Digest / May 29, 2026
Claude Code Issues Digest — May 29, 2026
370 new issues filed ·
460 resolved in anthropics/claude-code (UTC day).
New issues (370)
- [BUG] Session resume fails with 400 "Invalid model name" until /model is run (regression, started ~mid-May 2026)
- Subagent/Workflow creation forwards raw MCP tool schemas → 400 on connector tool with top-level oneOf/anyOf/allOf
- /resume command becomes unresponsive after window switch with NO_FLICKER enabled
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant message
- [BUG] Cowork installation fails with HRESULT 0x80073CFF on Windows 11 Pro - installer cannot remove orphaned package/service
- Claude Code repeatedly violates explicit 'confirm before action' instructions
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant message
- [Feature Request] Display full instruction contents in CronCreate loop confirmation
- [Feature Request] Allow /feedback command without including private session data!
- [Bug] Excessive token consumption in API requests
- [BUG] Workspace trust dialog re-prompts every session when CWD is home directory (non-git)
- [Bug] Fork from message history uses latest message instead of fork point
- context_window_size reported as 200000 for claude-opus-4-8 (1M-context model) → "100% context used" clamps prematurely
- [MODEL] Model repeatedly violates clearly stated instructions despite extensive reinforcement (memory, CLAUDE.md, in-conversation corrections)
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant message
- [Bug] Claude Opus 4.8 ignores MCP tool definitions, hallucinates parameters
- Channel notifications (notifications/claude/channel) buffered instead of injected when agent is idle
- [BUG] /goal stop hook fires after every assistant turn, interrupting user mid-message
- [BUG] API Error: 400 Failed to deserialize the JSON body into the target type
- [Bug] Claude Code consuming excessive tokens without completing simple tasks on Claude 3.5 Sonnet
- Opus 4.8 not selectable in CLI `/model` despite being available on account
- [BUG] v2.1.154 regression: custom agents via `/agents` fail with "messages[1].role: unknown variant `system`"
- [FEATURE] Desktop: expose worktree name/base input in UI + pass branch-prefix & base branch into the WorktreeCreate hook payload
- [BUG] Thinking text empty / "Thought for Xs" not expandable on Opus 4.7+ in VS Code extension — fallback thinking-config branch omits `display`
- [Bug] Claude Code API response latency degradation
- Phantom `⏺ Human:` lines appearing in transcript — assistant acts on instructions the user never typed
- [BUG] Unrecoverable `400 … thinking/redacted_thinking blocks … cannot be modified` permanently bricks a session — reproduced across two independent sessions (v2.1.154)
- Over 100% context and not triggering autocompaction
- Chat session rename doesn't update in sidebar until VS Code restarts
- [BUG] Intellij AI Chat Failed to initialize ACP session. Error: Internal error
- [Bug] Statusline effort.level reports "xhigh" instead of "ultra" for ultracode mode
- [BUG] API 400 "messages[1].role must be either 'user' or 'assistant', but got 'system'" (v2.1.156)
- Remote Control silently fails when HTTPS-scanning AV (e.g. Norton 360) intercepts TLS - Node rejects MITM cert chain
- API Error 400 with custom model via ANTHROPIC_BASE_URL when plugins are loaded
- Permission mode change away from "auto" doesn't stop the auto-mode classifier in the running session
- [BUG] Claude Code for VS Code API Error: 400 messages[1].role must be either 'user' or 'assistant', but got 'system'
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant messages
- Extended thinking + tool use: signature-only thinking blocks permanently break --resume (400 cannot be modified)
- [BUG] Opus 4.8: compaction partially overwrites visible chat scrollback above the summary (was previously retained)
- Feature request: Hard workspace boundary enforcement for project isolation
- /memory command fails to resolve symlinked user memory file in WSL
- [BUG] CLAUDE_CODE_DISABLE_1M_CONTEXT ignored in 2.1.156
- [FEATURE] Allowlist-only file access + deny rules for MCP tools (WebFetch, WebSearch)
- [Bug] Claude Opus 4.8: Tool call parsing fails when extended thinking is triggered
- Claude in Chrome (VS Code extension, macOS): @browser prompt injects + tab created, but mcp__claude-in-chrome__* tools never register (no native-messaging manifest / native host / bridge socket)
- [BUG] claude app crashes seconds after being opened
- [BUG] Bug Report: Loading spinner freezes + no response, correlated with repeated "malformed tool call" retries
- [Bug] Model performance degradation - Opus underperforming on code tasks
- asset_add_file MCP tool (Adobe connector) — file picker doesn't open in Claude Code desktop app (macOS)
- [BUG] Mismatched content block type content_block_delta thinking
- [Feature Request] Add timestamp display for conversation interactions
- [BUG] "Desktop app blocked by VS Code extension processes even without computer use enabled"
- [Bug] Screenshot tool repeatedly requests accessibility permission instead of using provided screenshot
- [FEATURE] Add `/mode <mode>` command to switch Claude Code mode
- [BUG] API 400 Error - role: system deserialization failure
- [FEATURE]
- [BUG] Opus 4.7 halts turn early after PreToolUse hook emits additionalContext on 2 calls
- SessionEnd hook reports 'Hook cancelled' even when hook exits 0 immediately
- Click-to-copy affordance on agent-designated content
- [Feature Request] Separate cached token tracking for brainstorming sessions from main session
- [Bug] /effort shows "Ultracode needs dynamic workflows enabled" even when dynamic workflows are on
- [BUG] /compact fails with cyber safeguards false positive during legitimate defensive security session
- [BUG] Fullscreen TUI default (~v2.1.150) breaks macOS VoiceOver / screen-reader access in Terminal.app
- [BUG] /terminal-setup writes to wrong path for VSCode installed in portable mode (e.g. via Scoop)
- [Bug] Session context lost after overnight inactivity or resumption
- [Bug] Insufficient error information provided
- [BUG] Windows: host CPU pressure starves Claude Code input loop — `claude agents` arrow keys/typing dead until load drops
- [Feature Request] Agent should prioritize Read tool before Write/Update operations instead of using cat
- [BUG] Claude Code keeps flagging on it's own injections
- [BUG] Server-side tool (advisor) bricks session: auto-rename system-reminder injected between server_tool_use and tool_result causes 400
- Headless + interleaved thinking + tool use: thinking block intermittently mutated → 400 "thinking blocks in the latest assistant message cannot be modified"
- Context/compact button triggers compact on single click with no confirmation or undo
- [BUG] VS Code extension sends invalid system role in messages array
- [Bug] Excessive token consumption during framework analysis and code generation
- [Bug] Anthropic API Error: thinking blocks in assistant message cannot be modified
- [BUG] Claude code admits to costing me money and tokens
- PowerShell tool disabled under MINGW64/Git Bash on Windows
- Max plan: Claude Code may compress context at 200K instead of using extended context window
- PR widget: GitHubPrManager drops --repo when session cwd is a non-git parent dir
- [BUG]
- [BUG] Claude desktop installer fails with HRESULT 0x80073CF6 on Windows 11 Home
- [Bug] Tool invocation fails with corrupted strings on WSL
- [BUG] Claude Code doesn't work (you write any question, thinks for 5 minutes and stops), Claude Cowork and Claude Chat works.
- Feature request: Add cursorLineStart/cursorLineEnd actions for Chat input
- [ultrareview] Two consecutive crashes on large diff — free credits consumed without findings
- [BUG] Claude Opus 4.8 makes changes outside the current worktree for the current session
- Session crashed with "Usage credits required" error after auto-switching to 1M context mode
- Built-in `Grep`/`Glob` tools unavailable when `ENABLE_TOOL_SEARCH=1` (not in active set, not discoverable via ToolSearch)
- claude.ai/code constantly redirects to /code/family (Firefox)
- [BUG] Session history empty in VSCode extension when working directory is on a mapped Windows drive
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant messages
- [Feature Request] Expose Fast mode state to statusLine scripts
- [BUG] Switching devices leads to stale prompt replay if one device has network partitioned
- [BUG] ClaudeCodeManager-VM Checksum Mismatch on macOS Apple Silicon — v1.9659.2
- [BUG] /context shows 0 tokens for all MCP tools on Bedrock — countTokensWithFallback fails with 400 tools.0.custom.eager_input_streaming: Extra inputs are not permitted
- [BUG] Setting only disableAutoUpdates locks Cowork and third-party inference panes as org-managed (macOS com.anthropic.claudefordesktop)
- [BUG] VS Code sidebar chat: API Error 400 - messages[1].role unknown variant system
- I'm unable to generate a GitHub issue title from this report because the description "best in the new ultracode" doesn't contain enough technical information about a specific problem, error, or feature request. To create a proper issue title, I would need
- [BUG] OTel api_request model attribute drops [1m] suffix while runtime serves 1M context — Cost-by-model dashboard misattributes ~50% of Opus spend
- [Bug] Session restore creates truncated chat history from claude agents list
- Model fabricates tool output (and even a user instruction) when a parallel batch is partially cancelled / appears empty
- [Bug] Commands hanging/freezing indefinitely during session execution
- Workflow tool: script parse errors always blame "TypeScript syntax" even for plain-JS errors (misleading)
- Cloud environment "Environment variables" aren't available to the Setup script
- [BUG] Native memory leak (~1.5 GB/min RssAnon) in interactive mode — regression between 2.1.140 and 2.1.150
- New site
- v2.1.154+ breaks third-party API compatibility - system role injected mid-conversation
- Claude Code inside tmux: scroll hits input not scrollback, no rich mode ("waiting for output"), transcript not saved
- [Bug] Opus 4.8 on Bedrock: "thinking.type.enabled" not supported — regression from #51439
- [Regression 1.9659.2] computer-use tools disconnect after first use in Code mode (Windows)
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant message
- [BUG] Account connector sync overwrites claude_desktop_config.json, wiping manually-added MCP servers and re-adding removed ones
- [BUG] Stale "Pull request status couldn't be checked" on archived sessions whose repo/worktree was deleted
- [BUG] No way to reassign or remove session-to-project associations after the assignment feature was removed
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant message
- [Bug] Advisor tool result split across messages causes 400 on session resume
- Claude provides incorrect guidance about its own /agents feature, causing ~2M token waste
- [FEATURE] In-product feedback-to-remedy loop for Max plan users
- [Feature Request] Add diff view for plan updates to track changes across iterations
- [Bug] Agentic loop terminates prematurely without completing task
- [BUG] Permission prompt immediately steals focus while typing (regression in 2.1.139+)
- [Bug] Frequent ECONNRESET errors with failed retry logic in autonomous sessions
- [BUG] | v2.1.156: settings.json "model" field silently ignored for interactive sessions
- [Bug] Auto mode display broken when tabbing with Claude 4.8 models
- feat: /mcp query <tool> <args> — invoke MCP tools directly from the slash command interface
- [Bug] Anthropic API Error: Cannot Modify Thinking Blocks in Assistant Messages
- [BUG] Claude Code on windows Dispatch
- Behavior report: over-orchestration + unverified factual claim + self-violated instruction on a small docs task (Claude Code vs Codex)
- [BUG] /claude-api bundled skill saturates context unconditionally — ~77% spike from neutral question
- [Bug] Anthropic API Error: 529 Overloaded
- API 400 error: thinking blocks modified in /plan mode conversation history
- [BUG] Compatibility issue with DeepSeek API: "unknown variant 'system'" error in messages
- Unable to use Practitest app using chrome extension
- [Bug] Anthropic API Error: Model temporarily unavailable blocks Edit operations in auto mode
- [Bug] Anthropic API Error: Output blocked by content filtering policy
- [Bug] Claude Code ignoring user requests to perform visual checks
- [Bug] Anthropic API Error: Cannot modify thinking or redacted_thinking blocks in assistant message
- [FEATURE] disableBypassPermissionsMode should accept booleans (managed settings)
- [Bug] Parallel tool calls trigger persistent "Cancelled" errors
- [Bug] High-effort session tier not propagated to sub-agents with pinned models
- `claude agents` doesn't show running agents
- [Feature Request] Expand all collapsed output blocks with single command
- VSCode extension: assistant tool calls rendered as literal text in chat, not executed
- [Bug] Anthropic API: Opus 4.7/4.8 Missing Tool Use Block with tool_use Stop Reason
- [Bug] JSON parsing error in Claude Code CLI
- [Bug] Opus 4.7 & 4.8 — silent stuck-turn: stop_reason=tool_use with no tool_use block (Sonnet 4.6 unaffected)
- AbortError in hook callback wedges query mid-stream; stdout silent, no result, no recovery
- [BUG] Skills upload returns "Failed to fetch" on Claude Desktop
- [FEATURE] Claude Desktop - Archive individual conversations from the sidebar
- [Bug] Thinking block signature rejection causing repeated strip-and-retry with increased latency
- Session list PR status chip always renders greyed out despite gh + backend reporting correct state
- [Feature Request] UI colors should respect system dark/light mode preferences
- [Bug] autocompact 调度路径在 v2.1.156 静默失败: 状态栏正确报告阈值但压缩动作不触发
- claude -w <name> --tmux parent process does not exit after the spawned tmux Claude session ends
- [FEATURE] CLAUDE_CODE_SHELL applied to the shell used by hooks
- [security] Built-in secret redaction at jsonl write-time in ~/.claude/projects/
- [Feature Request] Hide /install-slack-app from the command menu once the Slack app is already installed
- [VSCode Extension] Laggy keystroke input in chat box
- Bash tool does not inherit PATH from launching shell environment
- [Cowork] computer-use cu-esc handler kills session with pending permission dialog
- False verification: confirmed code changes were live without reading files
- [BUG] effort: frontmatter silently ignored in commands/*.md slash commands (works in skills/SKILL.md)
- Office 365 MCP connector: add send email capability (currently read-only)
- [BUG] Long-session compaction loses Write/Edit tool-use history → model misattributes its own work to the user
- Opus 4.8 asserted a false answer on a trivial lookup without verifying first
- [BUG] Opus 4.8 repeatedly emits malformed tool_use blocks, entire response discarded (4.7 works fine)
- [BUG] Claude responds again to a /btw question every time detailed transcript ( Ctrl+ o ) is toggled
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant messages
- [BUG] API Error 400: thinking blocks must remain as original response - claude-opus-4-8 incompatibility
- [Bug] Harness modifying thinking blocks causes Anthropic API rejection
- Illegal Instruction crash on CPUs without AVX since v2.1.113
- [Bug] Excessive token consumption for SVG legend modifications
- `/model` picker stores `opus[1m]` then next-turn validator rejects it as "may not exist" (Claude Max, Linux, `--resume`d session)
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant messages
- You've hit your session limit · resets 6:20pm (Europe/Helsinki)
- You've hit your session limit · resets 6:20pm (Europe/Helsinki)
- [BUG] Korean word '영역' excessively and incorrectly inserted in responses — Opus 4.7, 4.8
- [BUG] Claude Code fails BASH invocations
- [BUG] No logout mechanism in VSCodium extension — credentials irrecoverable without CLI
- Feature Request: Session rename and folder/tag grouping
- [BUG] CCR manual trigger hangs at "Setting up a cloud container" — scheduled runs complete fine
- [BUG] Stop hook evaluator fails with 400: "output_config.format: Extra inputs are not permitted" (Bedrock)
- Claude Code Access Disabled After Payment - Account: seodepartment1@gmail.com
- [BUG] [Desktop App] Claude desktop app not launching on Windows 11 ARM64/Snapdragon X
- Windows: MCP server spawning causes brief console window flash (missing CREATE_NO_WINDOW flag)
- [BUG] Disabled plugin state in ~/.claude/settings.json survives uninstall + reinstall
- [DOCS] `plugin update` docs do not specify whether existing dependencies of a meta-plugin are re-resolved
- [Bug] Anthropic API Error: Mismatched tool_use_id in advisor_tool_result block
- Auto-compact ignores CLAUDE_AUTOCOMPACT_PCT_OVERRIDE on 1M-context Opus — no compaction at 52% (522k) with override=40 [v2.1.156]
- [Bug] Anthropic API Error: 500 Internal Server Error should auto-recover
- Opus over-confirms on already-authorized low-risk steps, then silently blocks
- [Bug] /context display shows >100% (190%) due to wrong denominator — output token budget used instead of input context window
- [Bug] Agent view attachment lacks scrollback history for background sessions
- Intuit QuickBooks connector fails for accounts without QuickBooks Payroll
- [BUG] Cowork Linux sandbox never mounts on Windows MSIX — VM bundlePath resolves to un-virtualized Roaming\Claude while bundle data lives in the package container
- [Bug] /compact fails with "Usage credits required for 1M context" even after /model set to Sonnet 4.6
- Bug: Segmentation Fault in native binary on Ubuntu 26.04 (exit code 139)
- [BUG] Pro account, Cowork tab greyed out, yukonSilver not supported (status=unsupported) in cowork_vm_node.log.
- [Feature Request] Support Bedrock Guardrails Prompt Attack filter by injecting guard_content input tags
- [BUG] [Plugin: Telegram] Inbound notifications never delivered to session — outbound works fine
- [BUG] mcp__claude-in-chrome__* tools are advertised but not callable after upgrade to Claude 4.8
- Mobile app autolinker includes markdown bold ** delimiters in URL href, causing 404 on tap
- [Windows] Skill shell execution opens interactive mintty (Git Bash) window that does not close
- [BUG] Claude hangs with no output on Linux Mint despite all-green
- [BUG] Claude Code unable to reproduce a canonical DOCX design across varied source docs — hours of iteration with no working result
- Claude ran 'docker compose down -v' without user permission, destroying database volumes
- [Critical] advisor() double-counts its sub-inference into top-level context usage → silent mid-task context destruction + uncapped uncached re-billing; spec violation; advisor double-count open & unfixed for 5+ weeks (#53065) in a class reported since Feb (#24976)
- Allow model selection for Slack-triggered Claude Code sessions
- [FEATURE] Hookable Workflow lifecycle events (completion) that support additionalContext
- Title: Remote control: cloud→local input channel broken (account-scoped);outputchannelworks
- Background task notifications should include exit code and output tail
- [BUG] Claude Desktop blurry on high-DPI ultrawide display due to MSIX packaging blocking DPI override
- Claude Desktop: Auto-update loses spaces.json, causing empty project list
- Add custom session label field in extension header
- [BUG][Desktop App] Context token counter removed from bottom bar
- /model picker does not list newly available models (Opus 4.7, 4.8)
- [Bug] Auto-update mechanism not functioning
- [Feature] Per-invocation --fast flag for headless claude -p
- [Bug] Workflow retry loop exhaustion due to malformed JSON field construction
- [BUG] Default route in Settings not applied when starting a new project
- [BUG] rate_limits absent from statusLine JSON for Claude Max plan (v2.1.156, Windows)
- [BUG] In the terminal, when scrolling up through Claude Code's output panel, the same content block often appears duplicated/repeated multiple times.
- Heads-up: ultracode / dynamic workflows can chain-spawn local processes and spike host memory
- [BUG] Claude Code Desktop App sessions not tracked in Organization Analytics
- enabledPlugins in settings.json not persisting across restarts
- [BUG] ultraplan: cannot launch remote session
- Window controls overlap app UI on Windows with Hebrew (RTL) locale
- Claude Code silently escalates to 1M context model and bills usage credits without user consent, bypassing auto-compaction
- Versioned install path causes TCC re-prompts on every Claude Code update (breaks MDM PPPC; CoWork scheduled actions may be affected)
- [MODEL] Ignoring .md, my instructions
- [BUG] Auto-mode permission levels don't compact context.
- Up/Down arrows jump to history instead of moving cursor through wrapped multi-line prompt input (all platforms; regressed ~2.1.15)
- Versioned install path causes TCC re-prompts on every Claude Code update (breaks MDM PPPC; CoWork scheduled actions may be affected)
- [FEATURE] Show model name in subagent/session list view
- [Bug] Anthropic API Error: Invalid JSON Schema in Tool Definition
- [Bug Report] Stop hook (goal condition) loops with no user input when the condition is subjective/unverifiable
- [BUG] Session working directory diverges from the directory Claude Code was launched from
- [BUG] Cannot selectively copy text from assistant response — only full-message copy works
- [BUG] Using the 2.1.156 version of the Claude Code for VS Code extension
- [BUG] [Tool result missing due to internal error] is unattributed, undiagnostic, and hangs the agent
- [Bug] Incomplete bug report submission
- [BUG] git-subdir plugin install leaks sparse-checkout into the session's working repo, silently collapsing the tree
- [Bug] Tool Call Parsing Failed - Unable to Parse Model Response
- [BUG] marketplace add silently overwrites existing marketplace when two repos declare the same marketplace name
- Model switcher dropdown in chat UI (like Continue.dev)
- [BUG] Agent Teams (in-process): unbounded team-lead inbox stalls teammate→lead message surfacing at ~1MB / 1000+ msgs
- Your organization has disabled Claude subscription access for Claude Code · Use an Anthropic API key instead, or ask your admin to enable access
- Opus 4.8 (1M context) — frequent "tool_use malformed" client errors despite tools executing successfully
- [Bug] Anthropic API Error: Usage credits required for 1M context window
- Web hand-off session cannot commit or push — work stranded in ephemeral container (commit-signing 400 'missing source' + no remote/gh)
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant message
- Feature request: more descriptive and customizable session names in --resume
- [BUG] Customize → Skills file-tree only renders the first 10 files in a folder; additional reference files are hidden (desktop app + claude.ai)
- [FEATURE] Expose model configuration for subagents in dynamic workflows
- [DOCS] Plugin .mcp.json example incorrectly uses mcpServers wrapper
- Warning 'Native installation exists but ~/.local/bin is not in your PATH' fires even when ~/.local/bin is in PATH
- [BUG] Regression in 2.1.153: claude-in-chrome browser tools no longer load in VS Code extension (worked in 2.1.152)
- [FEATURE] Configurable pushback / proactive review mode
- [BUG] Workflow subagent: one stall is retried 6× at full token cost (no across-retry progress check); the no-progress signal can't tell slow-but-alive from hung
- [BUG] Admin settings: "Allow remote control" toggle is greyed out unless "Code on the Web" is enabled (Team plan)
- [BUG] "You've hit your usage limit"
- [Feature Request] Add periodic reconnection retry logic for network disconnections
- [BUG] VS Code sidebar extension ignores CLAUDE_CODE_DISABLE_1M_CONTEXT on Pro plan (Windows)
- [BUG] API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited
- App Build with Terrible Results
- [FEATURE] Archive (hide-but-keep) chat sessions in VS Code sidebar
- [FEATURE]
- [FEATURE]
- [FEATURE] A Health Bar for tokens
- Model Opus 4.8 False positive safety block: innocent Gmail MCP question blocked as "cyber threat"
- Feature Request — Conversation sidebar organization (folders / project grouping)
- Voice input stopped transcribing after updating to 2.1.156 in VS Code on Windows 11
- [Bug] Output duplication and formatting issues with Claude Opus model
- [BUG] Stopped due to usage limits being reached, but I’m only at 64%.
- [BUG] "Resuming your cloud container" stalls forever and now I have lost all my work.
- sed on Windows silently converts CRLF to LF, breaking files
- [BUG] Claude desktop renderer crashes - MacBook Pro
- [Bug] Anthropic API Error: Output blocked by content filtering policy after token consumption
- [Bug] Context window exceeding 200k limit with Sonnet-4.6 causes unrecoverable session state
- [Bug] Anthropic API Error: Tool Call Parsing Failure
- [BUG] Web session stuck indefinitely on "Resuming your cloud container": never starts Claude Code
- [BUG] /desktop command fails when Claude Desktop is installed via Anthropic's MSIX installer (Windows)
- [BUG] [Enterprise] CoworkVMService deadlocks MSIX updates on Windows 11 — fleet-wide
- [Bug] Unable to remove marketplace-installed plugins via spacebar/enter
- /clear deletes chat history permanently
- Workflow tool triggers on keyword 'workflow' in normal conversation
- [Bug] Clicking links causes Claude Code to hang or freeze
- claude-in-chrome: CLI bridge never connects (list_connected_browsers empty) despite extension installed + logged in
- [Feature Request] Improve "workflow" keyword bypass UX and documentation for meta key binding
- Claude Code persistently strips Spanish accents (tildes) and ñ in generated output despite extensive user mitigations — bimodal, all-or-nothing per file
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant messages
- Agent/subagent spawn fails with 'Prompt is too long' even for tiny prompts (large MCP tool set)
- Edit tool silently substitutes Unicode smart quotes for ASCII double quotes in shell scripts
- `/insights` HTML report truncates bar-chart category labels
- [Bug] GitHub auth defaults to token over SSH, leaking secrets in chat and /feedback transcript
- [Bug] Anthropic API Error: 529 Overloaded
- [Bug] Claude model hallucinating in image analysis and getting stuck in circular reasoning loops
- [Atlassian MCP] Add getJiraIssueWatchers tool
- [BUG] Grep and Glob Tools Unusable
- [Docs] Workflow tool args parameter is undocumented; nested-array gotcha causes runtime errors
- [BUG] Plugin installer uses incorrect (lowercased) home directory path when username contains uppercase letters or dots
- [BUG] Voice input stopped transcribing after updating to 2.1.156 in VS Code on Windows 11
- Field notes: StructuredOutput as a control-channel (a multi-agent workflow that died) + Agent-SDK sharp edges
- Regression (2.1.149 → 2.1.156): 1Password `op` CLI session broke in Bash tool calls on macOS
- [BUG] Max plan requires Usage Credits balance for 1M context — should be included
- FleetView: Archive completed agents into a collapsed group
- [Feature Request] Isolate subagent execution environments to prevent tool call (and other) conflicts
- MCP channel push: non-blocking delivery mode for active sessions (multi-agent coordinator use case)
- [Bug] Auto-submit on paste prevents editing in ultracode mode
- [FEATURE] Reduce duplicate MCP server processes when multiple plugins wrap the same binary
- Opus 4.6 failure report: confident wrong claims, unverified agent output, not listening
- [BUG] AUP/cyber-safeguard false positives on legitimate own-software hardening; one hit contaminates entire session
- [Bug] Overly aggressive content filter blocks legitimate infrastructure debugging of timeout issues
- [BUG] Worktree branch prefix preference (ccBranchPrefix) ignored in v2.1.156 — hardcoded claude/MMDD- used
- [BUG] Cowork — account enrolled in two A/B variants (`testfoo` + `0526`) causing skill misfires + connector failure; `ENABLE_TOOL_SEARCH` mismatch
- [FEATURE] Add workflow tool to cloud scheduled tasks
- [BUG] Auto-mode classifier blocks benign signed-API diagnostic with no in-CLI escalation path
- [DOCS]
- [FEATURE] expose session ID to external tools inspecting a running claude process
- [BUG] Workflow tool determinism guard false-positives on new Date(/Date.now(/Math.random( substrings inside prompt strings
- Allow annotating settings.json entries (esp. permissions.allow) — JSON forbids comments
- [Bug] Anthropic API Error: Usage credits required for 1M context with 1M context disabled
- Dynamic workflows ignore the subagent tools: allowlist (always grant Write/Edit)
- [MODEL] generated a migration that dropped a column without explicit instruction, on a live database.
- [BUG] Opus 4.8: Repeated/duplicated output blocks in Claude Code CLI, worsening with session length
- Renamed session titles don't persist + first user prompt ignored + panels reflow on new content
- [BUG] Bash and Read tool outputs arriving one turn late
- [BUG] Desktop app PR status bar shows closed/merged PR on main (still reproduces on 2.1.156; #48694 closed without a fix)
- [DOCS] Need a process for migrating from a locally-installed Claude installation to one provided by the Debian/Ubuntu Package Manager
- [BUG] /model switch to standard context doesn't clear 1M context credits gate — /compact still fails after switching back
- [DOCS] Plugins in `.claude/skills` auto-load not documented — marketplace not required
- [DOCS] `claude plugin init <name>` command missing from plugin documentation
- [DOCS] Plugin docs missing autocomplete for `/plugin` arguments: subcommands, installed plugin names, and marketplace plugins
- [DOCS] `claude agents` dispatched sessions missing documentation for `agent` setting and `--agent` flag
- [DOCS] EnterWorktree mid-session worktree switching not documented in worktrees guide
- [DOCS] tool_decision OTel event documentation missing tool_parameters attribute
- [DOCS] Worktrees page doesn't document that Claude unlocks worktrees when the agent finishes so `git worktree remove`/`prune` can clean them up
- [DOCS] Claude directory swept-paths table omits `.claude/worktrees/` for background agent worktrees
- [Feature Request] Add customization options for AgentView component
- [DOCS] Worktree docs do not clarify `--worktree` and `--tmux` behavior when invoked from inside a linked git worktree
- [DOCS] WSL: dragging images from Windows Explorer into Claude Code terminal is undocumented
- [DOCS] `/terminal-setup` GPU acceleration disablement for VS Code/Cursor/Windsurf undocumented
- [DOCS] Agent view dispatch input slash-command autocomplete matching behavior is undocumented
- [DOCS] Workflows docs missing backspace shortcut for dismissing workflow trigger keyword
- [DOCS] Settings page missing "Workflow keyword trigger" setting added in v2.1.157
- [BUG] `telegram` plugin (claude-plugins-official) — frequent poller crashes, `bun` launch failures, and silent inbound drop
- [BUG] PreToolUse hook for Write/Edit can be bypassed by writing files via the Bash tool
- [MODEL] Claude reasoned around its own PreToolUse hook and executed a bypass without user notification
- enter key not working
- [BUG] API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited
- [Bug] Session transcript corruption when executing commands during server tool execution
- [Bug] Anthropic API Error: Unable to Continue
- Opus 4.8 + dynamic tool loading (ToolSearch) re-modifies signed thinking blocks → "messages.N.content.M ... cannot be modified" 400 storm → strip-and-retry every turn
- [Feature Request] Allow "workflow" as non-reserved word in prompts
- [Bug] Unauthorized workflow tool execution consuming tokens without user consent
- [Bug] Opus 4.8 performance regression: excessive latency with degraded quality and context loss
- [Bug] Anthropic API Error: Model unavailable prevents safety classification, unsafe actions proceed
- Bash/Read tool results intermittently return empty (then flush late, redundantly) on Linux — NOT length/concurrency-correlated; same symptom as stale-closed #36038
- [FEATURE] VS CODE STUDIO'DA KONUŞMA PENCERELERİNE İSİM VERMEK.
- Cycle active agent from the prompt (agent:next / agent:previous) without entering the footer
- [Bug] opus 4.8 hangs indefinitely on tool calls after extended usage, freezes on messages ending with "let me..."
- [BUG] Windows: Sessions show "Session not found on disk" after project folder moved - no UI recovery option
- [Bug] Claude Code ignores claude.md instructions and prioritizes expedient solutions
- [BUG] Reply to earlier message wipes all subsequent conversation history from context
- [UI Bug] Large markdown paste causes conversation overlap + no collapsible 'Show more' on chat bubbles (VS Code Extension)
- /insights: prescriptions need repo-grounding; one trust-eroding security false-positive
- AskUserQuestion blocks invisibly behind the thinking spinner (no prompt shown) when extended thinking is active
- Claude-initiated context compaction: working workaround + native implementation proposal
- MCP client partial reconnect: one streamable-HTTP namespace stays disconnected after server restart while sibling namespaces re-bind
- Cowork Dispatch: results from long-running spawned sessions never relayed back to the thread (after-the-fact readback broken); persists across updates on 1.9659.2
Resolved issues (460)
- [BUG] Cowork installation fails with HRESULT 0x80073CFF on Windows 11 Pro - installer cannot remove orphaned package/service
- scheduled-tasks dispatcher stopped honoring SKILL.md `model:` frontmatter on/around 2026-04-27 ~01-02 UTC
- 2.1.154 regression: bridge/remote-control session replay re-sends persisted (empty-text, signed) thinking blocks -> 400 'thinking blocks cannot be modified', session unrecoverable
- [BUG] Claude Agents Plugin Skills Autocomplete is inconsistent with a normal session
- Permission mode change away from "auto" doesn't stop the auto-mode classifier in the running session
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant message
- [BUG] API 400 Error - role: system deserialization failure
- Claude in Chrome (VS Code extension, macOS): @browser prompt injects + tab created, but mcp__claude-in-chrome__* tools never register (no native-messaging manifest / native host / bridge socket)
- Max plan: Claude Code may compress context at 200K instead of using extended context window
- [BUG] Claude desktop installer fails with HRESULT 0x80073CF6 on Windows 11 Home
- [BUG] security-guidance hook blocks markdown/doc writes containing "exec(" substring
- security-guidance: Stop-hook LLM review broken on Windows when worktree has many untracked files (WinError 206)
- security-guidance plugin: eval_injection rule falsely flags ast.literal_eval()
- [BUG] VS Code sidebar chat: API Error 400 - messages[1].role unknown variant system
- v2.1.154+ breaks third-party API compatibility - system role injected mid-conversation
- [BUG] Permission prompt immediately steals focus while typing (regression in 2.1.139+)
- [BUG] Compaction fails with "context window limit" error even when context usage is low (e.g., 20%) — regression in v2.1.153
- [Bug] Unannounced quota changes
- Cloud environment "Environment variables" aren't available to the Setup script
- [BUG] 鼠标滚轮在 Android Studio 集成终端中无法滚动（功能回退）
- Sandbox cannot create subdirectories under project tmp path
- No UI to restore/unarchive archived sessions in Desktop app
- CLI-created sessions invisible in Desktop app (reverse direction broken)
- ultraplan: 'Claude GitHub app must be installed' despite app being installed
- [BUG] 双击 ESC 回滚会话后，Task 列表未清理
- [BUG] When opening a terminal in the company's internal CodeLab container and entering Claude Code, the output often shows garbled characters, especially for Chinese characters. English characters are mostly fine.
- [Bug] Non-verification error causing repeated failures in Claude Code actions
- Tool-call opening tag gets stuck on a malformed form and is unrecoverable within the session
- [FEATURE] Show pending diffs in IDE before edit execution
- [VSCode Extension] AskUserQuestion panel hides conversation context with no collapse shortcut
- Drag-drop and screenshot clipboard paste fail on Linux; works only after re-copy from image viewer
- [BUG] [VSCODE] Algebra formatting broken on VS Code extension
- [Bug] Environment variable expansion not working for $HOME in paths
- [Bug] Mac Desktop App Crashes Repeatedly During Session
- [FEATURE] Support audio file attachments (WAV, MP3) in Claude Code
- [Bug] Ultrareview synthesis step crashes after backend completes findings analysis
- [Feature Request] Add command to manually mark task as finished in agents view
- [Bug] Malformed tool calls in long sessions due to in-context few-shot poisoning
- [bug] CC pinned tasks panel + macOS terminal scrollback → process hangs in __write_nocancel, kill -9 ineffective
- UX feedback: hard firewall between Claude Code and Claude Cowork makes cross-surface work invisible
- [Bug] Anthropic API Error: Empty Text Context Blocks in Messages
- [FEATURE] Remote session revocation / active sessions management
- Verbosity fix: gate on decision load per response, not line count
- [BUG] Desktop app Claude Code: cached auth state survives quit/relaunch after billing restoration (personal Max)
- Feature Request: Programmatic Context Clear (Tool or Skill Primitive)
- Claude Code web: image upload fails with API Error 400 'cache_control cannot be set for empty text blocks' on messages.2 in new session
- [FEATURE] Add persistent defaultSessionColor setting
- Bash tool stuck at 'Waiting...' - permission prompt not displayed
- Bug: Claude Preview MCP panel opens in wrong session (multi-session Desktop)
- [BUG] Claude Code Web sandbox can't pull Git LFS
- [Feature Request] User-defined instruction priority weighting for custom rules
- [Feature Request] Make diff syntax highlighting colors configurable
- [FEATURE] Streaming response hook for external voice and accessibility integrations
- [BUG] Statusline shows 0% context usage after exit/re-enter, contradicts `/context` command
- [FEATURE] Append Current Timestamp to Every Output
- [Feature Request] Add API support for research context exemptions in safety filters
- [BUG] Illegal instruction: 4 (SIGILL/BMI2) on 2013 Mac Pro (Xeon E5 v2 / Ivy Bridge) — regression after 2.1.30
- Docs: skill discovery across nested git repos via parent-walk and `additionalDirectories` doesn't match documented behaviour
- Semble Hooks — semantic code search injected into every prompt
- [CCR Beta] Routine sessions stuck in infinite loop after API error — Esc key does not stop execution
- [Bug] Changes applied without user confirmation in manual edit mode
- You've hit your weekly limit · resets May 28 at 7pm (Europe/Madrid)
- [BUG] VSCode Extension Description in left side bar not updating
- [Bug Report] Incomplete bug report submission
- [BUG] Worktree isolation spams duplicate custom-title entries, causing title 'sticking' across sessions
- You've hit your session limit · resets 1:10pm (Europe/Madrid)
- ✨📲 {{ 🎀^【【%#@&*^#🎀^✚卍 卍Flynas infolinia polska 24h %#@&*^#🎀^】✚卍 卍】✨📲 Czy Flynas oferuje całodobową obsługę klienta?%#@&*^#🎀^】】✚卍 卍⁂}}{{ 🎀^
- 🍉 ⋆ 🍓 🎀 ✚卍 卍 卍 +48 58 8810 086 卍 卍 卍✚ 🎀 ❀ 💗 🍓 ⋆ 🍉卍【２➃𝕙ㄖ尺𝔸Ŝ {{{Flydubai infolinia polska 24h}}】Numer kontaktowy Flydubai 【２➃𝕙ㄖ尺𝔸Ŝ{{Jak skontaktować się z obsługą klienta Flydubai? ?⁂}}{{ 🎀^ }}}】卍✚２➃𝕙ㄖ尺𝔸Ŝ{{{❀ 💗 🍓 ⋆ 🍉
- 🍉 ⋆ 🍓 🎀 ✚卍 卍 卍 +48 58 8810 086 卍 卍 卍✚ 🎀 ❀ 💗 🍓 ⋆ 🍉卍【２➃𝕙ㄖ尺𝔸Ŝ {{{Qatar Airways infolinia polska 24h }}】２➃𝕙ㄖ尺𝔸Ŝ{{ Numer kontaktowy Qatar Airways【 Jak zadać pytanie na temat Qatar Airways? 】}}卍✚２➃𝕙ㄖ尺𝔸Ŝ{{【❀ 💗 🍓 ⋆ 🍉
- [BUG] Payment declined for Claude Pro subscription from Nicaragua with multiple cards
- [Generative UI Widget] 默认字号过小，建议提高默认值或支持用户自定义字号
- Plan mode: allow inline back-and-forth commenting on the plan (Google Docs style)
- Auto mode: classifier that picks model (Haiku/Sonnet/Opus) and thinking effort per task
- [Feature Request] Add prompt history/undo functionality for multiline input
- [BUG] macOS Desktop: Permission prompt has no Enter keybinding — requires ⌘⇧↵ / ⌘↵ only
- Cowork runaway task — agent runs 4+ hours without self-terminating, UI timer active but no meaningful progress
- Duplicate .claude/rules loading in worktrees nested inside .claude/
- [BUG] Claude Code Can't Remember It can Set Crons
- [Bug] Unintended git reset --hard executed without user confirmation
- [BUG] Cowork tab not appearing in Claude Desktop — Windows 10 Pro, Pro Annual Plan
- [MODEL] Opus-4-7
- [BUG] Skills with `context: fork` fail with "disable-model-invocation" when invoked via Skill tool (v2.1.113)
- v2.1.98 regression: rw bind mounts inside ~/.claude shadowed by parent ro mount
- [BUG] UI Issue on Mac laptop
- [BUG] SSH session startup: 13+ min per-session delay from sequential per-file plugin sync + no UI toggle for bundled Anthropic skills
- He ampliado mi límite mensual y he recargado dinero, pero...
- [BUG] Claude Code: recurring verification-gap and decision-boundary failures across sessions; session refund requested
- [BUG] Atlassian Plugin (Claude Marketplace) Bug
- Permission allow rules with ** glob not matching paths outside working directory
- Expose current permission mode in statusline JSON
- [BUG] I cannot uninstall Claude Code because I do not have an active paid account.
- Bulk-delete old session transcripts in the desktop app
- [BUG] Remote control doesn't work after you hit your limit
- [BUG]
- False positive AUP block during legitimate CI/CD VM provisioning session (claude-opus-4-7, 1M context)
- Feature: expose session token usage to hooks and statusline commands
- [BUG] Mac desktop app: Black overlay from sidebar covers chat content
- @ file picker should support fuzzy filename search in additionalDirectories
- [MODEL] Explore subagent system prompt has contradictory instructions for Bash vs built-in tool usage
- [Bug] Terminal output exceeds viewport width, requires scrolling to view complete responses
- Agent Teams: model parameter does not support version pinning
- [FEATURE] Add hook event for extended thinking / LLM reasoning blocks
- There's an issue with the selected model (claude-haiku-4-5-20251001). It may not exist or you may not have access to it. Run /model to pick a different model.
- Cold start context usage increased significantly between versions
- [BUG] who am I?
- [FEATURE] env var that can resolve to marketplace location
- [Bug] Agent performance degradation after instruction violation feedback
- [BUG] Cypress commands are not performing properly in the VS Code native UI
- Persistent list of spawned tasks instead of chip-only UI
- Deny rules with root-level /** patterns silently match nothing
- Fixed role-based numeric hotkeys (1/2/3) for permission prompts — consistent, fast, safer
- [BUG] Cowork fails - VM cannot reach Claude API, Default Switch is Internal type with no external uplink (Windows 11 Pro, v1.3883.0)
- [Bug] Claude continues autonomous execution after context auto-compaction without user confirmation
- Compaction fails: max_tokens must be greater than thinking.budget_tokens
- Feature request: context trim-tail option (preserve recent, drop oldest)
- [Bug] Excessive token consumption on simple queries (25k tokens for "what is 1+1")
- Feature: Selective context compaction — allow discarding context before a checkpoint
- [FEATURE] Bundle a `/browse` skill that delegates browser MCP work to a subagent — prevents per-call context bloat in main session
- [BUG] Claude session keeps getting Compacting Conversation
- GrowthBook experiment tengu_amber_redwood silently reduces autocompact window to ~80k on Sonnet 200k (v2.1.92+ regression)
- [BUG] Switching from Sonnet back to Opus after "prompt too long" error leaves session in broken state
- [BUG] Claude Compaction Completely Broken for Opus
- Massive token burn: 80% → 11% in ~5 minutes on a single session
- Session startup context overhead consumes ~20% of context window before first user message
- [Bug] Post-/compact continue references wrong session context (cross-session memory bleed)
- Claude ignores memorized instructions and repeats corrected mistakes
- [Bug] Excessive token consumption from system prompt and slow response times in Claude Code
- [BUG] Scheduled Claude Routine stalls on deferred tool schema loading — requires manual input to continue
- [BUG] markdown links dont work in Claude code on Desktop
- [BUG] VM download aborts after
- Claude Code web sandbox proxy allowlist blocks user-owned custom domains
- [Bug] CLI only consuming extra-use tokens despite available credits
- [Bug] Anthropic API Error: Token Exhaustion and Inconsistent Model Behavior
- [Feature Request] Add repository categorization to organize sessions across multiple repos
- Real-time file tree & multi-file preview panel for non-developer workflows
- [Bug] Agent fails to discover project-local skills when resolving slash commands
- Permission allowlist changes don't take effect mid-session — breaks scheduled/autonomous tasks
- [BUG] Docker daemon not started on web environment
- Session failure post-mortem 2026-04-26
- [BUG] undefined is not an object (evaluating '$.input_tokens')
- [Bug] /autofix-pr command not functional despite proper GitHub app installation
- [MODEL] camelCase double-vowel sequences at word boundaries are systematically misread, causing wrong characters in generated code and scripts
- [BUG]
- [FEATURE] Desktop App: allow launching a session directly in a user-level agent (equivalent to CLI claude --agent <name
- [Feature Request] `--resume` flag should limit sessions to current directory
- claude-in-chrome MCP: 'Browser extension is not connected' despite --chrome flag and remote debugging enabled (macOS, Chrome 147)
- [Bug] File deletion confirmation still blocks after explicit approval
- [Bug] Excessive token consumption causing rate limiting on conversation resume
- [Bug] Installation state remains "installing" after completion
- [BUG] Constant API Error: Stream idle timeout - partial response received
- [stream-json] Emit Anthropic API errors as a distinct event type, not as assistant text blocks
- Cannot delete Routines from Claude desktop app sidebar on Windows
- [BUG] Hook stdin contains unescaped U+0000–U+001F control characters in JSON string fields, causing jq parse error and silent bypass of security hooks
- Feature Request: Native token compression for CLAUDE.md and instruction files
- Dashboard 'Favorite model' counts text mentions, not actual inference model
- [FEATURE] Claude Cowork custom connector options
- Sandbox cgroup silently disables transparent huge pages (THP)
- [BUG] preview_screenshot MCP tool always times out — preview pane visibilityState permanently "hidden" when chat pane is focused
- [BUG] "illegal hardware instruction" on Intel Core i7-3615QM @ 2.30GHz
- [FEATURE] Allow customizing auto-generated conversation tab titles (length / format)
- [Bug] Message editing submits previous version instead of edited content
- [BUG] Cowork Tab Not Appearing on Claude Desktop for Windows — yukonSilver Unsupported
- [Max20] 200k context window makes complex project workflows unusable — 1M context urgently needed for Max20 subscribers
- Desktop App v2.1.111: New conversation auto-loads last project folder, no folder picker
- 在那裡說明：「上傳圖片時若文字欄為空，會觸發 API 400 錯誤 ‘messages: text content blocks must be non-empty’」
- [BUG] CCR remote agent commits to local branch without pushing — work lost when sandbox terminates
- [BUG] Organization plugins not available in Desktop Code tab
- Cloud-spawned tasks have no discoverable handle — invisible in desktop UI, unstoppable from agent
- [BUG] Durable cron tasks written to ephemeral git worktree instead of main repo
- [FEATURE] Enterprise usage bar support
- Hooks can't gate assistant chat text — only tool-call and user-prompt boundaries
- [BUG] Desktop app does not save OpenRouter API key
- Plan mode indicator shows on, but Edit/Write still execute — no bypass involved, no defaultMode set
- [Bug] Security: Bash commands execute without user approval
- ultraplan / ultrareview commands not visible despite eligibility
- [FEATURE] Headless claude status --json for current session state
- Feature request: Expose `resume: session_id` on the Claude Code `Agent` tool so rate-limited sub-agents can be continued after limit reset
- [Bug] Model not completing all requested tasks in single response
- WSL: 'Exec format error' for versions > 2.1.62, binary not executable
- [FEATURE] Option to COMPLETELY disable "accept edits" on
- [Bug] Auto-mode classifier incorrectly blocks parallel tool execution after allowing similar tool calls
- [BUG] Cowork scheduled task reads stale filesystem snapshot between invocations (Windows)
- [BUG] CJK IME multi-character commit causes text flickering/jumping in Alacritty (bracketed paste mode)
- [BUG] Large paste is silently trimmed — only partial content reaches the input
- Skills edited via Customize UI not reflected in Code tab (storage path mismatch)
- Permission prompts on heredoc body lines corrupt settings.local.json with garbage allow entries
- Inconsistent temporal context: models often operate without knowing the current date
- Feature request: add a true no-prompt dangerous permission level to the permission UI
- [BUG] AWS Cloudwatch Output token count is too high for Claude 4.5 Sonnet
- [Bug] Anthropic API Error: Incorrect Usage Policy rejection for benign PDF debugging request
- ultrareview crashed but consumed free credit (no findings produced)
- Expose message timestamps to the model in Claude Code sessions
- [Bug] Socket Connection Closed Unexpectedly During API Request
- [BUG] bug in Claude Code's PDF parser
- [BUG] Fullscreen mouse movement generates garbage characters in input command line
- Feature Request: Support LC_TIME locale for /usage command time formatting
- Leading whitespace before slash commands is no longer stripped — command text sent as message instead
- claude -p exits before Task(run_in_background=true) subagents complete, truncating output
- Ultrareview: add structured report view & download on web session page
- [BUG] [BUG] Create PR notification above query line in Claude.app Code window shows changes to untracked files in counts.
- [Feature Request] UltraReview: Add markdown export and cross-session/cross-device findings retrieval
- [Bug] Process exits with code 1 without error message
- [BUG] Claude Code process exited with code 1
- [BUG] .rgignore patterns block REPL startup for ~7s on large repos (synchronous glob matching)
- Claude Dispatch: no way to delete conversations
- [Bug] 1M context window not auto-applied on Max plan in Claude Code CLI v2.1.114 (plus: `--model` flag silently drops `[1m]` suffix)
- [BUG] [Cowork] macOS Seatbelt blocks write to /sessions/.../tmp — Bash completely broken
- [Bug] Claude infers Slack workspace membership from URL subdomain without verification, presents unverifiable assumptions as facts
- PermissionRequest hook does not override .claude/** sensitive-path prompt on macOS 2.1.119
- Feature request: Manage Claude Code / CLI conversation history from claude.ai browser interface
- WebFetch can fabricate confident factual claims by misreading page boilerplate as live status
- Sandbox blocks Unix domain socket IPC with credential managers (rbw/Bitwarden CLI)
- Text cursor (blinking caret) invisible in light mode — can't see typing position
- Follow-up to prior feedback — reasoning failure persists at xhigh after explicit naming
- False-positive Usage Policy refusal on Russian text typed in QWERTY layout
- [FEATURE] iOS Remote Control: show "connected" confirmation after handshake
- [BUG] Claude in Chrome (Beta) feature blocked on Windows native — flag exists but slash command unavailable
- [BUG] Cowork shows "Virtualization is not enabled" on Windows 10 Home despite all virtualization features enabled
- [Bug] Claude Code ultraplan fails to initialize on Windows 11 on both wsl2 and cmd
- [Bug] /remote command not implemented or account not enabled (Max20)
- [BUG] False-positive AUP refusal during legitimate AI chatbot plugin development
- This appears to be user feedback/commentary rather than a bug report that needs a GitHub issue title. However, if you need me to extract actionable issues from this feedback, here are the potential GitHub issues: **[Feature Request] Add source attribution
- [BUG] : Claude Code usage increasing without active prompts
- Skill tool's forked execution describes plugin skills instead of running them; ignores `context` frontmatter field
- [BUG]
- [BUG] Stop begging me to upgrade every 5 seconds it's HIGHLY ANNOYING
- [BUG] Claude Code Desktop GUI - exit code 1 on every session (Pro plan, Mac)
- Cowork VM SDK install on Windows ARM64: hundreds of Operation not permitted errors during node_modules cleanup
- Security alert — malware repo typosquatting Claude Code: github.com/Enxs969/skiller hosting Trojan:Win64/Lazy.PGPK!MTB
- Feature: Support .claude/rules.local/ for project-specific gitignored rules
- [DOCS] Advisor tool: clarify mid-session course-correction value, not just upfront planning
- [MODEL] Keeps hanging and not executing code edits/writes and API Error (stream idle timeout - partial response recieved)
- Claude Code repeatedly ignores agreed approaches, hacks instead of reading code, and wastes user time on trivial problems
- You've hit your limit · resets 11:20pm (Africa/Cairo)
- [Bug] Claude fails to analyze codebase before applying fixes, ignores instructions
- [Bug] Claude 4.7 ignores established project constraints and over-generates code/docs
- Usage indicator randomly flickers between different percentages
- [BUG] Max 20x gift subscription auto-canceled after 3 days — account reverted to Free plan
- [FEATURE] Render images returned from MCP tool calls inline in chat UI
- [Bug] Claude command not found despite PATH configuration
- [Bug] Duplicate output displayed three times for single command execution
- [BUG] `sandbox.filesystem.denyRead` silently bypassed when target is inside a `denyWrite` directory
- "Claude Code process exited with code 1" on macOS — real cause is folder privacy permission, no helpful error shown
- Claude applies partial fixes: changes the reported line but does not trace or verify the full data path
- Claude hallucinated identifiers: writes column/field/config names from memory instead of reading source files
- Desktop app: partial text selection copies entire message instead of highlighted portion
- Sub-agent audits produce false reports due to silent file truncation and missing grep verification
- Claude hallucinates framework APIs: writes method calls that do not exist in the framework version used
- [BUG] autofix-bot may or may not still exist, but certainly doesn't work
- [BUG] Tool calls (Bash mkdir and Write) executed without prompting for approval despite the user selecting manual approval mode for the session.
- [BUG] Aws Auth Refresh broken on versions >0.92
- Extra blank lines rendered around SessionStart hook output and first response
- [BUG] Status bar shows "plan mode on" after plan mode has exited
- [BUG] Windows MSIX: node-pty fails to spawn shell — 'Error: File not found:' in windowsPtyAgent.js (terminal panel broken)
- [BUG] Skills installed via npx skills add (~/.agents/skills/) not recognized by Skill tool — 'Unknown skill' error
- Queued-message ordering is undocumented and appears LIFO in practice
- [FEATURE] Cmd/Ctrl+click on file paths should open in default external editor
- [BUG] Cant change Claude Desktop App back to 300px width on Mac
- [FEATURE] Using `CLAUDE_CODE_OAUTH_TOKEN` to start a remote session
- Claude Code Access Disabled After Payment - Account: seodepartment1@gmail.com
- resolved
- Versioned install path causes TCC re-prompts on every Claude Code update (breaks MDM PPPC; CoWork scheduled actions may be affected)
- [MODEL] Ignoring .md, my instructions
- [BUG] Regression in 2.1.153: claude-in-chrome browser tools no longer load in VS Code extension (worked in 2.1.152)
- [BUG] Opus 4.7 1M context missing in the /model selector
- [BUG] Compatibility issue with DeepSeek API: "unknown variant 'system'" error in messages
- Web hand-off session cannot commit or push — work stranded in ephemeral container (commit-signing 400 'missing source' + no remote/gh)
- [BUG] Windows: host CPU pressure starves Claude Code input loop — `claude agents` arrow keys/typing dead until load drops
- [Bug] Session context lost after overnight inactivity or resumption
- [SECURITY] PreToolUse hooks not enforced on subagent tool calls, allowing security bypass
- Voice input stopped transcribing after updating to 2.1.156 in VS Code on Windows 11
- security_reminder_hook.py uses Unix-only /tmp/ log path (breaks on Windows)
- [BUG] Opus 4.8: Repeated/duplicated output blocks in Claude Code CLI, worsening with session length
- [DOCS]
- [Bug] Statusline effort.level reports "xhigh" instead of "ultra" for ultracode mode
- Developer → Open Hardware Buddy… menu missing after enabling Developer Mode
- Monitoring cycle 18:03 UTC — 20k pending orders accumulation in ft_orders
- Directory connector visible to one Owner but invisible to another Owner in same Team organization
- [BUG] Windows 11 Claude Desktop App at launch asks for Security code from Email to Log in: Email has a log in button that launches Edge and logs me in.
- [Bug] pdf-viewer MCP built-in server - UtilityProcess spawn timeout (Claude Desktop, Windows 10)
- [Bug] Claude Code unable to create accounts on localhost platforms
- [Bug] Claude Code corrupts Git repository state
- [Bug] Differential refusal behavior between subprocess and main turn in same session
- Model re-attempts denied tool call after user explicitly says 'no'
- Auto-compact triggers at ~8% context usage (84k/1M), 882k tokens free
- [BUG] MSIX package path resolution failure - "Open in Word" and "Show in Folder" completely broken in Cowork
- Auto-compaction blocks session for several minutes mid-task
- [Feature Request] CLAUDE.md re-injection deduplication for nested contexts
- Repeated context injection via JSONL attachments causes linear token bloat across long sessions
- Memory/active plan system is write-only in practice.
- Claude forgets its own file operations within the same 1M context window
- [Bug] WhatsApp bridge context accumulation causes exponential token consumption without warning, Clauxe Max Usage metrics are hard to retrieve
- [Bug] Claude Code 4.7 High Token Usage and Context Limitations
- [Bug] Unexpected rapid token consumption with 5x plan
- Plugin marketplace fails with EFAULT on macOS 26 Tahoe
- token 用量暴涨10倍，模型质量下降，代码改了一遍又一遍改不对
- [Feature Request] Improve remote control performance for GPU-intensive tasks
- [BUG] Dispatch non-functional since 29 March incident — messages marked "Read" but no response delivered, reproducible on desktop directly (Max plan)
- Context compaction too aggressive on 1M context window (Opus 4.6)
- Vi mode: add 's' (substitute) command
- [BUG] Bun has crashed. Segmentation fault at address 0x35C23800000
- [BUG] Weekly usage not reset during Feb 27 global reset + cycle date shifted (Max 5× subscriber)
- [BUG] User-input markdown autolinker treats `.md`/`.rs` filename extensions as TLDs, mangles `<word>_<filename>.md` references in chat
- [BUG] MCP edit_file $-stripping regression after latest Claude Desktop update
- [FEATURE]Don't trigger system notifications when the Claude Code session is already visible on screen
- [FEATURE] Ability to convert a Cowork task into a Cowork project
- [FEATURE] Update context when moving Cowork task into project
- [BUG] CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 disables TUI scrollback in v2.1.119 (terminal can't scroll up to view earlier turns)
- [Bug] Hourly rate limit consumed without user request or API call
- Claude Desktop: orphaned MCP subprocesses accumulate after 4-min tool-call timeout
- [DOCS] Skill - /Schedule
- [BUG] Mac app: orphaned claude-code CLI from ccd_session chaptered sub-task spins at 100% CPU indefinitely after parent exits non-gracefully
- [BUG] Claude in Chrome: programmatic scrollTop on virtualized third-party lists triggers runaway auto-prefetch loop that hijacks the page
- VSCode extension 2.1.117: @ file picker returns 'No files found' when system rg is not installed (bundled rg not used, findFiles fallback also fails)
- Plugin marketplace serves oldest version instead of latest when multiple submissions exist
- [Bug] /status reports costs for subscription plan when no billable usage occurred
- Plugin 'ido4shape' shows Published but not in community marketplace (last sync Mar 31)
- [FEATURE] TodoWrite items in the right-hand Tasks pane, not just inline in the conversation.
- [FEATURE] VS Code Extension should warn when reading/writing unsaved files or use unsaved version
- [BUG] /insights report: bar-chart labels truncated (width: 100px too narrow)
- Agent killed unidentified processes after being told to only kill unused dev servers
- [Feature Request] Recall & edit previous prompt in Claude Code desktop app (Mac)
- Sonet 4.6: 400 status with "Output blocked by content filtering policy"
- [VSCode Extension] Add stall detection and auto-retry settings for hung AI responses
- [Feature Request] Expand AskUserQuestion "Other" input field to support multiline responses
- [Claude Code CLI] (1) Voice-toggle strips typed slash commands; (2) Right-click "Attach message as context" double-pastes body (Windows 11)
- [BUG] Windows CLI: subscriptionType: max but interactive session fails with "Your organization does not have access to Claude"
- [BUG] Claude Code on macOS does not recognize active iOS Apple-billed Max subscription, login succeeds but CLI says account has no access
- Claude Desktop cowork VM startup permanently breaks WSL2 interop (vsock accept4 failed 110)
- [BUG] [BUG] Switching from Opus 4.7 1M to Sonnet mid-session (~700k tokens) breaks session completely — no recovery possible
- [BUG] AskUserQuestion: partial submit discards all already-provided answers
- [Bug] Claude model refuses to continue after single communication failure
- [BUG] VS Code Connexion compte Google
- [BUG] claude doctor to work with nvm
- [BUG] Dispatch mobile shows "Can't reach desktop" despite desktop app running and Cowork working normally
- Plan mode silently exits / auto mode overrides "plan carefully" requests on follow-up work
- [Feature Request] Improve model response quality and guideline adherence
- [BUG] /ultrareview Find step fails with empty [] for local repos — 3 attempts, all failed, and consumed all 3 free credits
- Claude Code 2.1.118+ hangs on macOS Sequoia 15.7.2 with OpenCore Legacy Patcher (Non-Metal GPU)
- [DOCS] skills priorities
- Feedback: Claude can iterate 8+ times on same layer without considering 'wrong layer' as hypothesis
- Feedback: Claude can iterate 8+ times on same layer without considering 'wrong layer' as hypothesis
- Feedback: Claude can iterate 8+ times on same layer without considering 'wrong layer' as hypothesis
- [Bug] Sandbox write-deny on .claude/commands blocks git worktree/checkout for repos with tracked slash commands
- [FEATURE] DISABLE_NONESSENTIAL_TRAFFIC silently disables security updates — unbundle DISABLE_AUTOUPDATER
- [BUG] SSHFS/network drive mounts fail directory validation in v2.1.82+ (regression from v2.1.81)
- [BUG] Bash tool hangs indefinitely on 'git remote show origin' — blocks session and remote-control clients
- Simple ops tasks dragging for days — orchestrator + agents ignore memory/CLAUDE.md constraints
- [Bug] Session/weekly token limits deplete disproportionately fast with Opus orchestrator + Haiku subagents
- [BUG]API Error: 500 {"error":{"code":"","message":"rate_limit_check_failed (request id: 202604260831354254128128268d9d641UJyrko)","type":"new_api_error"}}
- [FEATURE] Pycharm Claude terminal app: The resume picker should display key hints (↑/↓ to navigate, Enter to select)
- Sandbox mode leaves read-only bind mounts that persist after disable
- [BUG] Extra usage isn't working correctly - it's using up tokens for the next 5-hour session
- [BUG] otelHeadersHelper missing in Desktop App
- [BUG] Error running ultrareview: session URL (session_01GpHVnyZfLi9F1auBhLhyZs) and task ID (rin5ozzna)
- [BUG] Resources of type 'application/pdf' are not currently supported.
- Project not appearing on claude.ai/code despite remoteControlAtStartup: true and autoUploadSessions: true
- Skills: no escape syntax for literal `$<digit>` in command bodies
- [BUG] Claude CLI v2.1.119 crashes with SIGTERM (exit -15) after ~10.5h background running with Telegram plugin on macOS
- Plugin skill discovery: multi-scope installs silently load intersection, project-scope leaks across unrelated cwds
- [BUG] Claude Code Web version Failed to load scheduled tasks
- Edit tool's "must Read first" rule conflicts with successful-edit hint, causing repeated stalls
- [BUG] CLI / Claude macOS app Issue | Cannot execute Claude Code prompts.
- [FEATURE] Auto Debug Breakpoints
- [BUG]
- [BUG] CCR scheduled routines stuck at "Setting up a cloud container" since 2026-04-24
- [FEATURE] /resume should support filtering to named sessions only
- [BUG] Plugin install no longer copies symlinked skills to cache (regression in v2.1.117)
- The thinking words are gone
- [BUG] 🚨 Excessive Token Usage in Claude, Small Prompts Consuming more than 80% Quota, Think before claude subscription
- [BUG] iphone notifications not coming when task via remote control is done
- [FEATURE] find string (like cmd+f on mac) on the iphone claude code app
- OTel enhanced telemetry beta: streaming mode (Agent SDK / ACP) only emits claude_code.llm_request — interaction/tool/tool.execution spans are missing
- Opus 4.7 long-session pattern: using external review as iteration loop instead of design exhaustion
- 5-hour session reset time shifts forward by ~4 hours when new sessions open near window boundary
- Plugin loader silently breaks when installed_plugins.json and known_marketplaces.json drift after marketplace refresh
- [Bug] Claude Code references non-existent built-in CLAUDE.md policy causing false attribution
- [Bug] Error: Contains backslash-escaped whitespace validation too strict
- Hindi/Devanagari text renders broken in terminal TUI — characters scattered across columns
- Cross-domain work penalized by token consumption — users forced to throttle model capacity to survive session limits
- [BUG]
- Claude acts on tasks without user approval — implements instead of listing when asked to list
- Claude argues against non-traditional designs without testing — repeated pattern
- [Bug] Agents and sub-agents ignore dontAsk permission model
- [BUG] Claude.app (macOS): ShipIt falsely reports successful update but bundle swap never completes
- [FEATURE] Claude Code - Session history persistence between sessions
- Number keys collide between in-message numbered options and feedback shortcuts
- [BUG] PRO plan 5-hour session consumed after a few md file edits
- Bring back token-by-token streaming (line-by-line streaming since v2.1.78 feels worse)
- Bedrock makes 3x more tool calls and API round-trips than direct API for identical tasks
- [BUG] Claude code not expand inside tmux session
- [Bug] Model hangs on simple tasks with excessive token consumption
- You've hit your org's monthly usage limit
- [BUG] Allow Ctrl+B passthrough for readline cursor navigation
- Add conversation naming, numbering, and summaries for easier reference
- [BUG] Project memory silently drops when session cwd differs from user's main project directory
- [BUG] Unable to resize Claude Desktop App on Mac
- [Bug] Opus 4.7 optimizes for narrative closure over verifiable outcomes in multi-session workflows
- [Bug] Claude 3.5 Opus 200K context window incorrectly charged as extra usage on Max plan
- [FEATURE] Configurable path for all Claude files that are written in my home dir
- Session auto-titler reverts manual rename in Cursor extension
- [Bug] Claude overreaches and behaves inconsistently across sessions despite explicit instructions
- [BUG] [Feedback] Persistent drift, optimistic completion bias, and unverified-claim cascading in extended technical debugging sessions
- [BUG] UNC paths in working directories cause settings.json permissions to be ignored
- Proposal: On-device AI redaction primitive for Claude Code memory
- Same assistant response rendered 16 times in terminal for a single user message
- [BUG] ccd-cli fails on multi-user Linux: /tmp/claude-settings-{hash}.json collides between users
- [MODEL] Sonet and Opus 4.7 behaves like dumb
- Add browser rendering / DOM inspection capability to Claude Code
- Claude suggests global `docker network prune -f` as "safe" — wipes networks across unrelated projects
- [FEATURE] Support URL imports in CLAUDE.md (`@https://...`)
- [BUG]Claude Code on Windows should look for bash at C:\Program Files\Git\bin\bash.exe (or check both paths).
- /ultrareview crashes silently — opaque "Review crashed" error with no root cause or mitigation
- [BUG] Planviewer interrupts while I am writing my comments and ALL comments are NOT saved!
- I don't have access to the content of issue SP-702. Please provide the bug report details, error messages, or description of the problem you'd like me to create a GitHub issue title for.
- 总是返回错误400
- I don't have access to issue SP-702 or any external reference system. Please provide the actual bug report details so I can generate an appropriate GitHub issue title for you.
- [BUG] Code mode broken after v2.1.78 update — .app bundle binary not spawnable by desktop SDK
- [BUG]
- [BUG] Web client: resuming after stream idle timeout sends empty text block with cache_control, API returns 400
- [BUG] app bounces then opens but doesn't render - Claude Desktop blank screen on M4 Pro MacBook (macOS 15.7.4)
- Terminal freezes on /remote-control opt out in linux
- [BUG] Claude conflates planning permission with execution permission
- [Bug] Unauthorized Git repository modifications by Claude agent
- [FEATURE] Clade Design: prevent left scroll to go back in chrome
- [BUG] [Claude Cowork] Managed config managedMcpServers and inferenceModels arrays silently dropped from .mobileconfig — writer↔reader type contract mismatch
- You're out of extra usage · resets 3:50pm (Asia/Calcutta)
- Claude Code Rendering Issue in Terminal
- claude code folder access problem over Claude Desktop
- [BUG] degraded thinking performance since Dec/Feb still persisting (stop ignoring your paying customers pls)
- [Feature Request] Support keybindings for user-defined slash commands
- Notion MCP: Claude consistently misuses page icon — pastes emoji into title text instead of (or in addition to) the icon slot
- [BUG] Claude Desktop causes repeated WindowServer crashes on macOS 15.7.3 (Sequoia) — VirtualDisplayListener deadlock
- [BUG] Desktop App: Claude Code process exited with code 1 - skills plugin flag incompatibility
- Feature: Configurable status bar / footer with user-defined items
- Bash tool: env vars leak into spawned shells as literal string "undefined"
- [BUG] API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages: text content blocks must be non-empty"},"request_id":"req_011CaU2tSNQ2e97yzawT5KPs"}
- ENABLE_LSP_TOOL: LSP subprocess stdin is immediately EOF, server shuts down before receiving any LSP messages
- Can no longer tell Claude to change its working directory mid-session (regression)
- Free ChatGPT 5.5 outperformed paid Claude Opus 4.7 (Max plan, /effort max) on a basic multi-file reconciliation task
- Bash cwd no longer persists between calls (regression, even without CLAUDE_BASH_MAINTAIN_PROJECT_WORKING_DIR)
- [BUG] Safari mobile is terrible
- Slash-command/skill discovery walks node_modules, polluting context with hundreds of fake skills
- Plugin install path resolution: outdated project-scope entry shadows user-scope install (still reproducible on 2.1.121)
- Auto-updater overwrites FreeBSD ports installation, breaking claude
- [BUG] Remote Control: Desktop app drops messages during model switch lag
- Shell snapshots generate empty if-then-fi block
- Run Claude Code as a dedicated system user to enable OS-level file access enforcement
- [BUG] Desktop App — Sidebar overlaps chat content without left margin (macOS)
- VS Code extension: add a verbose toggle equivalent to CLI Ctrl+O
- [BUG] Claude Code Issue: Subagent Responses Not Returned to User
- Feature request: expose a `rename_session` tool to Claude in Code Cloud / Desktop
- [Bug] MCP resource_link silently ignored, PDF blobs not processed, and no docs on supported content types
- Plan mode approval UI missing when using --dangerously-load-development-channels flag
- [BUG] Claude Code crashing application
- [FEATURE] Add syntax highlighting to fenced code blocks in the VS Code extension chat panel
- [Bug] Auto-update failure notification displayed persistently
- [Bug] Chrome extension permission prompts appear unreliably and difficult to locate
- [Bug] Claude fails to complete final deployment steps for website publishing
- [BUG] Claude Code 2.1.121 Ink TUI incompatible with GNOME Terminal on Ubuntu 24.04