Digest / May 7, 2026
Claude Code Issues Digest — May 7, 2026
267 new issues filed ·
230 resolved in anthropics/claude-code (UTC day).
New issues (267)
- [BUG] --settings <json> writes to shared /tmp/claude-settings-<hash>.json, blocks all but first macOS user account
- [Bug] Claude presents token-presence checks as semantic verification, masking content loss or inversion
- [BUG] Claude Desktop 1.6259.1.0 — MCP servers show "running" in Settings but tools never reach servers from chat
- [BUG] /mcp displays list in a fixed scroll container, cannot view/take action on items outside of the default view
- Session hangs indefinitely with spinning thinking indicator — 3 confirmed variants (+ /btw MCP teardown)
- [BUG] Chat responses disappear/overlap with user commands
- [BUG] Cyrillic Homoglyph Characters Injected Into English Output During Long Sessions
- /doctor permanently reports `autofix-bot` not cached even when not installed
- [Bug] Session timeout occurring prematurely within minutes of activity
- [BUG] Claude Code on the Web auto-commits and pushes — task-agent system prompt hard-codes commit/push instructions, overriding CLAUDE.md
- [MODEL]
- System reminders inserted after every tool call disrupt long-running multi-step work
- [BUG] Claude Desktop - COWORK not present in app - Issue #25136
- Sub-agent returning "Tool result missing due to internal error" silently hangs parent — no error propagation
- 🚨 Claude Code agent repeatedly lies about task completion (declares OK without verification) — 12-day documented pattern with evidence
- [Bug] Permission pattern matching fails when `:*` suffix follows another `*` wildcard
- MCP server "ide" connection fails immediately on startup - IntelliJ IDEA plugin 0.1.14-beta
- [BUG] Claude Desktop 1.5354.0: stdio MCP servers double-spawn from independent CCD/LAM coordinator registries
- [BUG] MCP tool output not rendered in terminal when using native TextContent (fastmcp 3.x)
- [BUG] switch model doesn't work in Claude Code for VS Code version 2.1.132
- [BUG] Hook output and tool-result capture files in /tmp/claude-{uid}/<cwd>/tasks/ accumulate ~19MB null padding per file (missing ftruncate)
- [BUG] click registered in UI but empty payload reached the assistant
- claude-in-chrome: "Multiple extensions connected" error is unrecoverable without manual intervention
- [DOCS] [Environment variables] CLAUDE_CODE_SESSION_ID not documented in env-vars reference
- Add `OnContextThreshold` hook for programmatic mid-session handoff
- [DOCS] [Environment] Missing `CLAUDE_CODE_DISABLE_ALTERNATE_SCREEN` env var documentation for opting out of fullscreen rendering
- [DOCS] Image paste from clipboard lacks loading/feedback indicator
- [DOCS] [CLI] `--permission-mode plan` not restored on `--resume`/`--continue` after v2.1.132 (docs say it isn't, but it now is)
- Safety: Claude pushed to upstream remote instead of fork without warning
- [DOCS] Interactive mode docs don't clarify that pasted text starting with `/` is treated as literal input
- [DOCS] [MCP] Stdout parsing edge case not documented — servers emitting non-protocol data cause unbounded memory growth
- [Bug] MCP HTTP OAuth token exchange ignores authorization URL scopes, uses metadata instead
- [DOCS] [MCP] Document retry behavior and `/mcp` status for servers that connect but fail `tools/list`
- [DOCS] [MCP] Document "needs auth" vs "failed" status distinction and non-transient 4xx retry behavior in headless mode
- AskUserQuestion: Escape destroys typed answers (breaks vi/vim users)
- [BUG] Sidebar Recents should support cross-worktree aggregation for monorepo / multi-worktree projects
- [BUG] Wildly incorrect usage reporting in two different instances this week has severly impacted my workflow and am now uncertain what to trust.
- [BUG] claude hangs silently when ~/.claude/ is unwritable
- [BUG] Excel Agent repeatedly claims task completion without executing changes; fabricates verification details for unwritten cells
- [Billing Bug] Claude Max payment charged but account reverted to Free plan - invoice shows "Paid"
- [Bug] Agentic loop stuck requiring multiple retries to recover
- [BUG] Max Account downgraded from Max to Free Plan mid-subscription
- [Bug] Anthropic API Error: 400 Could Not Process Image
- [BUG] `/effort <level>` on first line of multiline message fails with misleading "Invalid argument" error
- [Bug] Model hallucination causing incorrect code generation and failed validation
- [BUG]VS Code extension: slash command triggers API Error: Content block not found
- [BUG] Desktop app: previously saved sessions not displayed in UI after reopening project (Windows v2.1.128)
- "Open Claude Code" handoff from claude.ai doesn't carry conversation context
- v2.1.128 update silently destroys conversation history for pre-JSONL sessions
- [BUG] While I was using Claude inside Cursor after 45 minutes, it got stuck and I have closed Cursor and reopened and it said, it didn't have the memory file to see what it was doing last. So whatever it worked for 45 minutes is gone?
- [BUG] Claude consistently makes unverified technical assertions that cascade into incorrect development decisions
- [FEATURE] Allow plan-preview UI for skill-generated plans/specs, not only ExitPlanMode
- [BUG] PowerShell tool returns exit 1 with no output for all commands (regardless of allowlist matching)
- [BUG] Enter key executes suggested user action when prompt out of view
- Expose /rename'd session name in model prompt context
- [BUG] /clear implies that I will get a refund on tokens
- [BUG] `Skill()` entries in skill frontmatter `allowed-tools` are ignored — invoking a nested skill prompts for approval despite the parent skill granting arbitrary script execution
- Make autonomous Claude Code actually viable: tiered Opus brains + Sonnet workers + persistent state
- [BUG] CLI process orphans at 100% CPU after window/terminal close (2.1.111, 2.1.121)
- title: Feature Request: Persistent Claude Identity / "My Claude" across conversations
- [FEATURE] Provide a way to select your data center
- [Bug] Claude Code generates mixed CJK characters in Korean text instead of pure Hangul
- Dynamic autosuggestions for slash command arguments
- [BUG] Usage limit reached prompting upgrade when only 11% usage
- [Feature Request] Built-in cloud sync for projects, sessions, settings, and conversation history
- Preview panel renders blank on open — fixes only after window minimize/restore
- Feature Request: Token Limit UX — Configurable Threshold Alerts, State Preservation Before Cutoff, and Grace Period Buffer
- [BUG] VS Code extension: Slash-command autocomplete traps history navigation
- [DOCS] 5 hour limit removed, but weekly limit remains? No more capacity just able to hit limit faster?
- Auto-set VSCode terminal icon and color when Claude Code is running
- Make `/fast` (Opus 4.6 fast mode) available to users without extra-usage enabled
- CLAUDE.md: @ file import fails silently for paths containing spaces (e.g. iCloud Drive on macOS)
- [BUG] /agents Library list does not auto-scroll viewport — selection past first page is invisible
- [FEATURE] Allow plugins to restrict install --scope via plugin.json
- [BUG] Agent Teams: background sub-agents idle for 30+ minutes after work completion; SendMessage ping required to wake them and trigger final report
- I'm ready to generate GitHub issue titles for Claude Code bug reports. Please provide the bug report content and I'll create a concise, technical issue title following the specified format.
- [BUG] VS Code Claude Code extension fails to respond to a prompt, with no response to attempts to cancel
- Allow disabling or customizing terminal tab title (currently locked to CC version)
- [BUG] Session crashes on accumulated screenshots — no recovery, no transcript export (regression of #9170 / #42558 / #51042)
- [BUG] forceLoginOrgUUID is incompatible with setup-token OAuth tokens — breaks all unattended/CI usage
- Notification hook does not fire on permission prompts (Windows 11)
- [Feature] /mcp reconnect-all: batch restart all MCP servers mid-session
- Silent crash on Windows when plugin autoupdate hits an orphaned installed_plugins.json entry
- Feature: PreToolUse block hooks should support custom tool_result (redirect without error framing)
- [Bug] MSIX Installation Timeout: AddPackage failed after 5 minutes
- autotrading bot
- [Bug] Whitespace prompt fires for allowed read-only commands in iCloud Drive projects
- [UI] MCP marketplace shows 'Install' button for already-installed servers
- [FEATURE] Surface pending-approval state in session JSONL (or via API) so external tools can detect blocked sessions
- [FEATURE] Claude Desktop user preferences don't apply to Claude Code
- [BUG] Edit tool corrupts GBK-encoded Chinese comments in source files
- [BUG] claude.ai/Design - no access on Max subscription, UK account.
- [BUG] IntelliJ plugin: ACP agent and `claude` child survive `/exit` and tool window close (orphaned processes)
- [BUG] [BUG] Claude Desktop installer fails with HRESULT 0x80073CF6 - MSIX package stuck in wedged state
- Voice mode: add multi-language support (currently English-only)
- [BUG] Claude in Chrome extension unusable for Apple Hide My Email users — magic link hijacked by Chrome Google session
- [BUG] Claude Code Desktop SSH to Hostinger KVM4 VPS - process.spawn exits with code 1 on every message
- [BUG] Directory Mismatch for Directory
- [BUG] Connector tool permissions (Always Allow) reset to default after every Claude Desktop upgrade
- [BUG]
- Add setting to hide/merge the 'auto mode on' status bar indicator
- Model cannot distinguish its own prior output from user messages, enabling self-directed action loops
- [Bug] Incorrect quote character conversion: smart quotes replacing straight quotes
- Claude-3p + Bedrock: managed-settings allowedDomains not honored, sandbox pinned to 4 hosts
- [BUG] Severe memory leak — claude/node process reaches 108GB on M5 Max macOS
- [FEATURE] Persist argument-hint outside the input box while typing arguments
- [Behavioral] Plan-Mode reminder persists after /plan-complete + /session-end
- Sub-agent attempts `exit_plan_mode` as Bash command, fails with Exit code 127
- [Feature Request] Auto-return to plan mode after plan execution
- Claude in Chrome v1.0.70: `permission_required` regression continues from #53630 — approval popup never renders, all MCP navigations blocked
- [Bug] Skill listing truncated due to insufficient context budget allocation
- Bug: /plugin marketplace add writes trailing comma in known_marketplaces.json, breaking subsequent runs
- C:/Program Files/Git/resume collapses final text response in focus mode
- [BUG] : The missing session/week usage in /usage
- [BUG] CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1 is being ignored since v2.1.129
- [VSCode Extension] File:line references in chat panel are not clickable/navigable
- Auto-memory system not used proactively during multi-session troubleshooting; user forced to re-explain 30+ hour saga
- Claude defaults to outdated training over current internet search, then cascades down wrong paths instead of grounding fixes in present-day reality
- [Bug] Claude ignoring instructions from CLAUDE.md configuration file
- HTTP MCP client doesn't follow OAuth authorization redirects, kills session with uncaught Unauthorized
- [Sonnet 4.6] Increased sycophancy after recent updates
- [FEATURE] please reuse the same environment ID when claude remote-control restarts QwQ (dead clones keep piling up forever)
- Graceful handling when hitting usage limits mid-session
- [MODEL] Search should be a default verification step on time-sensitive queries — cheap check, expensive mistake
- /status "Total cost" is confusing for Max-plan users — looks like a bill
- [Bug] Context limit reached prematurely despite low token usage percentage
- Add 'Yes, but...' option to tool approval prompt
- [MODEL] Opus 4.7 - "prioritize the users comfort over his safety"
- [BUG] CLAUDE_CODE_EXTRA_BODY thinking config breaks WebSearch and WebFetch on Opus 4.7 [1m]
- v2.1.132 (Cursor): parallel agent panels share one shell, sub-agent calls reset cwd mid-run
- [BUG] /ultrareview crashes in Dedupe stage after Verify completes — findings produced but never delivered to CLI
- Settings: Authorization tokens page missing bulk revoke and full pagination
- Mobile remote control fails with 'GitHub repository access check failed' for private repositories
- Agent with multi-destination addressing loses tool-use patterns after auto-compaction
- [BUG] Desktop 1.6259.1 adds model name validation that rejects non-Anthropic model names — breaks third-party gateways
- You've hit your limit · resets 1:40pm (America/Sao_Paulo)
- [FEATURE]
- [Bug] Platform Migration Regression: File Upload and Transcription Failures on Ubuntu 24.04 with GPU
- Feature request: option to hide PR number from input prompt
- User messages sent during agent execution can be silently lost or queued without UI indicator, causing work to proceed past intended halt points
- Resume picker shows no sessions when cwd is inside a git submodule
- [BUG] .claudeignore is silently ignored
- [Bug] Dark mode sources URL font color unreadable on dark backgrounds
- Drive sharing blocked while broader-risk actions are allowed
- [BUG] My Claude Code conversations just vanished
- [BUG] /plugin lists user-scope plugins under "Project" tab, blocks disable without --scope user
- [BUG] Cowork ECONNRESET on Windows 10 Pro - Chat works fine
- [BUG] Co-work single window not working - Crash
- [BUG] imessage: reply loop on self-chats keyed on receive-only aliases
- [FEATURE] Comment on the response option when the `AskUserQuestion` tool is called.
- [Android] Back navigation broken on foldables after switching conversations on large screen
- [Feature Request] Add session reset/clean state option to reduce context confusion
- [BUG] Telegram permission reply after local fallback timeout is silently dropped — allowlist not updated
- [BUG] Claude Code web session loses GitHub push access mid-project
- Model stuck in extended thinking loops — consumed 38% weekly token budget with no usable output
- [Bug] CLI session token expires prematurely, causing frequent logouts
- [BUG] /ultrareview crashed while charding the acocunt >$19 and producing no results.
- Windows: PowerShell allow rule auto-written on 'approve & don't ask again' fails to match identical subsequent invocations
- I'd be happy to help generate a GitHub issue title, but I don't see the bug report details in your message. Could you please provide the bug report information? I need details such as: - What the issue is - What error message (if any) was encountered - Wh
- [FEATURE] Pass relative paths to files under cwd when invoking tools
- `/release-notes` should indicate which version is current stable (per `/doctor`)
- [Bug] Claude GitHub App authentication fails on specific personal repositories despite valid installation
- Bridge session registration (bridgeSessionId) broken in v2.1.132 — regression from v2.1.131
- Feature Request: Show TodoWrite task list in Claude.app Tasks panel
- macOS Claude app — Plan sidebar shows stale cached plan after /clear; "Open plan" button non-functional
- [BUG] /Advisor not available in Claude Code CLI 2.1.132
- Plugins in claude-plugins-official marketplace lack `version` field in plugin.json, displayed as "unknown"
- [BUG] Windows: pwsh exited with code 1: The command line is too long — regression starting with Opus 4.7
- macOS Tahoe: Claude Code 2.1.132+ writes block all non-Anthropic apps from reading files in ~/Documents (regression vs 2.1.128)
- SONNET4.6
- [BUG] Windows Desktop app + SSO (Team plan): oauthAccount not hydrated to ~/.claude.json, OTel metrics emitted without user identity
- Claude desktop app shows blank main content area on macOS 16 (Darwin 25.2.0) after update to v1.6259.1
- Add configurable color for user input in transcript
- [BUG] MCP salesforce not connecting set up as per document
- CLAUDE.md approval-with-modifications instruction not respected — conditional approval treated as green light
- [Feature Request] Add configurable autoCompactThreshold parameter for conversation auto-compaction
- [Feature Request] Toggle Claude Code session between terminal and native UI modes
- ToolSearch does not index tools from claude.ai-hosted MCP servers (visible in /mcp, invisible to deferred-tool search)
- Support VS Code's browser-sharing API so Claude can validate web UI changes
- [BUG] CoworkVMService DACL omits BA/SY permissions, blocking installer upgrades and recovery (root cause for #25136, #48367, #49655)
- [Bug] Custom themes diff overrides not applying
- [BUG] Subagent permission cascade-failure when multiple Agent tool calls in one message (2.1.126)
- Subagent fabricates tool outputs (Read, Bash) when defined with sonnet model — wc -l reports wrong line count
- I don't have any bug report content to work with. Please provide the bug report details so I can generate an appropriate GitHub issue title.
- [Bug] LLM hallucinating incorrect tool outputs or responsesHW
- [Bug] Severe performance degradation on high effort setting
- [Bug] Claude generates hallucinated content without grounding in provided context
- [BUG] Character under cursor is illegible when terminal has CustomCursorTextColor set (Konsole)
- [FEATURE] Custom colors for agents
- [Bug] Ctrl+C not interrupting frozen agent, 'x to stop' input not functional
- [Bug] Chat width reduces when scrolling up in terminal output
- Mac app: 'Background full checkout failed: other' when starting worktree session (empty-index regression)
- [BUG] Edit tool progressively fails on WSL2 9p mounts (statx ENOENT on existing files)
- [BUG] Sonnet-only usage consuming both "Sonnet only" and "All models" limits simultaneously
- Chrome MCP
- Plan mode shows stale plan from previous conversation after resume
- [BUG] Cowork crashes with "Claude Code process exited with code 3" — Bun segfault inside VM on Windows 11 Enterprise
- [Bug] Agent suggests destructive commands without safety warnings
- Cowork: artifact deletion from UI leaves orphaned folder on disk
- Compaction destroys persistent identity/relationship context despite memory system
- [FEATURE] Counterpart to /chat from kiro-cli in claude code
- Extension Signature Verification Failed: Claude Code for VS Code
- [BUG] permissions strings matching is broken claude prompts for permission repeatedly
- Tool output is fully expanded by default and the collapse toggle (▸) is non-functional
- Feature request: side-by-side diff option for Edit/Write/MultiEdit confirmation prompts
- [Bug] Claude echoes user input with "Human:" prefix and misattributes it as user input
- [FEATURE] Push changelog to GitHub Releases
- Add Hungarian language support for voice dictation Claude
- [BUG] /context command shows /200k instead of /1M on Windows CLI when using Opus 4.6/4.7
- [FEATURE] /cd slash command for switching project directories within a session
- [BUG] Title: Subagents don't inherit .env deny rules from settings.json
- [BUG] @-includes in CLAUDE.md fail to resolve when project root is inside a hidden (dot-prefixed) directory
- [Bug] Explore subagent rejects with "prompt too long" error
- [BUG] All claude.ai built-in MCP servers disappear on every restart — require /login each session
- [FEATURE] session_info: allow read_transcript to access the current active session
- [Bug] Rate limits reached prematurely with 5x max plan tier
- [FEATURE] Lazy-load skill catalog and deferred tool names to reduce per-turn context bloat
- Markdown tables in CLI output render with unaligned column pipes
- Copying text from CLI in iTerm2 inserts hard line breaks at terminal wrap boundaries
- [BUG] Input suggestions appear as ghost text with no way to dismiss them — this interferes with typing and the ability to concentrate.
- TeamDelete returns success but teammate tmux pane survives
- Surface live stdout/stderr of background Bash tasks in the Tasks panel
- explanatory-output-style plugin SessionStart hook hangs all input in 2.1.132
- /commit skill performs unsolicited code edits when project instructions mention linting
- [FEATURE] Treat user typing during agent turn as a soft-pause signal
- [FEATURE] looking up to repositories in self hosted git instaces such as forgejo, gitea etc.
- Plugin updater leaves stale version directories in cache; loader globs all of them
- File links to absolute paths outside project root are not clickable
- Setting to hide GitHub PR/branch widgets in Desktop session view
- Models tab: per-model token counts do not update when switching All / 30d / 7d
- Background agent notifications not delivering reliably (macOS)
- [FEATURE] auto /clear
- Feature request: expose Claude.ai Max plan usage counters in model context
- UI label 'Shell' does not match backend tool namespace 'Task', causing LLM lexical confusion
- TeamCreate: stuck agents cannot be terminated from lead session — user must close and restart Claude to recover
- [BUG] MCP server tools unavailable after /clear in VS Code extension
- [Bug] Claude ignores user directives to wait for approval before making production changes
- [BUG] Claude Code on windows desktop fails to recover from API Error 400
- [BUG] claude -p rejected with "out of extra usage" while Max plan 5h limit shows 35% remaining (not the HERMES.md trigger)
- [Feature Request] Add toggle keybinding to show running background agents with live output
- Alt-V No image found in clipboard on Windows 11 despite image being present
- Notion MCP: notion-create-pages validation fails with misleading error when title contains special chars + body has substantial content
- [BUG] Clickable file links in chat no longer open files (regression in VS Code extension)
- [BUG] Bash tool rejected as "user denied" with no permission prompt — regression of #9383 on v2.1.128 / Windows ARM64 / Cowork
- [BUG] Stale .git/index.lock left behind in worktrees during normal CLI operation (macOS)
- Claude Code freezes for extended periods (often hours) during normal sessions at low context usage
- Visual agent dashboard
- yukonSilver detection false-negative on AMI BIOS whitebox PCs causes infinite VM bundle deletion loop
- [BUG] Bot lied and admitted it to me
- Hookify treats enabled: no as enabled
- [BUG] Cowork: "No conversation found with session ID" on every new task (Windows 11)
- /doctor TUI dismisses immediately on Enter or 'f' keypress
- Built-in slash commands (e.g. /ultraplan) not enumerated in per-session system prompt — AI flags legitimate harness UI as prompt injection
- Claude Code feature request: skip paranoid auto-spawn when cwd is already inside a worktree
- [BUG] Ghost text appears in input field unprompted (Windows PowerShell + Kali Linux)
- Meta Ads MCP OAuth fails on Claude Code desktop: redirect_uri not registered
- MCP OAuth token exchange omits client_id in body, breaks public PKCE clients (Wrike)
- Tahoe regression #57024 expands: Claude Code's macl stamping now poisons sibling tool dirs (Cline) via parent propagation, breaking multi-harness backup pipelines
- [BUG] MCP slash command followed by task silently returns empty response in v2.1.132
- [BUG] Sub-agent permission mode inheritance broken — permissionMode frontmatter and parent acceptEdits both ineffective for Edit tool calls
- Feature: notation that forces inline expansion of references to prevent reference-following failures
- Kanban Framework Issues - Agent Orchestration Not Working
- Bring your own subscription to a client's Claude Code / Desktop session
- [BUG] Pro → Max upgrade fails with "Payment failed" — extra usage credits charge succeeds on same card
- [Bug] Anthropic API Error: Usage Policy violation blocks security research requests
- [Bug] Settings.json serialization drops enabledPlugins key and reorders hook properties
- Hook `command` field is wrapped in bash -c '...'; substituted paths with apostrophes brick the session
Resolved issues (230)
- [Bug] Session timeout occurring prematurely within minutes of activity
- [BUG] Claude Desktop 1.5354.0: stdio MCP servers double-spawn from independent CCD/LAM coordinator registries
- [BUG] MCP tool output not rendered in terminal when using native TextContent (fastmcp 3.x)
- [BUG] /model UI shows "Sonnet 4.6" but session jsonl logs claude-sonnet-4-5-20250929 for every assistant turn
- CronList display strips (one-shot)/[session-only] qualifiers; also no UI panel for cron/wakeup/loop entries
- /clear <name> argument not shown in /resume picker
- [BUG] pasted smal text as "Pastet text #N" randomly
- Context usage counter missing from terminal status bar
- AskUserQuestion: validation errors silently discard user-entered free-text answers and notes
- Autocomplete suggests space-separated plugin commands that don't resolve
- WSL2: Cannot paste image from Windows clipboard (missing PowerShell fallback)
- [BUG] Claude Desktop 1.5354.0: stdio MCP servers double-spawn from independent CCD/LAM coordinator registries
- [BUG] `/effort <level>` on first line of multiline message fails with misleading "Invalid argument" error
- [BUG] `claude mcp serve` Bash tool output is invisible in the MCP client's transcript
- [BUG] [regression] MCP tool results invisible in 2.1.88 — outputSchema safeParse guard returns null on schema mismatch
- [BUG] "Chat about this" on AskUserQuestion erases the original question from the visible transcript
- [BUG] Slash command autocomplete misses /terminal-setup on partial prefix /terminal-setu
- [BUG] /settings: language change in slash command is not persisted (Enter does not save)
- [BUG] /status footer hint shows "Esc to cancel" even when Escape is unbound in keybindings.json
- Mid-session prompt suggestions populate input buffer and submit on Enter
- [Bug] With autoScrollEnabled: false, viewport drifts by N lines per N new streamed lines once user prompt scrolls off-screen
- Slash command picker only activates at start of input, despite v2.1.0 changelog claiming "anywhere in input" support
- [BUG] /mcp displays list in a fixed scroll container, cannot view/take action on items outside of the default view
- [FEATURE] Allow plan-preview UI for skill-generated plans/specs, not only ExitPlanMode
- [BUG] /release-notes shows Version 2.1.92 on Claude Code v2.1.118
- [BUG] CLI process orphans at 100% CPU after window/terminal close (2.1.111, 2.1.121)
- Remote triggers: MCP tools fail to load and network proxy blocks outbound HTTP after initial run
- [BUG] model picker doesn't show gateway's models
- [Bug] Pasted text lost when prompt truncated; hidden line count inaccurate
- [BUG] Fullscreen mode: "Jump to bottom" overlay leaves color-block artifacts on wide characters (wcwidth-unaware clipping)
- [BUG] /agents Library list does not auto-scroll viewport — selection past first page is invisible
- [BUG] VS Code Claude Code extension fails to respond to a prompt, with no response to attempts to cancel
- [BUG] Bash tool completely unusable on AlmaLinux 9 — shell snapshot creation fails with "Unable to read input file: Is a directory", exit 127 on every command
- [DOCS] Amazon Bedrock guide shows `global.*` model ID despite region-prefix guidance
- [DOCS] Agent SDK MCP tool result docs omit `structuredContent` alongside image content
- [FEATURE] Feature request: cross-project session resume / global history view
- [Bug] MCP config parsing fails with invalid JSON in .mcp.json
- [Bug] Claude fails to consider alternative sync methods despite available tools
- [Bug] Claude 3.5 Opus degrades Danish language quality with mixed English words and unexplained code variables
- [FEATURE] Expose a command to resume previous session in VS Code/Cursor extension panel
- [BUG] No verification code received.
- [Bug] Claude fails to self-correct on file read-before-write errors
- [UX] Confusing project-open flow after recent desktop app update
- [BUG] OAuth session expires mid-session with 401, /login doesn't reliably refresh
- Allow disabling or recoloring autolink ANSI in chat output (markdown renderer)
- [Bug] MCP server configuration file not found: mcp.json missing
- [BUG] Desktop App Code tab crashes immediately on macOS 15 Sequoia
- [Bug] Keychain credential storage leaks orphaned entries on repeated /login
- [BUG] All claude.ai-connected MCP integrations show ✓ Connected but expose zero tools in Claude Code session (terminal CLI, macOS, 2.1.126, Opus 4.7)
- Claude deleted user files without permission and repeatedly fabricated technical claims
- [FEATURE] Show agent reasoning/thinking activity in Claude Code on the web
- [FEATURE] Add compactInstructions setting for default /compact arguments
- [BUG] Background subagents still prompt or deny Read/Glob/Grep in git worktrees despite global allow rules
- [FEATURE] Permission request command appearnce to be customizable
- [Bug] Anthropic API Error: Rate limiting consumes free ultrareviews without completing
- [Bug] Auto-update failed with npm installation error
- [BUG] Claude Desktop OOM crash on launch — JS heap hits ~3.4GB in ~12 seconds, macOS 26
- [Bug] Feedback mechanism accepts terminal input as feedback rating
- [BUG] Repeated edits to large HTML file silently break script execution
- [Feature Request] Add allow-list configuration for Skill-to-Skill loader permissions in settings.json
- [BUG] "trial ended" message with active subscription
- [BUG] --output-format json result field contains invalid JSON when model uses unescaped " in string values
- schedule skill: setup-time 'No connected MCP connectors found' but 9 connectors auto-attach at routine creation
- [BUG] Cowork: Stale rclone FUSE cache (cache_duration_s=3600) causes silent file corruption when Edit/Write tools and bash interact
- Refund Request: Significant Token Waste Due to Repeated Claude Code Errors - Account rorudit52@gmail.com
- You're out of extra usage · resets 9:50pm (America/Sao_Paulo)
- Mid-session hook configuration reload (settings.json + hook-script bodies)
- Web-Claude opens PRs from stale branches; runs git as root in user trees (silent corruption path)
- [BUG] Fenced code blocks in Claude responses render as inline code in VS Code extension
- Voice input does not recognize Russian speech on Windows 11 despite ru-RU language setting
- Routines: GitHub App install gap — OAuth authorize flow doesn't trigger install, errors are opaque
- Optional eager-schema-load hint for MCP servers / plugins
- [BUG] Review failed
- [Bug] Local commands hang after idle session periods
- Pendo MCP: every tool call returns "Anthropic Proxy: Invalid content from server" (related to #51812)
- [BUG] Weekly usage warning re-appears every turn after dismissal (desktop app)
- `claude update` hangs indefinitely on version-check when current == latest
- Bridge error: OAuth token forbidden (403) — missing scope or org access (scope user:inference user:office)
- API Error
- You've hit your limit · resets 11:20pm (Europe/Paris) : session reaching 100% instantly
- [BUG]
- [BUG]
- [Docs] terminal-config: color token reference covers ~35 of 69 tunable tokens
- Did not follow directions
- [BUG] 429 mid-send leaves branch unrecoverable with No message found error despite parent UUID present in local JSONL
- Cloud routine sandbox: local LFS proxy at 127.0.0.1:37713 returns HTTP 502 on every git lfs pull
- [Bug] Ultrareview session crashed without producing findings
- [BUG] GIT LFS not available in my remote routine
- [Feature Request] Add setting to disable auto-generated terminal tab titles
- [BUG] Symlinked .md files in .claude/commands/ are silently dropped from autocomplete
- Feature request: support Brave browser for /chrome integration
- [Bug] Anthropic API Error: Output blocked by content filtering policy
- Feature request: Read claude.ai conversation history from Claude Code
- [BUG] Remote Control push notifications on iOS: silent banner, no sound, no vibration
- Claude Code VS Code chat panel: code blocks overflow horizontally without wrapping
- [Feature Request] Add connection status indicator and last-received timestamp to detect dropped connections
- [BUG] ~/.claude.json can become corrupted if disk space is low
- Pre-execution token/cost estimate before approving plan-mode tasks
- 在 compact context 进行中时，用户可以重复点击按钮触发多次操作
- Warn users before hitting conversation rate limit
- /ultrareview: free credit consumed when remote review crashed before producing findings
- [BUG] TypeError: Cannot read properties of null (reading 'alwaysThinking') on new session launch
- [BUG] /ultrareview blocked: GitHub access fails on PR mode, repo too large for no-arg mode
- [Bug] Model violates self-diagnosed patterns within same conversation context
- [Bug] @-reference resolution fails with nested Git subrepositories
- [Bug] Memory retrieval not applied to conversation context
- Bug: Plugin command /omp:guide shows Unknown command but works with --plugin-dir
- [BUG] CLAUDE USES EXTERNAL SPEAKER IGNORING POLICY WHEN EARPHONES PLUGGED IN
- [FEATURE] Warn users prior to account suspension
- [BUG] Can not copy chat content in terminal, and can not drag images directly into terminal.
- [Bug] Slider interaction not responding to user input
- [BUG] Claude desktop 1.5354.0 hangs after launch — no window, no renderer process spawned (macOS 26.4.1)
- Status line: sessions missing and Conductor sessions initially miscategorized as Terminal
- VS Code linter validates SKILL.md frontmatter against DXT schema instead of skill schema
- Feature: Show usage reset countdown timer in status line
- [BUG] Dispatch cross-device session isolation and notification routing
- [BUG] Plan mode fails to restrict write tools when isBypassPermissionsModeAvailable is true
- Plugins cannot be enabled - toggle has no effect (v2.1.88, macOS, VS Code)
- [BUG] v2.1.87 idle process grows ~500 MB/min, OOM-killed within minutes on 16GB WSL2 system
- Claude Code Eclipse 플러그인(v1.2.1)의 패키징 버그
- Stale task/plan name persists in status bar across restarts
- [BUG] 2.1.89 Opus 1m no longer available
- [Bug] Context from `/btw info` command not retained in current session
- [BUG] ESC key interrupts Claude response when canceling IME composition (Chinese input)
- [Bug] Crash occurring during active session execution
- [BUG] Claude.ai Web - Playwright Plugin page shows no buttons and plugin marketplace loads infinitely
- bug: EnterPlanMode/ExitPlanMode disabled globally when channel plugins exist (should gate per-prompt)
- Phantom messages: teammate SendMessage content injected as Human: turns
- [Bug] CLI and desktop app generate different project memory paths for directories with spaces
- Auto-updater incorrectly detects native install as npm-global after migration from npm
- [Bug] Incorrect version number in release: 2.1.81 should be 2.1.88
- Bash tool `start` silently fails with Unicode/Arabic filenames on Windows
- Disabled MCP servers still show tools in deferred tools list
- I'm ready to help you generate a GitHub issue title for Claude Code. Please share the bug report or issue details you'd like me to create a title for.
- [BUG] Account Access Issue: Incorrect Deactivation of Professional Account
- Feature: configurable context compaction threshold (e.g. compact_at_percent: 66)
- [BUG] Windows flashes terminal window running "npm list" periodically
- Telegram plugin: inbound notifications/claude/channel not delivered to session (v2.1.89, macOS)
- API Error: 400 due to tool use concurrency issues on session start
- Slash command exact match should take priority over fuzzy match
- [Bug] CLI hangs or stalls during execution, performance significantly degraded
- bypassPermissions mode not persisted — resets mid-session and prompts for file edits
- Telegram plugin: multiple instances compete for getUpdates, stealing messages
- MCP server instructions from initialize response not injected into context for HTTP/remote servers
- [BUG] Cowork "Virtualization not available" on Windows 11 Pro — Hyper-V fully functional but HRESULT 0x80370102
- C:/Program Files/Git/resume collapses final text response in focus mode
- [BUG] Incorrect usage reset time for daily limit, for a 5:20-10:20 window, at 9 it is indicating 4 hours left to reset, weekly limit is on 50%
- Agent with multi-destination addressing loses tool-use patterns after auto-compaction
- [FEATURE] Controllable inference depth parameter for Opus 4.7+
- Add 'Yes, but...' option to tool approval prompt
- Plugins in claude-plugins-official marketplace lack `version` field in plugin.json, displayed as "unknown"
- [Bug] Plugin uninstall fails with case-sensitivity mismatch in settings.json lookup
- Frequent re-authentication required with multiple concurrent Claude Code sessions (OAuth refresh token race condition)
- I'd be happy to help generate a GitHub issue title, but I don't see the bug report details in your message. Could you please provide the bug report information? I need details such as: - What the issue is - What error message (if any) was encountered - Wh
- [Bug] Unable to determine issue from empty bug report
- [BUG] Enter key executes suggested user action when prompt out of view
- [Feature Request] Add configurable autoCompactThreshold parameter for conversation auto-compaction
- Add opt-out flag for auto-synced claude.ai connector MCP servers
- [Feature] /mcp reconnect-all: batch restart all MCP servers mid-session
- Cursor Style Interference: Claude Overrides Terminal Cursor Behavior
- Session environment cache not invalidated on /resume, causing stale env vars
- [BUG] --worktree <name> silently falls back to parent repo cwd when worktree name already exists
- [BUG] File locator doesn't include enough directory entries when searching
- [BUG] Title: Subagents don't inherit .env deny rules from settings.json
- [BUG] All claude.ai built-in MCP servers disappear on every restart — require /login each session
- Bundled bfs find rewrite holds ~65k DIR fds per call on large monorepos, contributes to ENFILE
- CLAUDE_CONFIG_DIR not respected by VS Code shell-integration lock files
- [BUG] --resume picker does not show session when project path contains underscores
- [FEATURE] auto /clear
- /usage tips block leaks into terminal after dialog dismissal
- Alt-V No image found in clipboard on Windows 11 despite image being present
- Enhance Internationalization Documentation with Unicode Examples
- Enhance Internationalization Documentation with Unicode Examples
- AWS Bedrock / SSO Authentication for Claude.ai Desktop & Web App
- [BUG] approaching usage limit
- [MODEL] its not responding Properly
- [Bug] Claude Code process exits with code 1 on crash
- [BUG] Claude Code extremely slow- Conversation disapearing begining of session but still running - 30min+ for a small plan mode.
- [BUG] Unexpected SSH connection to GitHub on startup triggers FIDO key Touch ID prompt
- Agent bypasses git pre-commit hooks using --no-verify, stash, and quiet flags despite explicit deny rules
- [MODEL] Claude Code repeatedly violates its own documented protocols, wastes CI/deploy minutes, claims tasks complete without verification
- [FEATURE]
- [BUG] Memory/feedback system does not reliably change agent behavior across sessions
- Bug: Claude Code ignores explicit user instructions (working directory), wasting tokens and money
- --dangerously-skip-permissions flag no longer bypasses permission dialogs
- [BUG] Bash tool bypasses settings.json path deny rules — shell commands not subject to file tool access controls
- Claude Code attempted to post confidential project information to public repository despite explicit rules prohibiting it
- Feature request: Auto-generate session name from conversation content (like Gemini)
- [BUG] Tamil vowel signs (ோ, ே, ா) render incorrectly in spinner
- Standalone .md skill files in .claude/skills/ not discoverable via slash command search
- [BUG] - Endless loop and excessive tool calls by sub agents contributing to massive spike in token consumption (Max plan user)
- Plugin submission: tebra-mcp-server — Healthcare practice management MCP
- [FEATURE] Memory-backed constraint enforcement and pre-action gating
- Exit message rendering garbled — text concatenates on one line
- Bug: prompt-type Stop hooks always fail JSON validation, forcing dangerous workarounds that cause runaway background processes and unexpected billing
- [Bug] Unintended prompt text injected into conversation context
- Haiku auto-downgrade mid-conversation destroys user work — please stop silent model switching
- [BUG] Token Refund Request] Abnormal token consumption due to system error
- Flash Card Options at each Decision Point
- [FEATURE]
- Dispatch session collisions cause message loss and transport failures
- Dispatch panel doesn't render results despite successful completion
- [Bug] Context amnesia in long sessions — constraints silently dropped as context grows
- Claude Code deflects blame and offers unsolicited alternatives instead of acknowledging failures and doing what's asked
- [Bug] Claude does not verify file edits applied correctly — blind edit-and-proceed
- [Bug] Claude never proactively surfaces its own mistakes — requires user to act as quality gate
- [Bug] Claude generates incorrect code/SQL and reports it as complete without validation
- [BUG] Explore sub-agents report false findings without verifying via Read/Glob
- clangd LSP plugin: all operations fail with 'trying to get AST for non-added document'
- [FEATURE] Preserve and reuse thinking trace when user interrupts mid-thinking
- [BUG] ABSOLUTE RULEs Considered as Advisory rule
- Plugin marketplace auto-sync fetches but never pulls, so updates are never applied
- Persist /color setting across sessions via settings.json
- [FEATURE] Attaching content via remote session
- [BUG] Claude Code ignores "No" permission responses and executes bash commands anyway
- Plugin skills: standalone registration removes namespace discoverability and namespace form lacks argument-hint
- Agent deleted 2,229 untracked source files without explicit user instruction — catastrophic unrecoverable data loss
- [BUG] Plan mode does not hard-block write tools — model can make edits and user approval bypasses the constraint
- Bug: Sandbox half-broken — writes hit real filesystem, reads sandboxed — destroyed entire project
- [BUG] --dangerously-skip-permissions does not bypass runtime tool execution prompts
- [MODEL] Main agent's autonomous subagent delegation produces poor scoping, missing context, and no result verification
- Telegram plugin: multiple sessions compete for getUpdates, messages route to random session
- Expose current plan name and file path in status line JSON data
- Feature Request: Add Traditional Chinese (zh-TW) UI localization
- Claude-in-Chrome: macOS screen recording indicator stays on with multiple sessions
- 529 Overloaded errors kill parallel subagents, no retry — work lost
- [FEATURE] Claude Desktop: show "Last turn changes" in the diff viewer
- [BUG] Timeout issues with Opus 4.6 persist despite status page resolution
- bypassPermissions on agents ignores settings.local.json allowlist
- [BUG] PowerShell tool returns exit 1 with no output for all commands (regardless of allowlist matching)