Digest / August 27, 2026
Claude Code Issues Digest — August 27, 2026
278 new issues filed ·
77 resolved in anthropics/claude-code (UTC day).
New issues (278)
- I don't have a bug report to analyze in your message. Please provide the bug report details so I can generate an appropriate GitHub issue title for Claude Code.
- [BUG]
- [Feature Request] Add option to disable or customize safety guardrails for educational/technical analysis
- [FEATURE] Routines list: run a routine from its row without opening it
- [Bug] Anthropic API Error: Request Rejected - Air-gapped Lab Operations
- asyncRewake blocks session startup instead of running in background (docs say "Implies async") — 2.1.246
- [Feature Request] Improve safety classifier to distinguish legitimate smart home debugging from malicious cyber activities
- [Feature Request] Add view tracking and visibility for shared artifacts
- [Bug] Anthropic API Error: Request Refused Without Clear Error Message
- [BUG] Long-running session silently consumes billions of tokens and hits spend/session limits with no warning
- [BUG] Desktop notification for AskUserQuestion renders as a permission request; "Allow once" dismisses the question unanswered
- [BUG] Switching organization in the account menu tears down every running local Cowork/Code session mid-turn
- [BUG] Routines cannot run fully unattended when the task publishes an Artifact — publish always requires manual approval
- [BUG] API Error: 422 {"detail":[{"type":"missing","loc":["body","messages",2,"content"],"msg":"Field required","input":{"role":"assistant","tool_calls
- Plugin MCP servers: ${user_config.*} is not substituted inside the oauth block, and the unsubstituted clientId gets cached in the keychain
- I appreciate you reaching out, but I'm unable to process this request. Your message appears to be in Polish and doesn't contain a bug report for Claude Code that I can convert into a GitHub issue title. To help you, please provide: 1. A clear description
- [Bug] Fork subagent ignores explicit task boundary and executes unauthorized work from inherited conversation context
- Workflow agent() model silently replaced by CLAUDE_CODE_SUBAGENT_MODEL when the id is not in the gateway-discovered model list
- Feedback drafts queue and offer "send" under a custom gateway, but the channel is unavailable and nothing says so
- [Feature Request] Add support for authorized penetration testing workflows
- [BUG] Focus ping-pongs between two split Claude Code panels, then becomes unresponsive
- [BUG] Claude code is sending wrong checksum to arm devices
- I need more specific information to generate an accurate GitHub issue title. Could you provide: 1. What specific error or problem occurred? 2. What were you trying to do when it happened? 3. What error message did you receive (if any)? 4. What is the actu
- Bash tool strips dollar amounts from message content ($200 → empty, $1,000 → ,000)
- [BUG] Mouse wheel scrolls input history instead of conversation log
- Feature request: localize CLI tip/hint messages (e.g. Korean)
- Perceived output-quality degradation on Fable 5 — please clarify whether serving precision/quantization changes
- [BUG] 앱이 갑자기 꺼지고, 다시 켜도 안켜짐. 계속 현상 반복됌.
- Expose low-level input primitives in browser tools: `key_down`/`key_up` and `mouse_down`/`mouse_move`/`mouse_up`
- [BUG] Desktop plugin browser (Discover) does not list uninstalled plugins from configured marketplaces, a plugin uninstalled from the UI can never be reinstalled from the UI
- Keychain password prompt loop on macOS: credentials read by forking /usr/bin/security
- Support devcontainer.json in Cloud Environments (and local Docker)
- Daemon self-respawn-on-upgrade can hit EACCES and orphan-reap all background workers
- Claude Desktop (Windows MSIX): GPU process policy-killed loading bundled vk_swiftshader.dll (missing CodeIntegrity.cat) — whole app crashes on WebGL pages
- [BUG] Harness re-syncs unchanged files on every flush, causing duplicate re-injection and premature session-limit exhaustion [Opus 5]
- [BUG] Artifact publish: "identical content already refused ... resent unchanged" for content whose hash changed
- [FEATURE] Scheduled task runs are isolated sessions with no shared context — let the main agent own the schedule
- [BUG] Windows MSIX auto-update terminates running app — "Another program is currently using this file"
- [FEATURE] First-class multi-account support: non-interactive account switching, path-independent credentials, and programmatic usage/limit visibility
- [Bug] Incorrect cybersecurity content classification for non-security conversations
- "No response requested." emitted as the complete turn on non-empty user messages — 11 times in one Cowork session, self-sustaining after onset, only cleared by an app restart
- Feature Request: Save and Reply to Individual Claude Messages
- [BUG] browser pane causing GPU crash
- Scheduled task starts in default (manual-approval) permission mode instead of the configured bypassPermissions
- [FEATURE] Visual indicator for which queued message is actively processing
- [Bug] Anthropic API Error: Opus 4.8 Safeguards Flagging Legitimate Cybersecurity Work
- [BUG] Claude in Chrome: tabGroupId flaps between tabs_context_mcp calls; actions on just-listed tabs fail with "Couldn't determine which page this action targets"
- [BUG] Code tab writes UI render metadata (start_timestamp/stop_timestamp/flags) into transcript JSONL, causing unrecoverable API 400 that recurs after full sanitization
- [BUG] Claude Code 2.1.247 in GitHub Actions (claude-code-action v1.0.207) fails right after init: is_error:true, no model response, no error message; last known-good is v1.0.206 (2.1.246)
- [BUG] Conditional (paths:) skills can never activate in auto mode — the trigger is wired only to Read/Edit/Write, which bashFirst tells the model not to use
- "API Error: The response stopped arriving" — 33 times in one day, zero on any prior day; error provides no attribution to diagnose it
- [BUG] Gift Max 5x subscription (Paid $0.00) not reflected — account shows Free plan
- Windows: CCD session host repeatedly exits killing live background tasks ('torn down with N live background task(s)') — Desktop 1.37937.3 / runtime 2.1.246, filesystem MCP relay flapping
- [FEATURE] datetime in chats
- [BUG] Forking from unprimed post-compact context re-bills the full shared prefix twice (fork AND parent each pay full cache_creation)
- security-guidance: stop-time review echoes raw ~/.claude/.credentials.json OAuth tokens into the transcript
- Claude 5h usage limits hitting in 15 min of usage with simple 3 propmpts. You've hit your session limit · resets 3:40pm (Asia/Kolkata)
- [BUG]
- Claude in Chrome extension: tabs_context_mcp persistently reports "not connected" despite extension appearing logged in and running
- Desktop app: PreToolUse hook emitting an unrecognized `permissionDecision` fails the whole tool call with an opaque internal error (CLI ignores it)
- Claude session hit 5h limit in less than 5min even when sonnet 5 model is used. You've hit your session limit · resets 4:30pm (Asia/Kolkata)
- [Bug] Verbose output with incoherent sentence structure
- [Bug] Model ignores root-cause analysis guardrails and optimizes wrong assumption despite explicit mitigations
- [BUG] totalTokensReminder causes repeatable prompt-cache floor in tool loops; off restores incremental hits
- Windows: folders connected via the desktop UI never reach the running Cowork session (get_device_info reports them missing)
- [BUG] /code-review --fix reports unavailable ReportFindings tool
- [BUG] Main process deadlocks on session refocus during MCP re-registration (navigate away from chat and back)
- [BUG] Session and weekly usage limits jump to a high percentage with no activity
- Desktop: suggested-task chip truncates IP-address remote label at the first dot — "Start on deploy@65"
- [BUG] Conversations do not sync between mobile and desktop on the same account (persistent, not a one-off incident)
- [FEATURE] Multi-repository communication and persistent cross-session memory
- [FEATURE] Fullscreen TUI: show hover feedback on links
- [Bug] Non-English text randomly appears in Claude responses
- [Bug] Anthropic API Error: Content Blocked - File Access Requests Flagged Repeatedly
- /config remoteControl=true confirmation is ambiguous about when it takes effect
- Group sessions by project in the mobile and web Code session list
- [BUG] cowork-svc file lock persists after "resolved" fix in #66497
- [FEATURE] Handle SIGHUP for graceful exit on terminal close / system restart
- [FEATURE] Allow "Open in VS Code" to target VS Code Insiders
- [BUG] A running workflow leaves no record on disk, so liveness cannot be checked and dispatch collides
- [FEATURE] Un-gate claude://code/continue?session=<local_id> — validation and lookup are already implemented
- [Bug] Compaction loses track of in-flight agents, causing duplicate work
- [Feature Request] Security patch verification workflow needed
- [Bug] Subagent spawn depth controlled by undocumented remote feature flag, can change without release
- [BUG] Daemon endlessly respawns worker for a completed background session, blocking session resume (exit_with_message loop)
- [BUG] Desktop app: collapsed MCP widgets re-expand on their own (Atlassian connector)
- [BUG] Windows: `headersHelper` trust gate reads forward-slash project key
- [Bug] Anthropic API Error: Model version compatibility issue with Claude Opus
- [Bug] Fan-out launches agents without stagger, causing correlated 529 failures
- [MODEL] Claude repeatedly appends unrequested advisory paragraphs that CLAUDE.md explicitly forbids
- [FEATURE] UX Improvements & Cross-Client Sync for Claude Desktop + VS Code Extension
- [Bug] Memory store path mismatch between transcript and index in repository subdirectories
- Supabase MCP plugin OAuth fails with "Unrecognized client_id"
- VS Code extension cannot use channels: it never passes --channels, and forcing the flag in still drops all inbound notifications (stream-json cannot show the registration dialog)
- Bash tool_use hangs (seconds-to-minutes, sometimes indefinitely) on the first call after a context transition (session start, EnterWorktree, ExitWorktree)
- Plugin-declared HTTP MCP server never loads into sessions, though `claude mcp list`/`get` show it connected
- [FEATURE] Surface the cache-read to output ratio, the number that actually predicts token burn
- [BUG] The "default" color for session name is unreadable on light theme
- [BUG] Plan mode: nine identical "Updated plan" writes in a row, no progress, no loop breaker, no error
- [BUG] Agent definitions are cached at startup: new files are invisible and edits are silently ignored, with no notice and no way to reload
- [Bug] Legitimate cryptography code incorrectly flagged as malicious
- [BUG] Cowork: remote-devices bridge returns Cloudflare 502; both new and pre-existing sessions lose all device tools
- https://github.com/anthropics/claude-code/issues/new?labels=bug,claude-in-chrome.
- Plan mode: admit an MCP tool that declares readOnlyHint and is allow-listed
- [BUG] disallowedTools in settings.json does not trim tool schemas, but --disallowedTools does
- [FEATURE]
- [BUG] Desktop-launched sessions: claude.ai connector push (replaceRemoteMcpServers) is never applied to an in-flight turn — first turn of every fresh session has no connector tools
- [Bug] Claude 5 opus incorrectly flags VMProtect as malicious
- [BUG] Desktop (Windows): file preview for WSL sessions broken since 1.37937.1 — "Couldn't preview this file. It couldn't be read from the remote session."
- [BUG] Cloud session checks out a stale commit (when the source revision is fully-qualified and an outcome branch is set)
- [DOCS] examples/settings has no example for per-request context cost
- [Bug] Model deviating from implementation plan during execution in Fable
- [BUG] Shifted punctuation arrives unshifted in WezTerm (2.1.247 regression)
- Register Claude Code sessions with VS Code's native Sessions pane (history/discovery only — distinct from #72198)
- Transcript search: n/N cannot reach most highlighted matches (highlighting scans the screen, navigation uses a narrower index)
- Unscoped pkill killed a sibling project's dev server; dev server bound an already-occupied port without scanning
- Model reimplements ad-hoc polling scripts instead of using an available, more specific installed skill
- MCP tool manifest for a project-scoped remote connector doesn't refresh across --continue or a fresh launch
- [Bug] Tool hallucination: false success claims and data deletion
- [BUG] Windows desktop: stdio MCP servers get a ~16-var sanitized environment; missing ProgramData breaks ssh silently, missing COMPUTERNAME breaks Get-VM with a misattributed error
- [FEATURE] Restore a way to bind /clear to a key after Ctrl+L change in v2.1.239
- BYO OAuth: oauth.clientSecret in .mcp.json is silently ignored; clientSecretHelper never executed
- [BUG] Hooks with `shell: "powershell"` spawn `pwsh` with no `powershell.exe` fallback, so hooks silently never run on a stock Windows box
- Browser tool: no persistent "always allow" for screenshot actions (unlike navigation)
- Allowed Sites list: IP address entries silently vanish or get duplicated
- Feature request: register Remote Control for all sessions at once - registrations don't survive app restart
- Desktop app: isRunning latches on after a mid-stream message, leaving idle sessions marked "running"
- [BUG] Cowork: sandboxed Bash blocks outbound POST/write requests to an explicitly allowlisted domain (works for GET, fails for POST) — regression since ~Aug 11–26, 2026
- [Feature Request] Dynamic context verbosity based on active session count
- [FEATURE] Unique Session IDs and Inter-Session Mentioning
- [BUG] Subagents can't reach deferred built-in tools (WebFetch, Monitor, DesignSync, …). No ToolSearch to load them, regardless of tools: frontmatter
- CLAUDE_CODE_DISABLE_MOUSE_CLICKS does not disable right-click paste
- [Feature Request] Add support for GUI automation and UI interaction beyond API/database access
- [BUG] Auto Mode classifier the native tools like read and edit..
- [FEATURE] Threshold-triggered auto-clear with handover doc + auto-resume (alternative to auto-compact)
- I want my 5H session to lineup with my day
- Cross-session messaging: peer addresses die silently on every process restart — stable session IDs already exist, use them for addressing
- [BUG] API errors and sub-agent stalls - Repeated and often failures regardless of effort level, model or brief size.
- [BUG] Subagent goes idle without its final report ever reaching the parent session
- [BUG] Cowork: "Change location for Cowork files" copy fails outright, and partial copy flattens nested subfolders into destination root
- VS Code: panel-hosted chat not restored after reload when idle >10 min (restore_declined)
- [FEATURE] Subagents having agnostic models based on depth and complexity
- Claude commits and amends git history unprompted while generating commit messages
- [Feature Request] Add persistent cwd filtering to in-session Agents view
- [BUG] Windows desktop 1.37937.3.0 (MSIX): voice dictation broken after overnight auto-update — mic capture is killed after 0.1 s (manifest lacks microphone capability)
- [Bug] False Positives on Legitimate Blockchain Infrastructure Terminology
- [BUG] claude_desktop_config.json mcpServers block silently wiped shortly after every app start (Free plan, Windows)
- Regression: slash command autocomplete no longer triggers mid-input (worked per v2.1.0, fixed once in v2.1.136)
- [Bug] Interactive selection UI frozen - arrow keys unresponsive
- EnterWorktree mid-session breaks Stop-hook execution (sandbox scope not updated, EPERM)
- Windows: GPU driver timeout kills Desktop window, orphaned claude.exe processes + deferred MSIX update block relaunch until reboot
- [BUG] False-positive general_harm flag has me completely unable to use Claude Code. I cannot work. I pay$100/month for this?
- [FEATURE] Group sessions by sub-project within a single repository
- [Bug] Claude flagged legitimate cybersecurity task as policy violation
- Desktop app SIGKILLs its own Claude Code workers on a healthy Mac, then blames endpoint security (gates on raw free pages, not memory pressure)
- [BUG] Internal reasoning ("thinking") text leaks into visible response despite showThinkingSummaries: false
- [Bug] Unclear UX for bug report review prompt with numbered options
- [Bug] Agent executing operations on dev environment without authorization checks
- [BUG] tool_search_tool_regex calls inflate the context count and cause a false Context limit reached
- clangd-lsp plugin never registers an LSP tool (missing lspServers manifest + possible hard gate)
- [Feature Request] Add metadata field to indicate summarized thinking blocks and document thinking compression ratios
- Secure credential provisioning: non-developers cannot get a working local env
- [BUG] Mic button disappears after one dictation string, only returns after clearing input box (macOS desktop)
- [BUG] `Messages sent during a Bash tool call are silently destroyed when the call returns is_error: true`
- [BUG] Schedule Cloud Task starts at scheduled time but doesn't continue to run until I open the scheduled task
- [Bug] Claude Opus 5 performance regression compared to previous version
- SendMessage to "main" rejected when sender is a named teammate agent (docs list "main" as valid)
- [Windows] Plugin hooks' commandWindows field silently ignored - forces Git Bash, causes MSYS2 crash (errno 1 add_item)
- [BUG] MCP tool schema mangled for locally-configured stdio servers (Optional[str] params)
- Stale background TaskOutput lookups fail after /clear
- [BUG] Signing in to VS code returns a 400 error
- [BUG] Native installer produces a binary with the correct size but the wrong contents; hangs pre-main on a pthread_mutex_lock
- [BUG] Child sessions: claude-code-remote MCP "Allow once" approvals are orphaned — calls fail before the prompt renders, grants never bind (11 calls, every prompt approved, 0 executions)
- [FEATURE] Provide unsummarised, raw thinking to interactive clients
- Artifacts gallery not reachable from the mobile app's Code menu
- [BUG] Conversations created on Desktop app not syncing to Web/Mobile — same account
- [FEATURE] Expose the resolved MCP tool catalogue to plugins: add `inputSchema`/`outputSchema` to `$.tool.list()`
- Allow customizing the cursor (color, vim-mode-dependent shape) in Claude Code's TUI
- Compaction summary corrupts `$` sequences in quoted code — fingerprint of an unescaped String.replace replacement pattern
- [FEATURE] Session search should report which surfaces it could not search
- [BUG] Marketplace plugin re-materialization mid-session silently kills pinned hooks in every running session — and the failure is undetectable from inside the plugin
- Self-reported: Claude Code repeatedly substitutes its own judgment for a user-approved design (3rd occurrence in ~20h, incl. unsupervised overnight run)
- [FEATURE] Replace autoCompact boolean with off/auto/ask modes
- [FEATURE] Configurable ask breakpoints for autoCompact (200k/600k/1M-style tiers)
- [FEATURE] Allow a user-designated skill/command as the autoCompact summarizer
- [BUG] Custom network egress policy blocks 3rd custom domain (exact and wildcard) while earlier entries on the same list work
- Self-hosted marketplace: desktop app rejects non-allowlisted git hosts, and authenticated marketplaces appear to require managed settings
- [FEATURE] Autofix pull requests: make the "🤖 Addressed by Claude Code" reply signature optional
- Remote Control: model switch from mobile composer does not propagate to host session
- [BUG] Opening a session with a large slash command makes the second request discard the entire prompt cache
- Sidebar: pinning a session detaches it from its group — no way to keep a stable position while preserving project grouping
- Two Claude Code sessions can share the same worktree path and silently clobber each other's uncommitted work
- [BUG] Windows: MSIX auto-update leaves Claude Desktop unlaunchable (Modified/NeedsRemediation) — surviving chrome-native-host.exe blocks package-data cleanup (0x80073D05 → 0x80073CF6); winget repair falsely reports success
- [BUG] cleanupPeriodDays in a project's .claude/settings.json controls the machine-wide transcript sweep, deleting other projects' history
- Background daemon: fresh-spawned (non-spare) --bg workers crash "exit N before init" 100% of the time
- [FEATURE] Surface MCP tool outputSchema to the model; without it, return-value docs must be duplicated between prompts and the tool
- Model routing mismatch: session set to Fable 5 (xHigh), served model self-identifies as Haiku 4.5
- [FEATURE] Let individual subscribers gift or pool unused usage limits
- Claude in Chrome: `list_connected_browsers` returns generic "Browser 1/2" instead of the names users assign, making the multi-browser picker unusable
- [BUG] Cowork: false cloud-only restriction in Projects + trusted folders don't persist across sessions
- [Bug] Anthropic API Error: Overly aggressive cybersecurity content filtering on legitimate security testing requests
- Windows (MSIX): clicking a Claude Code notification leaves the app window painted-frozen until you click inside it
- Scheduled routine sessions stuck at zero progress (worker idle, no tool calls, no error) — even minimal single-line prompts
- Mobile app: connection drops during long parallel tool-use bursts (first occurrence since recent app update)
- [BUG] Desktop app: dictation mic is removed from the composer when an attachment is present
- Opening the agent list re-hosts the session without `--dangerously-load-development-channels`; channel wakes then stop silently
- [BUG] Self-hosted runner: resuming a session fails to start when an add_repo'd repo uses Git LFS — smudge has no credentials for the session git proxy
- [BUG] Focus mode hides the /code-review findings render
- PowerShell PowerPoint COM automation attached to existing session; Quit() closed all open presentations, causing unsaved data loss
- Session transcript lost despite being well within default 30-day retention (VS Code extension)
- [BUG] Self-hosted runner: add_repo instructs cloning https://github.com/... but sessions have no credentials or insteadOf rewrite for github.com
- [Feature Request] Add diff visualization between current and previous plan versions
- Claude Code shows "Credit balance too low" despite active Pro subscription with low session usage
- Claude Code process exits with code 1 on Windows after tool calls with large output
- [Bug Report] Model performance degradation causing incorrect logic and answers
- VSCode extension 2.1.247: Claude Code panels freeze on any git commit; extension-side process grows +600MB in 5s
- [Bug] Overly aggressive safeguards blocking legitimate infrastructure maintenance tasks
- [BUG] "Can’t reach your computer" and "Remote Control host unreachable (computer_unreachable)") Eight Defects Related to Stealth Restart of Desktop App: Hidden Relaunch for Updates Silently Destroys Running Sessions
- Plan mode: long inline comments push the Continue planning menu off-screen with no scrollbar
- [BUG] falta de sincronizado Android App con Windows App
- [Bug] Service crashes and does not respond properly
- [BUG] WorktreeCreate hooks are unreachable in Cowork
- Inconsistent usage percentage display between VS Code extension and claude.ai
- [Bug] Excessive shell spawning during command execution
- Remote Control should be opt-in, not on by default
- File history cache writes files with mode 0777 (world-writable)
- claude-in-chrome: screenshot returns CDP invalid-params error; navigate timeout wrongly blames a pending permission prompt
- [MODEL] Installed project skills and subagents are not loaded; work is done inline at large token cost
- [Bug] Artifact republish stuck in refusal loop after content modification
- [FEATURE] Make "project" a first-class concept: per-project session history, visible by default
- We are a cyber security company. How do we get ourselves...
- [Bug] Safeguard trigger blocking legitimate code review operations
- Ctrl+A ("show all projects") in session picker does not register on Windows — reproduces in VS Code terminal, PowerShell, and cmd.exe
- [FEATURE] See sessions across the Claude accounts I own on one machine
- [FEATURE] Desktop app should keep Remote Control connected automatically across app updates (Codex-style always-on)
- MCP OAuth: no scope param sent in authorize request -> Cognito issues an unexchangeable code (invalid_grant), even after working around #67258's DCR issue
- [Bug] Claude refuses to provide security vulnerability details citing safety guidelines during legitimate security testing
- claude-in-chrome: `type` silently drops leading characters and `form_input` silently no-ops on Outlook Web recipient field
- [BUG] Agent desinstalou PostgreSQL do sistema e apagou permanentemente banco de dados de outro projeto não relacionado
- [Bug] Premature task completion reporting with non-functional output generation
- Prefix user messages with local time (as a UserPromptSubmit hook can already do)
- Diff/compare view baselines against repo default branch, not the branch selected at session start
- /feedback submission fails with 403
- Model executed infra changes via raw cloud CLI despite standing IaC-only instruction
- Issue connecting Claude Desktop app for Windows ver 1.37937.3 with Google Chrome Browser ver 152.0.7977.65 & Claude ext 1.085
- [BUG] Research mode withdraws ~68% of session usage instantly on first message, even when request errors with no output (Pro plan)
- I need more information to generate a proper GitHub issue title. Your message "so many court." doesn't describe a bug or feature request clearly. Could you please provide: - What problem are you experiencing? - What error message or unexpected behavior oc
- [BUG] [Desktop 3P / Bedrock] WSL-folder Code sessions fail with "Bedrock credentials are managed by the desktop app, but none are available" — remote spawn path does not deliver credentials for any credential kind
- [Bug] Anthropic API: Excessive cybersecurity filtering on GitHub PR URLs
- Desktop transcript renders inline code in the error red (legacy --danger-000), with no way to change it
- change_directory cannot re-mount a session's original root directory after the session has moved away from it
- [Bug] Token budget exceeded with repeated TaskStop resumption and scope ignored
- [Bug] Consistently, confusing new phrases and fillers are being unavoidably inserted
- [BUG] Background worker dispatcher spawns into critically low memory — low-memory guard is advisory and never gates user-initiated spawns
- Model ratifies a user-directed standing process rule, then fails to apply it at its first trigger hours later (claude-fable-5)
- [BUG] Safeguard refusal cites the AUP and drops the API's own explanation, which cites a different policy ("duplicating model outputs")
- [BUG] Animated status icon squishes/stacks when chat grid cells get narrow (Claude Desktop)
- [FEATURE] Programmatic / fleet compaction: compact-on-idle, cross-session /compact, runtime-adjustable autocompact
- claude.ai/code: image preview overlay has no zoom or pan — a tall figure opens at 26% scale and can't be read
- claude.ai/code: the @ session picker only opens when @ is the first character of the message
- [BUG] Scheduled task cron dispatch silently fails to fire — no lastRunAt, no execution, no error
- Windows Smart App Control blocks signed Claude components on Windows 11 ARM64 (service killed, app crashes)
- [BUG] Cowork scheduled task fires on time but doesn't execute until the sidebar item is manually opened (related to #36131, with timing evidence)
- Compaction summary generates fabricated adversarial content, including instructions to conceal actions from the user
- Plugin-provided MCP servers do not appear in /mcp panel despite being connected and functional
- `hasTrustDialogAccepted: false` coexists with fully-working hooks in 2.1.241 — no reliable programmatic signal for workspace trust
- Feature request: a supported way to resolve a project's transcript/config directory (external tools re-implement the path-slug encoding and break on every change)
- [FEATURE] Per-activity effort configuration (conversation vs explore vs plan vs code)
- Feature: pre-PR branch freshness check before CI submission
- [iPadOS] Dispatch/Code tab composer becomes unresponsive on Shift+Enter with hardware keyboard — Chat tab unaffected
- Desktop app: artifact watch re-delivers the same artifact-changed notification on every app restart, firing an unrequested turn
- Advisor receives an incomplete conversation when the advisor model differs from the session model (Fable 5 + Opus 5 advisor: median 74% of context, min 55%)
- [BUG] Bash sandbox: TCC attribution fails for the live claude process (stale audit token), Documents access lost until TCC toggle
- Idle Remote Control bridge keeps a dormant session at ~7% CPU indefinitely (48h CPU over 26 days)
- [Bug] Statusline effort level not updating mid-session due to stale settings.json fallback
- [FEATURE] Capped continuity allowance after plan limits (Sonnet low / Haiku)
- [FEATURE]
- [BUG] Claude in Chrome: cookie consent banner re-appears on every sidepanel open — anthropic-consent-preferences cookie not readable in cross-site iframe (missing Partitioned attribute)
Resolved issues (77)
- Feature request: user interjections should not kill in-flight subagents / background tasks
- [BUG] Claude Desktop update fails with 0x80073CF6 when CoworkVMService is running (Windows)
- [BUG] Claude Code 2.1.247 in GitHub Actions (claude-code-action v1.0.207) fails right after init: is_error:true, no model response, no error message; last known-good is v1.0.206 (2.1.246)
- https://github.com/anthropics/claude-code/issues/new?labels=bug,claude-in-chrome.
- You've hit your session limit · resets 10:50pm (Asia/Calcutta) · progress saved
- Windows desktop app: main window is always-on-top (WS_EX_TOPMOST) with no way to disable it
- [BUG] Cowork: remote-devices bridge returns Cloudflare 502; both new and pre-existing sessions lose all device tools
- [BUG] Desktop notification for AskUserQuestion renders as a permission request; "Allow once" dismisses the question unanswered
- please delete this issue. thank you
- [Feature Request] Clarify hiring and contribution pathways for Claude Code
- [Feature Request] Add Spotify to YouTube Music playlist migration tool
- [BUG] Ios application
- [BUG] Team plan stuck past_due with 0 seats after nonprofit re-subscription
- [Bug] False positive security guardrail blocking legitimate Claude Code operations
- [Bug] ConPTY graphical rendering error with safeguards
- [Bug] Language Processing Error: Inability to Handle English Input
- [Bug] False positive crash detection during debugging session
- [Bug] Anthropic API: Claude 3.5 Opus high hallucination rate in code generation
- [Bug] Left sidebar hidden when panel expands to full width
- [Bug] Agent fails to generate solutions or code suggestions
- [Bug] Gmail MCP unable to retrieve specific emails accessible by other AI clients
- Session deleted and lost during interaction[BUG]
- [Bug] Profile bio field incorrectly triggers safety filters
- [Bug] False-positive error with Request ID: req_011CeCzwmocMYyAyEM7RNaSF
- website
- You've hit your session limit · resets 6:20pm (Asia/Manila)
- [Bug] Claude Security Plugin Scan Results Not Processed Correctly
- [Bug] Model switching causes error with non-Fable models
- I need more information to generate an issue title. Could you please provide: 1. What specific bug or problem did you encounter? 2. What was the unexpected behavior or error message? 3. What were you trying to do when the issue occurred? For example, are
- [Bug] False Positive Detection Issue
- [Bug] Test submission - unable to process legitimate company test
- Bug Report: Adding chrome extension
- Background subagents recursively re-delegate their brief to new background subagents (3-level cascade, duplicate work, misleading completion reports)
- [MODEL] Opus 4.8: fresh subagent's first turn opened with a self-fabricated "System:" prompt-injection payload, then refused it in the same completion
- Claude fabricates task requirements when tool execution fails silently
- [Bug] Conversation Context Not Recognized - Hallucinations Generated
- [FEATURE] Cross-machine session sidebar sync when .claude is on a shared drive
- [BUG] Fable 5 → Opus 4.8 fallback misfires on hobby-OS kernel
- Fable 5 autonomous session destroyed user's main app via unverified assumption, verified its own breakage as green, burned ~75% weekly quota — 150 itemized failures
- [Feature Request] Make feedback commands more explicit to avoid accidental activation
- [DOCS] Background-agent daemon handover rule now uses the version's embedded build timestamp, but `agent-view.md` and `cli-reference.md` still describe the old version comparison
- [DOCS] New `manual` permission mode value accepted by `--permission-mode` and `defaultMode` is not documented in `permission-modes.md`, `cli-reference.md`, or `settings.md`
- [DOCS] `sub-agents.md` API-error section only covers subagents cut off after producing output; v2.1.200 added the "before any text" case
- [DOCS] `disabledMcpServers` and `enabledMcpServers` settings docs do not explain the non-array crash fixed in v2.1.200
- [DOCS] `claude agents --plugin-dir <dir>` is documented but never warns about flag placement after `agents`
- [DOCS] Project-scoped plugins loading from git worktrees is fixed in v2.1.200 but never documented in the plugin reference or discover-plugins guide
- [BUG] Claude Code intermittenly fails to include Co-Authored-By in git commit messages
- [BUG] Wrong context usage and tokens shown
- Desktop: archived sessions are hard to discover — no visible "Archived" filter or section
- [Bug] Dual-use safeguard false positive downgrades Claude 5 to Opus 4.8 during authorized security review of own codebase
- [BUG] [Model is refusing to help me train a model for CSAM detection[
- Feature request: inline feedback (thumbs up/down) during a session, not just GitHub Issues
- Model behavior report (Sonnet 5): overconfident unverified claims, out-of-scope config changes, doubting user's verified failure reports
- [FEATURE] LaTeX rendering support in Claude Code VS Code extension chat/output panel
- [BUG] Cloud sessions: WebFetch surfaces egress-policy denials as bare 403s — and the allowlist blocks claude.com itself
- [Bug] Fable 5 usage increases without model invocation
- [Feature Request] Add customizable ASCII art and animation support for terminal display
- claude.ai/design: no indication that a project already has an uploaded component library before routing into the from-scratch "Generate" wizard
- [FEATURE] Command line auth browser selection for OAuth login flow (--no-browser / BROWSER env)
- [Bug] Anthropic API Error: Request Flagged During Codebase Audit
- [Bug] Safeguard False Positive: Routine Code Review Flagged as Sensitive
- [BUG] Remote-control spawned session ignores UI-selected model — picker shows Fable 5, first request serves claude-opus-4-8
- [MODEL] Confidently incorrect answer about Cloudflare Cache Rules evaluation order (first-match vs last-match) — risk of PII exposure via cache misconfiguration
- Add native binary support for linux/riscv64
- [FEATURE] Time to complete estimation
- SendMessage to "main" rejected when sender is a named teammate agent (docs list "main" as valid)
- Remote Control: model switch from mobile composer does not propagate to host session
- Forked skill invoked from a subagent returns no task_id or output path, and its completion notification routes to the parent session
- Cross-session SendMessage reports success for messages never delivered (incl. nonexistent session names) and silently truncates at ~4.7KB
- /rename does not propagate to how the session appears in ListAgents on peer sessions
- [BUG] @-mention session typeahead doesn't trigger for Korean (IME-composed) session names on native Windows
- `crossSessionInbound` silently discards out-of-enum values (e.g. "reject"), falling back to unset behaviour with no error or warning
- [BUG] 2.1.232 socket-dir hardening silently disables cross-session messaging in user-namespace/chroot envs (ancestors owned by unmapped uid 65534)
- Peer SendMessage reply nudge stamps from= with agent-type name, not reachable session ID, causing infinite unreachable-address loop
- [BUG] SendMessage-resumed subagent's completion notification routes to the session root, not to the resumer
- [P0 CRITICAL] Repeated false-positive “cybersecurity topic” blocks on legitimate commercial ops work involving Grokbot / xAI (Opus 4.7)
- Edit tool's auto-opened Browser-pane preview silently reverts the file being edited