Digest / October 13, 2026
Claude Code Issues Digest — October 13, 2026
257 new issues filed ·
187 resolved in anthropics/claude-code (UTC day).
New issues (257)
- I can't see the prompt which should be shown to ask for my approval
- [Bug] /compact blocked by safety classifier despite pre-compact warning (Opus 5.5)
- Claude in Chrome extension permission popup fires repeatedly within an already-approved session, ignoring Claude Code permission mode
- [Bug] Safeguard blocks benign agent message-analysis coding with no reason
- [BUG] Japanese voice dictation inserts a full-width full stop 。 at the start of each new dictation segment (macOS desktop app, Code tab)
- [BUG] Voice dictation sends the dictated text twice when a character is typed through the Japanese IME before Enter (macOS desktop app, Code tab; reproduces every time)
- [MODEL] Opus 5.5 - Palantir showed up in thinking? Model denied a root cause analysis and session got blocked?
- Subagent sent the user's email from session context in a User-Agent header
- [Bug] Legitimate security review of own code blocked with [cyber] flag
- [Bug] Opus 5.5 cyber safeguards falsely block legitimate security study materials
- [Bug] Repeated factual errors in summaries: incorrect counts, overclaims, misread context
- [Bug] Auto mode blocks harmless actions like reading project context files
- [GitHub integration] Missing Github option
- Preventing Claude within Projects from manipulating system systems or file dele…
- MCP tool place_stock_order fails with type validation error when passing qty or notional
- [Bug] Haiku subagent fails after Skill use: 400 system.4.type must be 'text'
- [BUG] Files of all types uploaded via both the plus (+) button and drag-and-drop cannot be recognized
- demorar en cargar
- Bug Report: Unspecified error in Fable 5 and Sonnet 5.5 desktop sessions
- [FEATURE] Desktop Code tab: allow pinning sessions to the bottom of the sidebar
- [MODEL] Restored a test deployment without checking for dangling links, leaving four on each of two hosts
- [GitHub integration]
- [BUG] - API Error: Fable 5.1's safeguards flagged this message (https://www.anthropic.com/legal/aup). Our intentionally broad safeguards allow us to deliver more capabilities faster, but can sometimes flag legitimate coding and cybersecurity tasks.
- it is consuming my plan's usage instead of the setup credit given(still valid)
- [Bug] Set-Clipboard leaves ownerless OLE object, stalling RDP clipboard
- [BUG] Selecting "Local" consistently creates cloud sessions instead
- [Bug] Auto mode safety check blocks writing skill files
- [FEATURE] Mods: a launcher surface for icons in the desktop toolbar, top nav, or a right-edge dock
- [Bug] Mac app requires TCC access to Desktop, Downloads, and Documents instead of per-folder access on Mac
- [FEATURE] Show which installed skills and connectors fit a prompt, and let the user choose
- [Bug] Claude Code Auto-Submitted Message as User on Startup in Bypass Permissions Mode
- [BUG] Agent installed an end-of-life dependency with 37 published CVEs after a failed registry lookup, with no vulnerability check
- Browser "Allowed sites" allowlist ignored — allowlisted site still prompts on every action ("Site-level permissions are disabled for this site")
- [BUG] Interactive sessions write the 5-minute prompt cache from session start, while `claude -p` on the same account writes 1-hour (2.1.296, Team seat, survives `/login`)
- [GitHub integration]
- [BUG] Long-running sessions on older builds stuck on ECONNREFUSED for every request until /exit + --resume; newer-build sessions on the same machine unaffected (error omits host:port)
- [Bug] context: fork skill returns before background subagents finish in interactive mode
- [BUG] Multiple panel sessions rapidly steal input focus from each other when VS Code window regains OS focus
- Coordinator relays to a Remote Control session arrive inside tool results when the session is mid-task
- [BUG] Claude Desktop sessions disappeared after reverting to Pro from Max (Linux)
- [Bug] Opus 4.7 Real-Time Cyber Safeguards Flag Legitimate Security Work
- Option to disable automatic conversion of image file paths into [Image #N] attachments in the prompt input
- Phantom mid-turn user message injected into session (desktop app, Windows)
- Severe: Claude Code runs a multi-hour multi-agent pipeline on a UI-only change and ignores explicit ordering instructions
- Bash tool: injected `grep` silently skips .gitignore'd and binary files
- [GitHub integration]
- [Bug] CLI redraws to original terminal size on resize, causing visual glitches
- [Bug] /clear and /new carry over the /rename session name to the new session
- [FEATURE] Show an estimated time until the current prompt is done (ETA) and its cost so far, by default
- [Bug] Agent skips mandatory checks, omits test run, and proposes hacky workaround
- ListAgents: peer rows don't show cwd/branch, so the agent can't resolve "the <task> session"
- Let a stream-json host declare an attended session, so background commands have no time limit
- [BUG] [Tool Routing] Deferred MCP tools lack descriptions in ToolSearch, causing Skills with overlapping capabilities to always shadow them
- [BUG] Claude is useless for testing anything
- po22
- [BUG] Remote Control: worker's custom skills still missing from Claude app slash-command autocomplete (follow-up to #62482)
- [BUG] Claude Desktop (Windows on ARM): extensions with server.type "uv" fail with "Checksum verification failed"; managed uv download never completes
- [Bug] Auto mode classifier denies Chrome MCP type actions after exiting auto mode
- [BUG] Claude Desktop (Windows): sessions + dynamic workflow + warm previews exhaust RAM and freeze the whole system; memory governor only evicts idle sessions ("evicted 0 of 0")
- [GitHub integration]
- [BUG] VS Code extension: voice dictation stops while the user is still speaking (hard 2-min cap, 15s timer based on transcripts instead of audio, no indication)
- [GitHub integration] I have connect github connector in claude web but AI it cannot find the tool connect
- [FEATURE] Mods: document the 4-concurrent-clip limit of $.audio.play
- [Bug] False-positive reasoning_extraction safety blocks interrupt Claude Code sessions
- [BUG] `"async": true` is silently ignored on `type: "http"` hooks — the turn still waits for the response
- [Bug] Image input processing is slow compared to ChatGPT and Gemini CLI
- [GitHub integration]
- [Bug] Safeguard triggered without contextual reason on hypothetical open source game setup
- Desktop (macOS): built-in browser "Import cookies" lists only Edge until each browser is allowed under Files & Folders; the dialog gives no hint
- [GitHub integration]
- [Bug] Unexpectedly flags benign requests as cyber-related
- [BUG] macOS: saving an MCP server's OAuth tokens wipes the shared Keychain entry, signing out every session (regression of #45551)
- Bug Report: Unclear issue, user asks why content is labeled "cyber"
- [GitHub integration]
- [Bug] Teammate agents: conversation logs and Bash output not visible
- Agent re-argues explicit user decisions and turns obvious fixes into questions
- [BUG] Plugin skills are reported as "custom_skill" in skill telemetry instead of their plugin:skill name
- [Bug] Sonnet 5.5 safeguards falsely flag benign herdr workload message
- [Bug] Package update workflow produced 100k+ unintended file changes
- Bug Report: Siemens PLC password unavailable for water-treatment pilot plant
- First session in a new folder does not load AGENTS.md (2.1.296)
- [BUG] Desktop Files pane in Remote Control sessions: unsorted (raw readdir order), folders not expandable
- [Bug] statusLine command in settings.json never invoked on macOS (CLI 2.1.296)
- [Bug] Anthropic API Error: Request Failed (req_011Cfz4vfzwvQ3jWMQZj3Xh6)
- [BUG] VS Code extension: after "Show more" on a long pasted prompt, "Show less" is unreachable (chat cannot scroll far enough)
- [GitHub integration]
- [BUG] safety classifier is *extremely* sensitive and impedes much of my work
- [Bug] Agent refuses benign cybersecurity report task, misclassifying it as harmful
- [Bug] Cyber safeguards triggered when analyzing user's own application
- [Feature Request] Show progress/percentage for background tasks in shells
- [FEATURE] Mods: engine-owned prompt.submit receipt linked to native turn IDs
- Please we need dangerously skip permissions on threads and instructor
- [FEATURE] Split and rearrange plugin (mod) panes in the desktop app
- [FEATURE] Expose MCP server connection status to statusLine commands and hooks
- [GitHub integration]
- [BUG] `claude` processes spawned from a desktop Code session ignore the session's account and fall back to the OS user's default CLI profile (silent switch to an org account: billing + data)
- Desktop split view: plugin AbovePrompt band and side pane only render in the primary pane
- [BUG] Mirror for improperly closed issue about dangling .cc-writes directories in project dir
- [BUG] Desktop resume restores a worktree without a current session lock during Bash execution (2.1.295)
- [FEATURE] Plugin slash commands with immediate: true should run without disturbing a pending AskUserQuestion in the desktop app
- [Bug] Claude refuses to generate install.sh due to safety restrictions
- wont code project after being corrected
- Tung [GitHub integration]
- [FEATURE] Plugin API: replay the main thread's last request as sent (cache pre-warm) for keepalive mods
- [BUG] statusLine.refreshInterval ignored in VS Code extension — script never re-invoked on interval
- [CRITICAL RISK WARNING] Model makes unverified security conclusions and false reports (proposed blocking 172k unrelated IPs)
- can't see the text sent by the coordinator on the website and on the claude app
- [FEATURE] Mods: render sites in the desktop app's empty space (sidebar bottom, chat margins)
- [BUG] Windows MSIX: shipped .msix is not streaming-ordered (AppxManifest.xml at 99.6% of the archive) — every URI-based auto-update fails with 0x87E80034
- [BUG] VS Code WSL authentication fails without actionable error when Windows and WSL have different effective internet connectivity
- [BUG] Microphone only in initial chats, than vanishes for all other older chats, or second statements (Chrome/Ubuntu 26.04) since yesterday
- Projects beta threads are now unable to modify excel. It was working earlier in…
- [GitHub integration]
- [BUG] Android App: Sharing media/files does not work since yesterday
- AskUserQuestion: render question bodies as normal-weight markdown (follow-up to #78204)
- 雲端連線中斷 無法正常寫回檔案
- [MODEL] Opus 5.5: False positive: safety classifier halted a legitimate reverse-engineering session
- Auto mode classifier blocks a read-only reachability check of the user's own remote browser endpoint as "Containment Escape"
- Desktop app: let the user name the worktree and branch when starting a session
- [Bug] Claude Code fails to follow explicit instructions and executes unintended SQL operations
- [BUG] File preview in Claude Desktop is unnecessarily thin. (Small)
- [Bug] Opus 4.5 falsely flags network switch management tool (Telnet/SSH) as intrusion attempt
- [Bug] Claude ignores source docs, misreads units, and acts on feedback-dialog input as work instruction
- Opus 5.5 cyber safeguard blocks non-security messages after session flag
- [Bug] Test report - no actual issue content provided
- [Bug] Request Asked Claude to Explain Its Reasoning, Response Not Shown or Logged
- [Bug] Unexpected feedback survey prompt intercepts numeric input
- [Bug] Claude Code refuses assistance during forensic tool development
- Claude gateway: let a session choose which group (billing code and budget) it bills to
- claude.ai renders a bare issue number (#101769) as a styled chip that is not clickable
- [Bug] Plugin API: a plugin's own hooks are skipped for $.agent.spawn made after its hook returned (auto mode then refuses the spawn)
- Cloud session sandbox proxy returns 403 for the Browserbase API, breaking skills that depend on it
- Auto mode classifier blocks reading a file from the user's own installed skill as "Containment Escape"
- Session context tells Claude that memory is disabled when the user has it enabled
- Gmail connector list_drafts has no way to limit output; four drafts overflowed the tool result
- [BUG] --ax-screen-reader: AskUserQuestion option text is truncated at 300 characters ("+N more characters") with no way to view it in full
- [BUG] Desktop app: prompt suggestions never appear in an SSH session
- [BUG] Claude Code spams my notifications with irrelevant information
- [Bug] Agent execution hangs for ~100 minutes, unresponsive
- claude.ai cloud session: auto mode classifier labels the user's own typed approval a "cross-session message" and blocks the approved action
- [FEATURE] Verifiable human approval for external tools in Remote Control
- [BUG] Refusal followed by terminal-only AskUserQuestion dialog blocks channel-driven sessions
- Je crie un nouveau plan. Je crie un nouveau plan. [Bug] Unverified claims and untested migration plan ignore CLAUDE.md rules
- [Bug] Prompt classifier incorrectly blocks valid prompt
- Here is the draft. Paste it into the feedback form or a GitHub issue.
- [FEATURE] Mods API: expose the session's current effort level
- [FEATURE] Client-side prompt comment (,,) that are stripped before API transmission
- [Bug] Agent fabricates content that contradicts facts in the codebase
- [Feature Request] Add setting to hide permission-mode badge
- [FEATURE] A current task for the task tools, with parent tasks and descriptions people can see
- Writing Permissions on Files in Claude Code (Code myself in Claude Code Desktop)
- Channels: the expanded view (ctrl+o) does not show a channel message's content; the one-line summary is cut at a fixed width
- [BUG] Mods: hardware cursor stays at the prompt while a Client has key focus, so IME composition (Korean) is drawn in the prompt box
- [BUG] Mods: Client onKey gets a lone space as undocumented "space" and unnamed keys (F1, Insert, F12) as raw escape sequences
- [BUG] Agent view: finished conversations stay in "Ready for review" indefinitely, with no way to move them to completed
- [BUG] Marketplace with plugin "source": "." passes `claude plugin validate` but fails to sync on claude.ai
- Auto mode classifier blocks a user-directed procedure on the user's own server as "Auto-Mode Bypass"
- Auto mode classifier blocks grep of the user's own installed skill file as "Instruction Poisoning"
- [MODEL] Did not mark a schedule lane in progress when its work was written and committed
- [FEATURE] Allow submitting an issue in one paste (free-form template), and fit non-terminal bugs
- [Bug] Opus safeguard over-triggers on benign requests
- [FEATURE] Provider-signed session receipts: verifiable "made by model X, unedited" for commits and PRs
- Plugin API: $.agent.register silently drops cacheTtl; no way to set a runtime-registered agent's cache TTL
- [FEATURE] Let Claude search existing GitHub issues (read-only, public repos) to prevent duplicate reports
- Bug Report: Agent takes destructive actions and ignores user instructions
- [BUG] VS Code: a session picked from History in a new tab opens in the sidebar instead, if the sidebar has had it open
- Claude Code Projects: i send a msg then nothing happens. claude says its typical {verb} to indicate i…
- Plugin API: a background agent from $.agent.spawn loses its report under acceptEdits (completion notification skipped as already-notified)
- [FEATURE] VS Code: keep the message box usable while Claude asks an interactive question
- Bash approval required but no prompt shown in embedded Agent SDK host
- [MODEL] Opus 5.5 identifies with the user and describes the user's actions as its own
- [FEATURE] Threads: ask follow-up questions under one message without leaving the history
- [Bug] Background agent doesn't receive task result until main session pings it
- You can't turn Bypass on in a Project's sub task
- [Bug] Opus 5.5 shows "This model requires a Pro plan" on Max plan
- [BUG] Claude Code and VS Code extension are not able to see all tools from an MPC
- whenever i try to control things on my laptop using the projects for co work it…
- [MODEL] Wrote that a tool fails over ssh as if it had run it there, when the failure was inferred
- [Feature Request] Store AskUserQuestion prompts and answers in transcript for checkpoint/resume/revert
- [MODEL] Sent remote commands through the login shell instead of piping them to bash, against a no-exceptions rule
- [Bug] Claude Code still prompts for Fable credits after upgrading to Max
- REPOSITORIOS
- Agent isolation: "remote" silently falls back to a local worktree agent with no indication in the tool result
- [GitHub integration]
- [MODEL] Wrote a command's output to a temp file in /tmp, against a no-temp-files rule
- [Feature Request] Allow customizing Ctrl+C to a single key press for accessibility
- [GitHub integration]
- [BUG] MCP tools/call rejected with "missing required resultType" from a server that negotiates 2025-06-18 (Microsoft Learn MCP, 2.1.296)
- Hover group doesn't update when content changes under a resting pointer
- This issue has occurred many times since the update. My typical usage involves …
- [FEATURE] A setting for Edit and Write to write exactly what they're given
- [BUG] claude.ai connector: transient 403 (mcp_request_blocked) is cached as needs-auth; later sessions skip the connector and re-authorizing can't clear it
- [BUG] Remote Control spawned sessions exit with "Process exited with error" mid-task; app reports host offline
- [Bug] Concurrent Claude Code sessions in same repo leak tool outputs across sessions
- [MODEL] Ran a lint gate that calls git, against a ban on running git or any script carrying it
- Plugin API: session.usage().rateLimits omits the per-model weekly window that /usage shows
- [MODEL] Handed over a script that ran a tool on a second host without checking that host's build of it
- [BUG] I was asked for permission to run something even on BYPASS PERMISSIONS
- Auto mode classifier denies an explicitly authorized Bash command, with no manual approval prompt
- [BUG] Desktop scheduled tasks stall waiting for permission approval, which defeats unattended runs
- I'm not able to share my project with anyone
- "Usage limit reached" message continues popping up after using reset
- [FEATURE] Syntax highlighting for R scripts (.R, .Rmd, .qmd) in the desktop app, like Python
- [BUG] Registry sweep deletes other hosts' live session records when hosts share $HOME and /etc/machine-id (HPC clusters, cloned images)
- [BUG] claude.ai connectors that differ only by URL query string are deduplicated to one in Claude Code
- [MODEL] Assumed a tool's current behavior on a host running an old build, twice, and a "dry run" wrote files
- [GitHub integration]
- [GitHub integration]
- [GitHub integration] Claude Code Cloud shows no repositories
- [Bug] Safeguards block legitimate router security incident response
- [BUG] /desktop on a terminal session with Remote Control on leaves two Desktop sidebar rows for one conversation
- [BUG] Desktop: `immediate: true` commands wait for the turn to end
- [MODEL] Named files by paths relative to a clone, against a rule that every file is named by its full path
- Claude Code webview blank in WSL after VS Code update, 401 on webview resources
- [Desktop] File pane: rendered Markdown locked to a narrow fixed-width column while tables span full width
- [GitHub integration]
- [GitHub integration]
- Windows: Ctrl+V can paste images in Windows Terminal, but chat:imagePaste makes every text paste take ~0.5 s (two PowerShell spawns)
- Browser tools aren't loading in new project threads
- I can't add folders to Claude Code
- [Bug] Channel notifications dropped after session handoff to background processthat
- [GitHub integration]
- [MODEL] Answered a why question at length, a day after being told a reply gives its answer in a few lines
- [Feature Request] Auto mode permission handling for directive file edits
- Built-in git instructions can't be removed by the user
- macOS: Keychain read timeout under load is cached as logged out for 30s ('Not logged in · Please run /login')
- Model picker shows "update Claude Code" for Opus 5.5 / Fable 5.1 even though CLI is on latest version
- Function-hook session.compact replacement re-saves kept messages as new entries, so the desktop app shows them duplicated
- Claude Code is not following my rules to merge PRs. I never want to merge PRs. …
- VS Code extension: effort level sometimes missing from the chat UI
- [BUG] Task chips from spawn_task don't show up in the Claude mobile app
- [BUG] Plugin Input empties on Enter while the hook still draws the same value
- Mods: a mod's own $.mcp.call passes auto mode in a terminal but is refused in the Desktop app, on the same build
- [BUG] Claude executed a pre-existing script after its own write to that file was rejected, causing unintended deployment
- I don't see the earlier report or the title I generated in this conversation. Could you paste the bug report again? I'll write a new title for it, and the response will contain only the title.
- Agent tool isolation:"worktree" spawns a seat whose every Bash call is refused on arrival (analyzer fails closed on plain commands too)
- claude -r picker hides sessions whose first message contains a pasted image
- [BUG] claude.ai "Add marketplace" hides marketplace_name_taken behind a generic "failed to add" error
- CLAUDE_CODE_OAUTH_TOKEN is removed from all subprocesses (Bash tool, subagents, hooks, MCP) without CLAUDE_CODE_SUBPROCESS_ENV_SCRUB; undocumented
- [Feature Request] Desktop split view: new session should open in the focused pane
- [FEATURE] Turkish (tr-TR) UI localization for Claude Code
- Bug Report: Unclear complaint about Claude Code "insubordination" and refund request
- [Bug] Unexpected Usage Charges, Refund Requested
- [BUG] Desktop "Allowed sites" dialog rejects every domain with "blocked by your organization's policy" until the site is added at claude.ai
- [Feature Request] Add Turkish language localization for TUI
- [BUG] Context overflow 400 on self-hosted model after updating to 2.1.296: compaction doesn’t prevent or recover from “maximum context length” error
- [Bug] Claude Code Refuses Remote Machine Troubleshooting as Cyber Attack
- [BUG] Claude Desktop (Windows): cowork-vm transiently holds ~18 GB host RAM, not attributable to any process
- Remote Control sessions in a Claude Project keep blocking approved actions with Trusted devices OFF.
- Extension host OOM-crashes within seconds to ~1 min of launching 2+ parallel background Agent subagents (Windows)
- [MODEL] Reported what a deploy skipped as fact, when the tool's output named nothing
- Opus 5.5 Session wrote a file to my computer through the desktop app link without being asked, and did not tell me
- [Bug] Invalid or malformed bug report submitted
- [BUG] Canceled foreground subagent provider request has no terminal llm_request span
- [Bug] Ctrl+G external editor prefill omits assistant text from multi-message turns and truncates to 50 lines
- Bug Report: Missing Issue Details
- Claude writes Monitor log watchers (`tail -F | grep | while … break`) that never exit, so the task shows "Running" up to 30 min after the build ends
Resolved issues (187)
- [BUG] Auto-compaction fires at 281K–845K tokens on 2.1.296 for claude-opus-5-5 (documented default ~967K)
- [GitHub integration]
- Desktop app bundles Intel-only node-pty prebuild, triggering macOS 27 "will not open in macOS 28" warning
- [BUG] transcript usage lines record the requested model id instead of the served one since 2.1.296
- Let users turn off the AI attribution footer in cloud sessions (GitHub posts)
- Autocompact thrashing error with a small autoCompactWindow: counter is per model response, message blames files/tool output, subagents are killed
- Claude refusing to make monoliths
- Bug Report: Siemens PLC password unavailable for water-treatment pilot plant
- [FEATURE] Claude desktop app chat (not Code): on-demand context usage query, readable by the user and the session
- I appreciate you reaching out, but I need to clarify: I'm designed to generate GitHub issue titles for Claude Code (Anthropic's CLI tool) based on bug reports written in English. Your message appears
- [Bug] SSH/Telnet connections dropping unexpectedly
- Thread hung and became unresponsive.
- [Bug] Login Authentication Failure on Local Machine
- [Bug] Opus generates repetitive agent loops and produces faulty tests requiring multiple edits
- [Bug] Anthropic API Error: Safety classifier blocking file creation for approved commands
- [GitHub integration]
- [GitHub integration]
- [GitHub integration]
- always having problems linking my computer
- AI do not stop. when i made the typo so i tried to stop the ai. the problem was…
- [BUG] MCP tool calls in CCR Routines fail with "requires approval" — no approval UI shown, reconnect does not resolve
- Assistant text block dropped (UI + transcript) when followed by an interleaved thinking block before a tool call
- [Bug] Anthropic API Error: Reasoning extraction access restriction
- Desktop: session group names sync across devices via account, sessions don't — no opt-out (privacy concern for isolated work/personal machines)
- [BUG] macOS 27 Golden Gate beta: Claude Code re-requests Files & Folders access on every auto-update (new TCC entry per version)
- Cross-session messaging (agents_cross_session_inbox) is enabled on Linux but not Windows for the same account
- [Feature Request] Daemon mode with background process management and session persistence
- MCP tool results missing content blocks: accessibility tree never surfaced, screenshot intermittently dropped
- Remote Control session goes permanently unresponsive (process alive, no crash) after a Bash tool call times out; mobile app shows it as unrecoverable 'Disconnected'
- OAuth refresh token rejected server-side after ~24h - forced /login every day (Max sub, single machine)
- claude-dedupe-issues.yml logs github_duplicate_comment_added on every run, even when no duplicate comment was posted
- scripts/gh.sh fails with "FLAGS[@]: unbound variable" on flagless invocations under macOS bash 3.2
- [Bug] Anthropic API Error: Invalid tool_use_id in advisor_tool_result after session compaction
- [Bug] Subagents loop 9 extra turns after end_turn, returning filler text instead of report
- [BUG] `protocol mismatch` error when trying to install Claude CLI
- MCP Apps widgets stopped rendering after staged rollout of server/discover version negotiation (2.1.234)
- [BUG] VSCode extension: markdown links to directories in chat are dead (clicking does nothing), though the system prompt instructs Claude to emit them
- [BUG] No document preview in Claude Code, despite Home tab supporting it (same account/product)
- Desktop (macOS): reopening an existing conversation shows a blank transcript (history not rendered), though context is retained
- [BUG] Korean (Hangul) text is garbled/corrupted in AskUserQuestion and TodoWrite card UI (VSCode extension)
- --print returns "Not logged in" when $USER is absent/wrong: keychain OAuth lookup keys on $USER with no uid fallback (macOS)
- Plan panel comments not relayed to Claude
- ToolSearch: the `+term` prefix matches descriptions too, so it does not "require the term in the name"
- [BUG] Status line disappears after assistant response and does not re-render (still reproducible in 2.1.205, refreshInterval workaround does not help)
- [BUG] Claude Code launched via wsl.exe-relayed WSL terminal triggers Windows kernel BSOD (0x1A/0x4477) — safe via SSH-remote WSL session
- Auto-mode safety classifier blocks user's own credentials for their own accounts (local menu bar app setup)
- Desktop app update wipes `claudeAiOauth` from .credentials.json, silently logging out Claude Code and killing Remote Control sessions (Windows)
- claude.ai connectors and local stdio servers yield zero tools at session start (desktop); only a manual connector re-toggle — which uses a different registration path (UUID vs display-name) — makes them appear
- [FEATURE] Priority-aware memory index: let entries be marked never-truncate, and trim by priority instead of position
- [BUG] Cowork Project Memory panel reads from a disconnected file store, silent mismatch on 12/12 tested projects
- Vim mode: no rebindable action for "jump to agents panel" (← is hardcoded)
- Persistent microphone permission banner cannot be dismissed (Windows desktop app)
- [BUG] Routine unprompted instant delete
- [BUG] AskUserQuestion: choosing "chat about it" erases the question and all its options from the transcript
- [Bug] Benign request falsely blocked as reasoning_extraction after batching_reminder injection
- [BUG] Incident report — Claude Code agent reported a false verification (2026-08-22)
- CLAUDE_CONFIG_DIR does not redirect user memory — ~/.claude/CLAUDE.md still loaded (undocumented)
- Scary to see the tasks that are running and then coming back with some stale messages minutes after the request is completed
- Simple session-history lookup took ~10 min, wrong answer given, recall.py silently truncates on UnicodeEncodeError
- Auto-mode classifier blocks routine API POST with no way to grant standing permission; !-prefix input hard-wraps long lines
- Windows: Claude Desktop AppData\Claude grows unboundedly, no relocation option; symlink/junction workaround breaks app on startup
- [BUG] ~/.claude/commands/ (symlinked onto Windows drvfs) entirely invisible to non-interactively-launched sessions
- [MODEL] Fable High Triggered Itself to Keep Coding After I Stopped It
- [BUG] When trying to maximize the terminal panel while a claude extension tab is open, terminal isn't maximized, claude prompt is focused instead.
- [BUG] Apps gateway serves Claude Desktop an `otlpEndpoint` pointing at its own bearer-gated OTLP ingest but no `otlpHeaders`, so every Desktop telemetry flush is rejected with `missing_token`
- Add a setting to disable/customize the terminal title (and refresh it so it doesn't go stale)
- [Bug] Incorrect reasoning extraction from non-Claude Code context
- [BUG] : Routine is not working properly in Claude Paid version
- Sandbox network egress allowlist not consistently enforced (non-allowlisted hosts intermittently reachable)
- [FEATURE] Request: Persist per-turn token/time usage as text in the session transcript
- [Bug] Claude Code loses context after API failures and denies error occurrence
- [Feature Request] Implement answer-first response discipline and require explicit permission for actions outside local scope
- [BUG] Plugin without "version" in its manifest records the enclosing git repo HEAD (~/.claude) as its version
- Session rename does not propagate to a connected Remote Control session
- [BUG] VS Code extension: chat panel blank on non-secure (HTTP) origins — unguarded crypto.randomUUID() in webview bundle
- CLAUDE.md (user-level, ~/.claude/CLAUDE.md) intermittently not loaded in new sessions (claude-desktop)
- Title: Bun runtime segfault crashes claude-vscode after long-running session (v2.1.229)
- Background subagent (Agent tool) wedges before first tool call, never self-recovers, must be manually killed
- Custom subagents fail to register after cold start when settings.local.json grows very large (348KB / 2438 valid permission entries) -- fixed by removing the file
- [BUG] MCP Apps widgets never render in third-party (custom-3p) deployment mode — empty feature-flag payload disables the epitaxy MCP Apps gate
- [BUG] Browser MCP crash (Browser MCP bridge crashes entire app when Windows username contains a space)
- First session in a new folder does not load AGENTS.md (2.1.296)
- [BUG] VS Code: "Resume session" on a conversation that entered a worktree starts a fresh empty conversation instead
- Tung [GitHub integration]
- [BUG] Claude is useless for testing anything
- [BUG] Desktop: new session's slash menu omits anthropic-skills:* until the first message (temporary query not given the skills plugin)
- [BUG] Windows: the PowerShell command-safety parser's encoded launch trips antivirus heuristics (6 alerts per affected command)
- Je crie un nouveau plan. Je crie un nouveau plan. [Bug] Unverified claims and untested migration plan ignore CLAUDE.md rules
- [BUG] Ctrl+Shift+Tab is partially intercepted by Claude Code and also triggers Shift+Tab mode switching
- Desktop usage card counts each API call about 2.2x (sums every transcript line, not each unique call)
- [Bug] VS Code extension: Tool use concurrency error after Jan 23-24 update
- [BUG] Claude Code and VS Code extension are not able to see all tools from an MPC
- Custom themes are ignored in markdown rendering: inline code uses the built-in dark colors
- bypassPermissions silently downgraded to Manual in daemon sessions; detach/re-attach strips bypass from an interactive session
- Al Hawara
- [BUG] Desktop (Windows): all custom sidebar groups wiped after usage-limit pause/resume, no app restart
- [Feature Request] Interactive companion pet with dynamic reactions and state awareness in Claude desktop app
- i can't add existing chats to my own project
- AIC designer
- [BUG] : <'CEO : M - [bet -ø : zet[a] : <concord 'BSL' : tree(coverage - > accesing : 'python-via :script(DLC)')>]'>
- [FEATURE]
- claude.ai web: "How is Claude doing this session?" banner keeps reappearing over the composer with no way to turn it off
- [Bug] Tool calls fail repeatedly requiring rewind to recover
- [Bug] Untracked file "forge/Immagine.png" preventing operations
- [Bug] Agent obscures or hides incorrect responses mid-execution
- [Bug] Unclear error flagging on present-tense capability questions
- [Bug] Unexplained stubborn behavior in command execution
- pretty much every single archive that this thread delivers is incorrect and it …
- [Bug] Model performance regression: Unexpected degradation in output quality
- [Bug] Security alerts triggered during development with protected application
- false-positive na "cybersecurity" promt to siema masz...
- [Bug] Claude Code not following user instructions and withholding information
- [GitHub integration]
- [GitHub integration]
- [GitHub integration]
- [Bug] @ mention picker not functioning correctly
- [GitHub integration]
- [Bug Report] Invalid Input: Unable to Process Non-Bug Report
- [GitHub integration]
- nao está a responder o projeto... por mais que tente..
- I fucking hate this update, i want to go back how it was. get me off of this sh…
- [GitHub integration]
- Response Information Keeps Deleting
- [GitHub integration]
- clode [GitHub integration]
- [Bug] Anthropic API Error: Reasoning extraction safeguard false positive on consecutive tool calls
- Agent discovery: initial scan silently skips some project .claude/agents/*.md; the same files' names absent all session while a newly created agent hot-loads fine
- /clear does not cancel a slash command queued behind it, and gives no warning
- GitHub app install flow is a dead end for organization repos - the working route (github.com/apps/claude) is never mentioned
- Feature request: read-aloud (TTS) for mobile Remote Control responses + screen-reader audit of the session view
- Advisor (Fable 5) returns "unavailable" whenever the transcript contains any prior tool call — one Bash(ls) suffices; Opus advisor immune
- Cannot copy long assistant output from the terminal (works fine in other CLI tools, e.g. Google Antigravity CLI)
- [Locked out — existing account routed to “Let’s create your account”, request human escalation
- [FEATURE] Detach the Browser pane into its own window (multi-monitor)
- Remote Control: a bridged session gives no indication of which machine is executing it — side effects land on an invisible host
- Idle notification gives no signal that a teammate's finished output still needs to be explicitly requested
- Mouse wheel scroll (and arrow keys) trigger prompt-history recall instead of scrolling transcript, in JetBrains (PhpStorm) integrated terminal
- [BUG] Empty server-managed settings (304 cached) zero out local managed-settings.json — deny/allow rules never enforced
- [BUG] Identity verification flow excludes South Korea and blocks Claude Code with API Error 400
- [BUG] Windows MSIX: opening the sandboxed Browser pane bricks a working install when CoworkVMService is half-registered after auto-update (0x80073D28)
- [Bug] Subagent status line JSON missing parent reference for nested subagent hierarchy
- [Bug] Subagent status line JSON missing effort field for non-custom subagents
- [Feature Request] Add token direction field to subagent status line JSON
- Session list: show status (running / waiting for input) and open PRs
- I LOVE CLAUDE CODE[Feature Request] Improve user feedback messaging for Claude Code CLI
- Desktop: focusing a session with many large subagent transcripts eagerly loads all of them, main process exceeds 7 GB and the app crash-loops (1.34493.1, Windows MSIX)
- Three wedged claude processes leak 92 GB and freeze macOS: startup failure path loops allocating instead of exiting (2.1.234)
- [BUG] Cowork local MCPB calls fail with -32003 without approval UI, even in Skip mode
- [FEATURE] Task-first sessions: infer the working directory instead of requiring it up front (Claude Desktop, local Code)
- Feature Request: 中文本地化支持 - Chinese Localization Support
- Tool-call blocks leak as plaintext, output repeats up to 184x, and a fabricated tool result was reasoned from — in very long sessions (Opus 5, Claude Code Windows desktop)
- [BUG] Ctrl+V image paste deterministically crashes CLI — Bun 1.4.0 SIGBUS at 0xBAD4007 on macOS 26.5 (any image, 3/3 repro)
- [BUG] ultrareview: CLI declares cloud review failed at 30 min while it is still running; free attempt consumed, findings never delivered to CLI
- [Bug][cyber] ClAudit false-positive while: “wire all of it together.... you did.. you got here to readin…” (req_011CcnFFi6C7cW65SFu9WLSb)
- [Bug][cyber] False positive block during memory scan/watchpoint debugging tool development and testing (req_011CcnFJZScjXWEQuM7jUVN4)
- [Bug][cyber] False-positive block on rooted-device SELinux/dev-mode setup for the user's own Android phone (req_011CcmzvbxgnDzowZjuRR3rJ)
- [Bug][cyber] Safety filter blocked routine analysis of RSA-SHA256 key material extracted from own memory dump (req_011CcnJ2yHjMz3YQrqvLefKo)
- [Bug][cyber] Safety block halted authorized reverse-engineering of my own drone's auth protocol, mid-secret-re (req_011CcnReWU19UF1bg7REcCRy)
- [Bug][cyber] Safety filter blocked continued work on a drone auth plugin (RSA key-based protocol handshake cod (req_011CcnWhgbceAtzYUrYPwG71)
- [Bug][cyber] Safety block halted routine memory-dump analysis of a mobile game process for reverse engineering (req_011CcnXGX2KkGqEv89H61eRm)
- [Bug][cyber] Safeguard blocked a routine "finish the task" continuation message with no security content (req_011CcnRnv4YvFFpaz45AjrWk)
- [FEATURE] Desktop app: show the GitHub issue as a footer chip next to the PR chip (statusLine and footerLinksRegexes are silently inert)
- [BUG] Remote Control banner active in only 1 of 3 open tabs
- [BUG] iTerm2: Cmd+Click on a TUI link opens two browser tabs — terminal and TUI each open it once (a split, not a duplicate click event)
- [BUG] Stray SGR mouse-report escape codes leak into input during fullscreen TUI mode
- [FEATURE] Roll up elapsed time and token spend to subtree totals in the Agent Panel
- Desktop: expose a persistent "keep sidebar open" setting; the pinned state (Ctrl+B) is undiscoverable and undocumented
- [BUG] .worktreeinclude patterns with leading `**/` silently match nothing during worktree copy
- IDE websocket silently proxied (and can outright fail) when lowercase no_proxy lacks loopback under NODE_USE_ENV_PROXY=1
- Session rename writes custom-title to ~/.claude, ignoring $CLAUDE_CONFIG_DIR
- `ask` permission rules are silently overridden by the auto-mode classifier — breaking change to a safety contract (~2.1.236–238)
- [BUG] [claude-generated] sandbox filesystem: allowRead re-bind of a directory is shadowed by a later --tmpfs on a parent path
- [Bug] Exit cleanup recursively deletes $HOME instead of session directory
- `/ultrareview` results become permanently unrecoverable once the task ID expires
- [Bug] False-positive safety pauses
- [Bug] Session interruptions and model switch prompts
- [Bug] False-positive safety pauses
- [Bug] Content filtering false positives
- Windows Application Always Stays On Top[BUG]
- Claude overuses the word "arm"
- [Bug][aup] ClAudit false-positive while: “can you look at the config of my groups now?…” (req_011Cd1bhCAiShXTWS4a1R6LS)
- [BUG] Theme setting missing from Settings → Claude Code → Appearance; code font from the same release is still present
- [Bug][cyber] False positive while analyzing local packet captures during device firmware update (req_011Ce1J4t6WgbqX3FKCWz9Ta)
- [Bug][cyber] False positive while inspecting captured device traffic for unencrypted HTTP flows (req_011Ce1L8LvNSkpJcCLt6VFay)
- [Bug][cyber] Local router DNS troubleshooting and firmware update packet inspection (req_011Ce1J6s72zjygpgBt5W5Jv)
- [Bug][cyber] Blocked during Android device inspection and rootability assessment (req_011Ce1mvbe6Ktr2PpNQBHbjy)
- [Bug] Invalid or malformed bug report submitted