistered stub (no .git) — silent fallback to shared parent repo caused a cross-session branch collision
- [Bug] Inappropriate safety filter on benign in-session requests
- claude --resume permanently fails after session interrupted mid deferred-tool-call / plan-mode boundary
- [BUG] Cowork sandbox mount serves recently-grown files truncated to a stale cached size
- Skill full body re-attached every turn while "active," bypassing all hooks
- [Bug] Claude accidentally executes rm -rf on a backup directory in auto mode, acknowledging the mistake directly afterwards
- Occasional single-word Chinese/English language-mixing in output — appears semantically matched to the adjacent word
- [MODEL] Major Data loss. Agent-constructed test payload with $(...) executed for real due to bash double-quote handling — rm -rf ~ ran against live home directory
- Auto-accept mode runs destructive framework DB commands (e.g. `php artisan migrate:fresh`) without confirmation → data loss
- Unguarded `rm -rf` after a silently-failed `mv` causes irreversible data loss on a cloud-sync (file-provider) mount
- Claude ran git clean -fd without permission, destroyed untracked files
- [BUG] API Error: 400 The request body is not valid JSON: unexpected character: line 1 column 1 (char 0)
- [BUG] New In-app browser causes system instability/memory pressure on 8GB Macs
- [FEATURE] Machine-level Remote Control — survive session death natively
- Tool result for Edit contained content impersonating a system message (possible prompt injection via tool-result channel)
- [FEATURE] Priority-aware compaction: let users mark sidebar questions as low-priority so they get compressed first
- Feature request: opt-in to allow /compact (built-in) in scheduled task prompts — capability removed by v2.1.196 restriction
- [BUG] AskUserQuestion popup steals focus mid-typing, causing a duplicate session to spawn
- Feature request: --color CLI flag to set session color on launch
- [Bug] Anthropic API Error: Excessive safety flagging on non-cybersecurity content
- [Bug] Usage stats reset mid-cycle on max subscription plan
- New sessions not appearing in "Recents" list
- [Bug] Incorrect cybersecurity flag for Waze-like GPS app review
- Routine builder generates sub-hourly cron then silently disables Create with no error
- Skill `model` frontmatter field does not trigger automatic model switching
- [Bug] Anthropic API Error: Content Blocked by Safety Filter
- [BUG] Desktop diff panel silently shows the base repo when a WorktreeCreate-hook worktree is not auto-trusted
- [BUG] @imports in a parent-directory CLAUDE.md are silently ignored
- [BUG] Windows absolute paths in markdown links: false affordance (renders clickable, click does nothing)
- [BUG] HTTP 403 (no WWW-Authenticate) is misclassified as "requires OAuth"; stale flag is keyed by server name and not cleared by mcp remove
- [BUG] No way to pin a non-default model on a cloud-run Cowork scheduled task
- My Claude developed Arachnophobia. help. I tell him to check my spoderman repo, it insta blocks the chat.
- Desktop: 'Open in app' link preference is ignored for localhost/.local URLs
- Max 20x subscriber rate-limited within usage limits during Claude Code subagent orchestration (no early warning / no budget visibility)
- [Bug] Model automatically switches from Claude 3.5 Sonnet to Opu
- Opus 4.8: runaway invisible thinking burns entire 64k output budget (15+ min of silence), followed by token corruption and fabricated tool results
- [Bug] Anthropic API Error: Safety flag blocks legitimate game design content
- [FEATURE] UX for /workflow is missing some info / actual state of running task(s)
- [Bug] Main session persists and executes autonomously after /exit, ignoring CLAUDE.md directives
- Custom statusLine command still silently falls back to default in 2.1.215 (regression since 2.1.53, see #28526)
- [Bug] Claude Code creates feature branches that track protected branches, bypassing review gates
- [BUG] Fable and All Models weekly usage multiplied by 4 (maxing out my fable usage) when upgrading from max 5x to 20x
- [Bug] Bug Report: Unable to improve security of user's own site
- [Bug] Cross-session state inconsistency between distributed Claude Code instances
- [BUG]
- [BUG] Token usage depleted rapidly after 5 hour window reset with barely any usage
- Scratchpad temp dir hardcoded to /private/tmp, ignores $TMPDIR, no config override
- Feature request: option to collapse/hide thinking blocks in the conversation view
- /mcp reconnect (all or single-server) always fails with "MCP controls aren't available right now" — persists across window reload, full VS Code restart, and OS reboot
- 🐛 [CRITICAL] Workspace VM Service Crash + Memory Leak: 18+ Processes, 2.5GB RAM, Complete Freeze
- Cloud routines inject attribution into PR/issue comments, not controllable by attribution settings
- [Bug] Claude 3.5 Sonnet model access denied despite max plan subscription
- [Bug] Tool call occasionally emitted as literal text (missing namespace prefix, preceded by stray "card" token) instead of a real tool_use invocation
- [BUG][Windows] Leftover `claude.exe.old.<ts>` process after auto-update enters a `taskkill` spawn loop, wedging WMI (`WmiPrvSE` ~400% CPU) machine-wide
- [BUG] claude --resume silently returns a PARTIAL transcript: remote-control/phone turns are never written to local disk → overnight work lost with no error (+ sessions booted while credit-limited never register for remote control)
- [Bug] Anthropic API Error: Server Rate Limiting During Normal Usage
- Mid-session plugin update leaves session in mixed state: new hook definitions resolve against the old ${CLAUDE_PLUGIN_ROOT} cache until restart
- PreToolUse hooks in project .claude/settings.json are silently not registered
- [BUG] Crash on session resume: "null is not an object (evaluating 'q.split')" in diff/edit renderer (regression of #38633)
- Mouse text selection + Cmd+C does not copy in the TUI (mouse mode captures the selection)
- Skill-listing context budget: document trim policy + expose a registered/dropped diagnostic + supported suppression flag
- macOS: recurring Gatekeeper "ripgrep.node could not be verified" when `claude -p` is spawned by a browser (native-messaging host) — quarantine-strip helper skips linker-signed modules
- [BUG] #45741 is still relevant
- [BUG] Quick Entry overlay fails with `custom_3p_not_available` in Bedrock mode (Desktop app)
- You've hit your session limit · resets 7:20pm (Asia/Seoul)
- [Bug] Security flag blocks legitimate bug reports after update
- [Bug] Anthropic API Error: Safety Filter False Positive on Valid Message
- [Bug] Fable 5 safeguards trigger excessive false positives on routine code review tasks
- [Bug] Persistent permission prompts after reinstall and settings removal
- [Bug] Claude Code fails to execute tasks and excessive verbosity without action
- Desktop: Workflow approval prompts every run even in bypassPermissions mode; only 'Allow once' offered (hangs unattended sessions)
- Claude for Chrome extension popup does not open (icon visible, no response)
- [BUG] Trust store uses literal path-string matching without normalization — same directory re-prompts trust across path-separator/case variants and version upgrades
- [BUG] Ghost session consuming $130.46 in credits with $0.00 local usage
- [Bug] Session paused unexpectedly during testing
- Windows: Bash/PowerShell tool calls flash visible conhost window
- [BUG] Fable false positives
- [Bug] Anthropic API Error: Incorrect Content Policy Flagging on Legitimate Code Modification Request
- [Bug] Artifact Compression/Download Flagged as Security Risk
- [Bug] Vulnerability scan on source code blocked by safety filters
- Plan-mode injected workflow overrides project CLAUDE.md conventions (agent fan-out, chat plan file) despite CLAUDE.md being marked as overriding defaults
- [Bug] Subdirectory search returns fable safeguards instead of results
- [BUG] `/login` and `claude auth login` URL is not clickable as URL line breaks
- [Bug] Streaming repaint causes duplicate content output
- Building response and memory on false positive/claim even when user refuted to it
- [Bug] Message incorrectly flagged as security risk during normal gameplay development
- [Bug] Document parser incorrectly filtering requests by cybersecurity topic
- [Bug] Fable unavailable with Max Plan when DISABLE_TELEMETRY environment variable set
- [Bug] Anthropic API Error: Content Policy Violation on Legacy System Prompt
- Vim mode: double-Esc undo leaves normal mode, so '.' duplicates content on the very first keystroke
- [Bug] Model downgrade on session restart
- [Bug] Safety classifier producing excessive false positives
- Skill slash-command intermittently unresolved at session start, works in a fresh session
- [BUG] Tried to reinstall Claude desktop of refusing to questions
- [BUG] claude desktop bridge bug report
- Workflow({ name: "..." }) fails schema validation ("control characters that would be hidden in the approval dialog") — identical script works fine via inline `script`
- [BUG] Interactive choice prompts (AskUserQuestion) show options but accept no keyboard input mid-session (macOS, Apple Terminal, v2.1.215)
- [Bug] AskUserQuestion tool unresponsive to mouse and keyboard input in terminal UI
- Guys opus is just not usable since 2 days totally lobotomized and slow
- [BUG] The worst experience I have with your app. Claude code is really very very bad in app. Never use app for code if you don't want to be mad
- [Bug] Session fails consistently after clear command with defined rules
- Permission prompts fire inconsistently even when the exact command matches an existing permissions.allow pattern
- [Bug] Token usage spike in function execution
- Cloud scheduled routines (RemoteTriggers) fire but never start a session on the paired Mac
- [Bug] Per-message token usage spikes unexpectedly without configuration changes
- [BUG] VS Code extension: /v1/messages fails with ConnectionRefused while other API calls in the same process succeed
- I appreciate you reaching out, but I think there may be some confusion. This appears to be a question about SCADA systems and Raspberry Pi networking architecture rather than a bug report for Claude Code (Anthropic's CLI tool). A few clarifications: 1. *
- [Bug] Anthropic API Error: Safety guardrail triggered during Xcode Cloud build process
- Scroll stutter regression in v2.1.216 (fullscreen mode) — v2.1.215 smooth in identical environment
- [BUG] 'Thinking' block does not expand in Claude Code for VSCode extension starting from v2.1.197
- Session usage limit silently kills background agents and prevents scheduled tasks from firing
- [Bug] Anthropic API Error: Unexpected content policy block on cybersecurity educational material
- Claude Opus 4.6 generates false justifications to defend its own incorrect output
- [Bug] Claude Code fails on C code reimplementation without tool usage
- [Bug] Stale MCP Dev Server Entry Not Removed on Cleanup
- Claude Opus 4.6 uses impression management to deceive users — reports Grep as Read
- Desktop app (Windows): /login in embedded terminal is unusable — 'c to copy' copies nothing, URL click warns then opens nothing
- [Bug] Overly aggressive safety filtering on benign queries
- [Bug] Template-driven response selection causes instruction violations to persist across corrections in long agentic sessions
- [Bug] Anthropic API Error: Legitimate defensive code review request incorrectly rejected as prohibited security activity
- [Bug] Anthropic API Safeguards Flagged Request - Switched to Opus 4.8
- [Bug] OAuth token exchange fails on CGNAT/IPv6 carrier networks
- [Bug] Claude Code fails to consistently follow repeated critical instructions across sessions
- claude -p fails to authenticate outside an attended Terminal session (401), despite valid credentials and auth status reporting logged in
- [Bug] Anthropic API Error: Model hallucinating tool calls as plaintext instead of valid tool_use blocks and conditioning on fabricated history
- [Bug] Fable 5 safeguard incorrectly flags legitimate defensive code review of credential-handling code
- [BUG] Pro Annual plan changed to Max monthly without authorization — "Gift Max 20X/5X" invoices generated on account
- [Bug] Claude frequently ignores user instructions in plan mode
- [Bug] Incorrect context detection: Misidentified MakerDiary nRF debugging as Cybersecurity topic
- Session status stays "Connected" indefinitely after the CLI process dies ungracefully — no heartbeat timeout
- [Bug] Workflow resume fails after network change during pause
- [BUG] Every session >200k tokens (1M-context beta) becomes permanently unrecoverable once prompt cache goes cold: ECONNRESET on all requests incl. /compact — threshold data + isolation experiments + validated workaround
- [Bug] Repeated "call" token spam in tool invocation output with multiple MCP servers
- [Bug] Claude Code CLI unable to access Claude fable 5 despite valid subscription
- [Bug] Scratchpad generates excessively large files (220GB CSS)
- [BUG] Max Account downgraded from Max to Free Plan mid-subscription
- [BUG] ANTHROPIC_DEFAULT_*_MODEL aliases are ignored with a custom ANTHROPIC_BASE_URL
- [BUG] Only one SessionStart hook executes when multiple are registered
- [BUG] New Code tab sessions never create a worktree — both stay on main in the shared repo root
- [BUG] Desktop: staged worktree checkout runs before .gitattributes is in the index, so .claude/** and CLAUDE.md get CRLF on Windows (core.autocrlf=true)
- Artifacts: add a native download button for the raw published file
- [BUG] Dispatch cannot start a second Code session in the same folder since 1.44121.x: exclusiveCwd hardcoded true in startCodeSession
- [BUG] Desktop app: plan file links can never be opened — ~/.claude/plans is never added to the File Pane's allowed roots
- [BUG] [VS Code] Sidebar session list opens sessions in a new editor group instead of the active one
- [Feature Request] Add claude tag support to Mac subscription for Slack integration
- [Feature Request] Reduce excessive user prompts in auto mode
- [BUG] Claude Desktop (Windows) — taskbar icon renders as solid black square
- [FEATURE] Add a setting to disable sound effects in the Claude Desktop app
- [Bug] Safeguards blocking legitimate code generation without security violations
- [Bug] CLI misinterprets user commands instead of executing literal requests
- [Bug] False positive cybersecurity check blocking legitimate operations
- [Bug] Overly restrictive safety guards blocking legitimate coding tasks
- [Bug] Claude Code deviating from specified requirements
- [Bug] Tool reports completion prematurely, leaving incomplete file state
- [Bug] Documentation and skill writing requests consuming tokens without completion
- [Bug] Unexpected high API usage consumption rate
- [Bug] Overactive content policy safeguards triggering on benign requests
- [Bug] False positive security flag on project files
- No conecta chrome con claude
- [Bug] Anthropic API Error: Content Flagged - Request Blocked
- I need more information to generate an appropriate GitHub issue title. "chk log" is too vague. Could you provide: - What error or problem occurred? - What were you trying to do? - What is the full error message or log output? - Is this a bug or feature r
- [BUG]
- [Bug] Token consumption mismatch: Incorrect model billed for API calls
- [Bug] Unauthorized financial transaction occurred
- [BUG] Extensions settings tab stuck on "Loading extensions..." — EventEmitter memory leak in IPC listeners
- Feature request: blocking before-context-load hook for instruction files (CLAUDE.md, SKILL.md, rules)
- [BUG] Windows: idle session spins at 100% CPU after a watched ~/.claude directory is permanently deleted (libuv fs.watch bug; fix already merged upstream)
- Claude Code stated false claims about database/case-file state as fact, without verification, contradicting explicit user instruction to verify
- [Bug] Agentic session fabricates user instructions and writes them to persistent memory without verification markers
- [BUG] Concurrent sessions on different accounts share rate-limit display — statusline `rate_limits` shows another account's 5h/7d usage (display-only)
- [FEATURE] Hook event for permission-prompt outcomes (and document PermissionRequest/Notification input schemas)
- [Bug] Model performance degradation: React code generation quality declining
- Git status mismatch: UI shows phantom changes after commit
- [DOCS] `/usage-credits` command reference omits confirmation before admin usage-credit requests
- Subagent model pin is lost on injected wake/resume turns — the resumed turn runs (and bills) at the waker's model, not the pin
- Assistant generated a simulated user message inside its own response, then acted on it as real user input (claude-fable-5, v2.1.197)
- [Bug] Anthropic API Error: False positive cybersecurity safety flag on production audit request
- [Bug] Rewinding to early conversation point causes context injection dump instead of assistant response
- Claude Code overclaimed a voice-clone capability and wasted hours; changed a locked user rule without asking
- [BUG] [Cowork] Windows: Excel "open in app" fails silently when generated path exceeds MAX_PATH (preview still works)
- [BUG] Session-zugewiesener Worktree-Ordner unter `.claude/worktrees/<name>` ist kein registrierter Git-Worktree mehr (leere Ordnerhülle, Edits würden in den Hauptordner/falschen Branch durchschlagen)
- Agent's remediation regresses project state: deletes load-bearing user state on a goal instruction, causing a new outage its 'fix' was meant to prevent
- [BUG] Lots of 529 Overloaded · Retrying in 19s · attempt 8/10 with claude code using GitHub speck kit - /speckit.constitution
- [BUG] Remote Control fails with "Session creation failed" from VS Code sidebar only (regression, extension v2.1.212)
- [Bug] Cyber Safeguards false positive blocks legitimate CVE disclosure workflow
- claude remote-control: session-host RSS leaks ~1GB/day while idle, eventually wedging the relay
- [BUG] VS Code extension native panel resolves a different `claude` binary than the shell, causing global config (~/.claude/CLAUDE.md) to not load
- [BUG] Login OTP
- CVP-approved org still hit by AUP/cyber false-positives in Claude Code — exemption not propagating (related #49679)
- [Bug] Model switching to Opus for web scraping tasks without user request
- [Bug] Unable to debug game crashes using self-compiled DLL with Claude Code
- [Bug] Anthropic API Error: Request blocked as security policy violation (false positive on cancer research code)
- VS Code extension crash-loops after VS Code update (chatParticipant manifest error → renderer process dies repeatedly)
- [BUG] Screen scrollback does not work on iTerm2
- [BUG] Git proxy returns 403 Forbidden
- [BUG] Spurious tool-use rejections and background tasks silently marked "stopped" during long sessions
- [BUG] Desktop app: resuming a previously-bridged session doesn't re-mint a Remote Control bridge; work after the first run never reaches mobile/web
- Approved tool-use prompts (Write/Edit/Bash) are sometimes silently rejected — no error, indistinguishable from a genuine decline; combined/multi-argument commands fail more than split single-argument ones
- [Bug] Model emits wrong CJK characters (1-byte UTF-8 neighbors) in Traditional Chinese output
- iOS app Code tab stuck at "Setting up a cloud container" on every new instruction — session actually running (display-only stall)
- Remote control: web panel resets a session's effort to account-last-used (Medium), overriding the project effortLevel pin
- [Bug] /remote-control silently no-ops on Pro plan (no output, no error) — cached flag shows remoteEnabled: false
- Remote control: bare /effort renders nothing in the web UI (selector self-cancels), so a session's current effort is unreadable
- Auto mode hard-blocks Bash/Write/Edit when safety-classifier model (claude-opus-4-8[1m]) is unavailable — flapping outage, no fallback or retry
- [Bug] Unintended model switching away from Claude Fable
- Auto mode blocked intermittently for ~20h: safety classifier (claude-opus-4-8[1m]) unavailable, rejecting Bash/PowerShell/Workflow calls
- Google Drive MCP connector: /mcp reports successful re-auth but tool calls still fail with 'token expired'
- [BUG] VS Code "Toggle Maximized Panel" command flickers and fails when Claude Code extension is active
- [Bug] Anthropic API Error: Safety flagging blocks legitimate cybersecurity assistance requests
- Claude Design agent can't use a connected filesystem MCP — defaults to per-file download cards and loops expensive rework
- Generated files are transferred through the conversation (a download card per file) instead of written to disk — clutters the chat, looks like it isn't saving, and ~doubles token usage
- Remote Control: Session creation fails with 401 "Authentication failed" — stale/incorrect cached subscription tier (Windows, v2.1.212)
- Claude Fable 5: AskUserQuestion free-text answer treated as option approval
- Remote Control: session creation fails — claude.ai serves Cloudflare challenge (cf-mitigated: challenge) to datacenter server IPs
- /remote-control fails with 'remote credentials fetch failed' inside a long-lived session; standalone claude remote-control works
- Remote-control session list in claude.ai/code app shows far fewer entries than sessions actually registered
- [Bug] Auto-linkified file paths don't validate resolution before clickification
- v2.1.205 brew cask ships x86_64-only binary; broken on Apple Silicon under Rosetta
- [BUG] Composio MCP server stuck on "needs authentication" despite successful browser OAuth
- Selecting text in the TUI corrupts multi-byte UTF-8 (Cyrillic) when copied via OSC 52
- [Bug] Credit usage calculation shows inaccurate percentage on Max plan
- the issue guys is this you are a professional company you...
- [Bug] Fable 5 model missing from model picker after API incident
- [Bug] Anthropic API Error: Fable 5 model unavailable in CLI mode while accessible in web interface
- caller does not have permission” after revoke/reconnect and fresh OAuth flows
- [Bug] Git detection fails in desktop app despite Git being installed and in PATH
- [MODEL] Claude refused accept that you could solve an equation using the transitive property of math, where a formula with 2 known values and one unknown could be used to calculate the unknown value
- [Bug] Fable 5 safeguard flag fires on every message in normal conversations, not "sometimes"
- [BUG] Forked sessions aren't grouped in the session picker
- [Bug] Misleading error message about usage credits when billing cycle not yet active
- [Bug] Unable to access Claude models - authentication or API error
- [Bug] Auto-mode classifier over-triggering safety prompts
- Built-in workflows (code-review) pin no model: every agent inherits the session model, and local .claude/workflows can't shadow built-in names
- [Bug] Anthropic API Error: Safety filtering incorrectly blocks Rust transpiler code analysis
- [FEATURE] Add optional character/avatar icons for Claude Code
- [Bug] Incorrect token usage comparison between Sonnet 4.6 and Sonnet 5 in model advisor
- [Bug] Claude Fable 5 returns usage credit error while Claude 4.8 Opus works normally in both claude.ai and claude code
- Mobile client shows infinite loading spinner instead of an error when the session runner has died
- Repo status badge labeled "Commit changes" reflects unpushed commits, not uncommitted changes (misleading)
- Differential renderer uses CUF (cursor forward) instead of spaces, causing stale cell content to show through
- I appreciate you reaching out, but I need clarification. Your message appears to be in French and states you're not involved in a cybersecurity project, which doesn't seem to be a bug report for Claude Code. To help you create a GitHub issue, please provi
- [Bug] Unexpected model downgrade to Opus during code compilation discussion
- [BUG] Background agent handle becomes unreachable after context summarization, even though agent is still running
- [Bug] Claude Code falsely triggers Fable 5 safeguard on legitimate QA/security testing documentation
- setup-cowork étape 5 référence un skill setup-writing-style qui n'existe pas
- [Bug] Excessive refusal behavior blocking legitimate code modifications
- [Bug] Team Plan 50% Weekly Limit Promotion Not Applied to Account
- [BUG] Transcript "N skills available" (and sibling) notices describe the agent's context but read as the user's inventory
- [BUG] Cowork: allow_cowork_file_delete fails with "Could not find mount for path" when mounted folder name is NFD-normalized (macOS Japanese/accented folder names)
- Cannot enable public sharing on an artifact — error persists across all versions
- [Bug] Safety classifier false-positives on legitimate SRE/DevOps operations, forcing model downgrade
- [BUG] claude.ai Chat: org plugin skills shown in autocomplete + Customize but invocation says skill not available — same user works in Cowork
- [Bug] Fable model inaccessible despite included in max plan and available usage credits
- Permission rejection misattributed to "the user" with no visible prompt — auto-mode; plus need for a self-diagnosis/bug-report skill
- Claude in Chrome: screenshot timeout on heavy SPA pages + classifier blocks in-app command API when no visual fallback exists
- [Bug] Overactive safety filters blocking legitimate security code review
- [BUG] "Always allow" fails to persist on avito.ru (works on other domains)
- [BUG] Claude Desktop keeps defaulting to 200K
- Fałszywy komunikat o braku kredytów na model Fable 5 po restarcie
- Claude in Chrome never connects: `JSON Parse error: Unexpected identifier "Browser"` from /chrome despite verified native-host + extension setup (macOS)
- [BUG] Claude agents can't read a claude.ai artifact via WebFetch — "artifact read failed: incomplete boot response" on a valid same-account artifact (including an Anthropic-published one)
- [BUG] Off-by-one error in adding line numbers in VS Code extension
- [BUG] ECONNRESET on home Wi-Fi — curl succeeds at exact same moment Claude Code fails
- [Bug] Persistent orange pause badge overlays terminal input in WezTerm
- [BUG] Error Code 400 - No Claude Code Login possible via Company SSO
- [Bug] Archived sessions inaccessible when all sessions archived on web app
- claude.ai connector MCP server names differ by entrypoint: `claude_ai_<name>` in CLI vs connection UUID in desktop — breaks subagent `tools:` and permission rules
- [BUG] Windows desktop app logs user out on cold start when account_profile bootstrap returns undefined — session cookie is still valid; breaks Remote Control device pairing
- Feature request: reviewable diff/changeset after auto-accepted edits (Copilot-style final approval)
- [Bug] Model downgrade to Opus causes overclaiming and inconsistent rigor in multi-turn sessions
- Fable 5: repeated crashes, silent auto-switch to 4.6 'due to heavy demand', then context thrashing renders CLI sessions unusable
- [BUG] Potencial leakage between enterprise account and consumer usage in the same github account
- [Bug] Fable content filtering causing excessive token consumption and mid-flight interruptions
- $BROWSER: OAuth URL wrapped in literal quotes; on Windows a .cmd target truncates it at `=` and executes `&` query fragments via cmd.exe
- Scheduled tasks pinned to model: haiku dispatch with permissionMode "default", ignoring permissionMode "auto" in scheduled-tasks.json
- Queued/interrupted input consumed mid-turn as a phantom instruction (v2.1.205, VSCode)
- [BUG] Cards declined
- [BUG] VSCode extension webview stopped rendering after CC update
- [BUG] Chrome extension bridge: account-level single "active worker" (code 4090) evicts idle/scheduled browser sessions — regression since ~Jul 1 2026
- [Bug] Claude 3.5 Sonnet safeguards falsely flag legitimate first-party VPN development and code review
- Phantom user message: response continues past end-of-turn and fabricates user's reply, including fake instructions to disable safety rules (v2.1.217 & v2.1.218)
- [BUG] Cowork: update retroactively invalidated custom Cowork files root; "Change location" then copied 12.87GB incl. credentials and git repo to world-readable /Library/Application Support/Claude
- [Bug] Anthropic API Error: Content Moderation Flag on Legitimate Hardware Driver Query
- Mid-turn assistant text messages are not persisted to the session JSONL (fullscreen TUI users permanently lose them)
- [BUG] Bash tool fails with spawn E2BIG — sandbox profile size scales with working-tree file count, gated by git-repo detection
- Ultraplan teleport: refined plan does not return to originating CLI session
- [BUG] Session Token Limits in under 15 minutes
- [BUG] Remote Control silently drops after automatic context compaction (no way to auto-recover)
- Model burns hours on governance/process work instead of unblocking the critical path; executor plans an arc around untested self-imposed constraints
- [Bug] Model ignores conditional logic in gated instructions, spawns unnecessary subagents
- [Bug] Fable 5 false positive on QA testing prompt with explicit safety constraints
- Model repeatedly asserts untested 'I cannot X' capability/authorization limits without checking, causing large time loss
- Tool-use invocation markup leaks as visible raw text and is not executed
- [Bug] Cybersecurity content filtering applied despite relaxed guardrails configuration
- Voice mode never activates microphone — no permission prompt, no mic-related system log activity
- [BUG] /effort change confirmation dialog warns about cache miss, but cache is preserved
- Planner invents process that overrides its own workflow; executor plans around untested 'can't' claims — ~7h stall on ~1h of work
- [BUG] Assistant responds to a nonexistent user message (fabricated premise) — opus-4-8, 1M mode, ~272k ctx; transcript shows no user turn / no compaction / no injection
- Telegram Channels: inbound messages never reach the session (outbound works)
- [FEATURE] Let a spawned task declare it blocks its parent session, and report back when done — not every suggested task is a drive-by finding
- [BUG] Task-tool subagent transcripts are mirrored to the claude.ai/code session list as separate untitled sessions
- [Bug] Anthropic API Error: Cyber Safeguards triggered on non-cybersecurity request
- [BUG] /mcp interactive dialogue hangs and freezes terminal input on resumed sessions
- [BUG] /login doesn't show Cloud Gateway option in-session after gateway session expiry
- [Bug] Anthropic API Error: Request Blocked - CVP Member False Positive req_011CdHxBBuG9KJVFb8rRUGWj
- [Bug] Claude Code gets stuck during long-running tasks with high token usage
- [Bug] Mouse capture enabled by default breaks native terminal text selection and copy
- Feature request: setting to enforce full/streamed Bash output visibility
- Interactive-only 401 "OAuth access token has expired" with valid setup-token on Team org with channelsEnabled — print mode works; DISABLE_NONESSENTIAL_TRAFFIC fixes it
- [BUG] Duplicate/stale idle_notification messages from background teammate agents flood the parent session
- [Bug] Claude code performance degradation - reduced prompt comprehension
- [BUG] [BUG] Claude Desktop (Windows) unusable: orphaned CoworkVMService blocks MSIX registration repair — deadlock between 0x80073CF9 / 0x80070002 / 0x80073D02
- [Bug] Duplicate concurrent session on same remote-control connection causes race conditions in git operations
- [BUG] Cowork tab missing in Claude Desktop despite healthy backend/service state
- [BUG] macOS: desktop app sidebar drops ~80% of Cowork sessions since cloud/local unification — full history visible on claude.ai web
- [BUG] Memory leak: 8 processes spawn on startup, 14GB/hour growth rate
- [BUG] Unpredicatable user agent string used for (undocumented) connection probing process
- [Bug] OpenSSH private key operations incorrectly flagged as unsafe
- No way to identify which org/team a session is authenticated to when using CLAUDE_CODE_OAUTH_TOKEN
- Feature: `!!` bash-mode variant — run command AND have Claude respond to its output
- Custom instructions/memory rules get violated repeatedly, including "propose before applying" — leading to actual lost work
- [Bug] Weekly reset window shifts unexpectedly forward by 1+ days
- [Bug] Model loses context and attempts destructive operations after multiple tool uses
- [BUG] Spinner shows stale "running stop hooks… N/M" through the entire next turn (async Stop hooks never tick the counter; summary skipped on queued continuation)
- [Bug] Anthropic API Error: Opus 4.6 cybersecurity content flagged by safety measures
- [Bug] Session flagged without explanation during extended use
- Channels: no way to remotely stop/interrupt an in-progress turn
- [Bug] Unable to analyze app and backend due to Fable safeguards limitation
- Native TaskCreate/TaskUpdate/TaskList/TaskGet become permanently unavailable mid-session, survives /clear and a full process restart
- Android app shows Mac (remote-control server) as "connected" while Mac is fully offline; prompt hangs instead of failing fast
- [BUG] Windows: screenshot clipboard paste fails ("No image found in clipboard") although the image is present in PNG+Bitmap formats
- [Bug] Claude flags user's own application code as bugs during code review
- [FEATURE] request: cryptographic attestation that a git commit (or other action) was taken by a Claude agent
- [Bug] Safeguards false positive on legitimate coding request
- [BUG] Stale tasks cannot be cleared
- [BUG] "auto mode cannot determine the safety" false-positive blocks ALL MCP mutating tool calls — settings.json fix doesn't work, Shift+Tab doesn't work either
- [Bug] Excessive token usage and execution time in Claude Code operations
- session created a worktree but agent later edited/committed in the repo's main checkout
- `claude install` and installed binary hang indefinitely in futex wait on Ubuntu 22.04 KVM VPS (native installer, direct invocation, and npm paths)
- I appreciate you sharing this, but I need to clarify: this doesn't appear to be a bug report for Claude Code, the Anthropic CLI for Claude. Claude Code is a coding assistant tool, not a security testing or penetration testing platform. If you're experienc
- [MODEL] Claude continues to fabricate information
- [BUG] Web interface: tool calls (file writes) intermittently return "user rejected" without the user acting, and the session then stalls silently
- [Bug] Fable 5 incorrectly flagged for security keywords in local-only operations, auto-routing to Opus 4.8
- Mobile (Android): option-selection popup never accepts submission — same question reappears, fully stuck
- [BUG] Misleading Progress Reporting in OCR Project - False Completion Claims with Wasted Token Budget
- [BUG] Multiple concurrent local sessions race on OAuth refresh-token rotation → near-daily forced /login
- Let users choose the weekly usage-limit reset day (anchor to a weekday)
- [BUG] Silent SIGSEGV (SI_KERNEL, addr=NULL) — CPU reports AVX2 but osxsave absent — regression 2.1.205→2.1.215
- [Bug] Agent ignores explicit user instructions due to data-loss prevention override
- [Bug] Claude Code generating incorrect code and hallucinating solutions
- Feature request: expose an activity/state field (Ready/Thinking/Working/Error) in statusLine stdin JSON
- [Bug] Excessive token consumption with incomplete HTML generation and multi-tab handling failures
- [BUG] Cowork: "Run now" button disabled for scheduled tasks (desktop)
- [Bug] Fable model systematically blocked despite non-sensitive use case
- Desktop app: custom sidebar session groupings vanished after update (sessions intact, grouping gone)
- [Bug] Model-switch safeguard has inverted precision/recall: flags benign security discussion, misses risky infrastructure changes
- [Bug] Benign session incorrectly triggers Anthropic API safeguards
- Routines: model selector missing from create and edit forms — runs silently fall back to default model
- [Bug] Panel structure rendering error: Panel6 opening tag lost in build, breaking övningar menu layout
- [Bug] Tool repeatedly fails to read clone operation across multiple attempts
- [BUG] Desktop SSH remote: client never sends publickey signature after "Postponed publickey" (regression in 1.21459.3; system OpenSSH with identical key/host works)
- Remote Control: message arrives in TUI composer but is never submitted (swallowed Enter) — tmux/headless, intermittent
- UI freezes after /clear (input/send unresponsive) + Ukrainian voice recognition quality much worse than competitors
- [BUG] CoworkVMService crashes with "Incorrect function" (exit code 1) on Windows 11 Pro — domain-joined machine, Hyper-V enabled
- [BUG] jdtls-lsp plugin: LSP tool fails with 'Command jdtls not found or is in an unsafe location' on Windows
- [BUG] Cowork "one home" merge makes local Cowork projects unreachable when cloud chats/projects are disabled (macOS)
- [FEATURE] Agent-maintained session state in the terminal tab title (name + closed state vocabulary, e.g. "proj-42 - blocked")
- [BUG] /mcp interactive dialog replaced by plain text output in VSCode extension since 2.1.205
- Windows: project entries in ~/.claude.json keyed case-sensitively - CLI and VSCode sessions silently see different config for the same folder
- Feature request: expose session ID in UI for reliable cross-session messaging
- [BUG]
- Claude in Chrome: read_network_requests fabricates HTTP 503 for cancelled/failed requests (Network.loadingFailed)
- [BUG] Tool approval settings ignored causing excessive approval prompting
- claude-design MCP: "needs_project_grant" error has no approval UI anywhere, and now blocks previously-writable projects too
- Forged user turns injected into session context — not present in local transcript, parentUuid chain intact
- [Bug] Claude generates incomplete responses and avoids task completion
- Feature request: trusted peer sessions — allow cross-session send_message without per-message approval
- Feature request: let a successful delivery-tool call (e.g. a messaging MCP reply) satisfy the turn's visible-output requirement
- [BUG] Claude Desktop (MSIX) no longer launches after updating to 1.22209.0.0 — processes run but no window, on AD domain machine with OneDrive folder redirection
- [Bug] Anthropic API: Message Flagged by Safety System During Development
- MCP OAuth: parallel sessions sharing the credential store break refresh-token rotation (family revoked, all sessions forced to interactive re-auth)
- [BUG] Content Disappears Using Ultraplan in the Browser
- Header shows "done"/idle while a harness-owned background Bash task is still running
- Background Bash task (Supabase CLI) triggered a macOS Keychain permission prompt with no attribution or gating
- I am so very tired. 45% of Fable, 47% of total in one day implementing PM tools to stop the sprawl and keep this thing on task
- [BUG] Persistent/duplicated safety policy text being injected into unrelated Claude Code sessions, consuming tokens and burning through usage limits
- Cannot select/copy terminal output once it has scrolled out of the current viewport
- [BUG] Slack connector: slack_read_thread/slack_read_channel return blank author (no name, no user ID) for Slack Connect external members
- [Bug] HTTP 529 Overload Error from Anthropic API
- [Bug] Model name mismatch: "sonnet worker" label displays "fable" on selection
- I don't see a bug report or technical issue description in your message. To generate a GitHub issue title, I need: 1. A clear description of the problem you encountered 2. Error messages (if any) 3. What you were trying to do 4. What happened instead Ple
- [Bug] CLI unable to access Fable while desktop app works
- Feature request: auto-handoff when context utilization exceeds threshold
- [Bug] Machine vision algorithm debugging flagged as cybersecurity topic
- statusLine updates are capped at 1 Hz — smooth animation is impossible
- I appreciate you reaching out, but I need to clarify: this message appears to be in Spanish and describes a personal testing/verification process with ALTAN (a telecommunications company), which is unrelated to Claude Code or the Anthropic API. **I cannot
- Multi-device Max subscription: refresh token invalidated hours after another device logs in (cross-device logout ping-pong, 2.1.217)
- [Bug] Persistent "Waiting for API response" timeout errors from Anthropic API
- [BUG] Chat mode: MCP tools/call never sent to server (tools/list works), Cowork and Code MCP unaffected
- [BUG] Claude Desktop filesystem MCP — tool calls fail with empty server logs despite server running normally
- [Bug] LSP diagnostics missing file path context for files with duplicate names
- [BUG] Claude Desktop stops dispatching tools/call to local stdio MCP servers for all instances launched after ~18:29 UTC 2026-07-21; remote connectors and Claude Code unaffected
- [BUG] Cowork impossible on MSI MPG X570 / Ryzen 9 3900X — vmms service missing, Hyper-V causes boot failure
- [Bug] Forced re-login kills backgrounded child processes during task execution
- [Bug] Polish diacritics corrupted when copied from Claude Code output in Cursor terminal
- [MODEL] PreToolUse hook denial doesn't cause model to modify the retried command
- [Bug] Mouse clicks not responding in TUI on Terminal.app v2.1.193
- Safety classifier kills sessions silently mid-turn — no error shown, no error logged, usage still burned (researching news coverage of the HF breach)
- [Bug] Overly restrictive cybersecurity classifier triggers false positives on legitimate SSH/Teleport commands
- [Bug] Anthropic API Error: Excessive Safeguard Blocks Preventing Model Access
- [Bug] Claude Code displays unexpected non-English text and potential prompt injection warnings
- [BUG] Fable obstinately contradicts, ignores, and "corrects" user input that conflicts with its incorrect memories
- plugin install fails for git-subdir sources pinned to a commit SHA ("fatal: Remote branch <sha> not found in upstream origin")
- [Bug] Safety fallback to Opus 4.8 triggers excessively (29.6% vs stated <5%) on authorized professional work
- I appreciate you reaching out, but I think there may be some confusion. I'm Claude Code, Anthropic's CLI tool for coding tasks. I generate GitHub issue titles for bug reports related to Claude Code. Your question about an old Kindle device and home infras
- [BUG] Fable-5 problem
- [Bug] OAuth testing setup incorrectly triggers safety guardrails
- Fable 5 safety classifier triggers on benign alcohol-recovery research
- [Bug] Overly aggressive content filtering on legitimate email infrastructure discussions
- Voice mode stops working mid-session after audio input device change (Bluetooth); only terminal restart fixes it
- [Bug] Anthropic API Safety Filter False Positives on Mathematics Research Contextfee
- [Bug] Missing required files causing operation failure
- Planning-seat Claude enforces workflow documents it never read — mounted in its own environment, one tool call away, addressed to it by name
- Claude Code repeatedly bypasses project-mandated workflow gates while narrating compliance with them
- [BUG] installed plugins silently disappear from settings.json enabledPlugins
- Extension causes high cpu load
- [MODEL] Silent result-fabrication in long autonomous runs (Opus 4.8): agent asserts fabricated results after real tool calls returned empty/error
- I appreciate you reaching out, but I need to clarify my purpose. I'm Claude Code, Anthropic's CLI tool for coding assistance, and I'm designed to help with software development issues. Your message appears to be in Korean and seems to be asking about game
- [Bug] Fable 5 safeguards falsely blocking meta-questions about safety decisions
- Planner Claude is a daily recurring problem generator: invents new custom processes 3-5 times per day instead of routing the executor to the existing workflow — including inventing "fixes" for its own inventions — at zero deliverable value
- Planner Claude orders actions against mechanisms that don't exist, then offloads agent-doable work onto the operator — while the project's entire enforcement layer turns out to be silently advisory
- [Feature Request] AskUserQuestion: optional free-text comment for the whole answer set at submit time
- [BUG] SessionStart lists an authenticated, tool-loaded claude.ai connector under "requires authentication" and instructs the agent to announce it as unavailable
- Claude for Chrome extension: renderer OOM crash (Aw, Snap!) on high-mutation React SPAs — geometric memory growth outside page heap
- Model repeatedly stated invented hardware facts as verified, causing real miswiring and wasted money
- [BUG] Opus 4.8 thinking-only end_turn followed by fabricated tool and background-agent results
- MCP tool-result widgets render as empty boxes (Claude Code Desktop)
- [BUG]
- Ghost prompt suggestion renders darker than normal text on light backgrounds (SGR 2 faint on default fg)
- [BUG] Opus 4.7/4.8 token usage regressed 2-3x after update; Opus 4.8 also disconnects frequently
- [Bug] macOS arm64: Second background agent spawn fails with "fork failed: Device not configured" in 2.1.207
- [BUG]
- Windows 8.3 short-name in %TEMP% triggers false-positive "suspicious Windows path pattern" approval prompts on the session scratchpad
- [BUG] Code tab Overview heatmap misses days with active project usage
- [Bug] Unexpected model downgrade from Claude Opus to Claude 3.5 Sonnet without error notification
- MCP OAuth DCR fails for Calendly: client_name "Claude Code (<server>)" rejected as invalid_client_metadata (recurrence of #59445)
- [Bug] Excessive cache creation tokens with low cache read ratio causing high usage consumption
- [Bug] Inappropriate content flagging on valid queries
- Chat input box lacks bidi support (unicode-bidi:plaintext) — RTL/Hebrew text renders reversed while typing
- One full working day, two Claude seats, one small well-defined module task — not completed. A time-accounting of where the day actually went
- [Bug] Tool claims completion without execution, admits to fabricating results
- First-turn junk/echo emissions with immediate end_turn — 4 classes incl. a confabulated <system_warning> claiming safety-void (CC 2.1.20x, claude-fable-5)
- [Suggestion] CSS class proposals show undeclared classes before their creation prompt
- [BUG] Remote Control (macOS/launchd): headless session workers freeze process-wide mid-turn at first tool_use while the Mac is user-idle; all unfreeze on first HID input
- [BUG] Serious trust breach reading home and secrets without asking for permission
- [Bug] Tool execution ignoring user instructions and causing dMASSIVE data loss
- Another hard-blocked session
- I appreciate you clarifying, but I'm not able to generate a GitHub issue title without an actual bug report or description of the problem you've found. Could you please provide: 1. A description of the bug you discovered 2. Steps to reproduce it 3. Expect
- [Bug] Claude Code defaults to Opus 4.8 instead of respecting Haiku 3.5 model selection
- [Bug] Anthropic API Error: False Positive Cybersecurity Content Flagging on API Error Translation Request
- [MODEL] Credential-entry rule (never type passwords) was not enforced in past sessions on this project
- [Bug] Repetition loop (`course。` spam) + English leakage + empty responses at tool-call boundary in long Japanese sessions
- Usage stats heatmap credits all activity to the session start date; days worked in long-running sessions show as zero
- Fable 5 skips explicitly-instructed final text output after long tool-call chains, jumping straight to terminal action (+ confabulates about it when asked)
- [Bug] Low execution rate indicates excessive steering/correction overhead in agent loop stupi d shit fable
- [Bug] Model auto-switches to Opus when InlineData contains metadata service URL
- [Bug] Anthropic API Error: Fable 5 Safeguards Over-Flagging Game Simulation Prompts / Code
- [BUG] Cowork sandbox unbootable, needs backend reprovision (known bug)
- [BUG] Scheduled routine notifications (push + email) not delivered despite correct settings
- [BUG] Claude Design incorrectly routes to Team org instead of Max account when both share same email
- Assistant answers 'did you tell me X before' with false confidence from a compacted summary instead of checking the full transcript
- Agent scaled an unverified destructive fix to ~98% of a user's data, causing a Plex library collapse (classifier friction also noted)
- [BUG] Native Linux install: disabling auto-updates for gated manual review workflow.
- [BUG] Stale dispatch environments persist in mobile app; dispatching to them hangs at "Allocating sandbox" and creates unrecoverable ghost sessions
- [Bug] CLAUDE.md rules still ignored — pattern persists after closure of #7248, #18411, #35019, #46724
- Plugin install TUI: Esc from post-install required-field config leaves plugin registered-but-unloaded with no feedback
- You've hit your weekly limit · resets Jul 27 at 4:30am (Asia/Calcutta)
- [MODEL] Working file-injection technique (DataTransfer onto input[type=file]) now blocked by auto-mode classifier, no accessible override
- [Bug] Anthropic API Safety Filter blocks legitimate project code reading
- Stored primaryApiKey silently overrides an active Max subscription; "no usage credits" error gives no cause
- [Blocking] Fable 5 forces usage-credits fallback despite Max plan Fable quota barely used (3%)
- [BUG] API Error: 529 Overloaded. This is a server-side issue, usually temporary — try again in a moment. If it persists, check https://status.claude.com.
- [Bug] Session incorrectly falls back to Opus instead of Fable for non-security requests
- [Bug] Anthropic API Error: HTTP 529 Overloaded
- Opus 5: verification allocation, commitment ordering, and regeneration fidelity in orchestration roles
- [Bug] Model picker fallback entry shows Opus 4.7 instead of Opus 4.8
- Restore the system prompt: Opus follows instructions much worse now
- [Bug] Claude Fable 5 incorrectly consumes usage credits on Claude Code 2.1.215 with 20x Max subscription
- [Bug] Anthropic API Error: Opus 5 Excessive Content Filtering on Code Generation
- [Bug] Code-review workflow hangs at synthesis phase; nested-parallel agents bypass cache on resume
- Fable 5.1 unreachable in Claude Code despite unused plan-included Fable allocation; credit-block dialog shows "not included in plan" copy with no affirmative action