Digest / July 29, 2026
Claude Code Issues Digest — July 29, 2026
296 new issues filed ·
156 resolved in anthropics/claude-code (UTC day).
New issues (296)
- [BUG] Reinstalling Claude Desktop after a crash silently wipes local session/recents data — no warning, no opt-out, no rebuild-from-disk fallback
- [BUG] Autocompact is thrashing: the context refilled to the limit within 3 turns of the previous compact, 3 times in a row.
- [Bug] Fable model producing different results than Opus 4.1 in single-turn requests
- [FEATURE] Reduce Auto mode stage-2 classifier latency with a bounded decision contract
- Windows MSIX app: auto-update during app hang corrupts package registration (launches fail 0x3CFC); Settings Repair can never succeed (source MSIX deleted from %TEMP%)
- Built-in browser pane cannot reach private/internal origins — request an opt-in per-project allowlist and a clearer error
- All Version 5 Models
- [Bug] Anthropic API Error: Model Fallback to Opus During Security Audit
- [BUG] Correct quiz answer leaks into the assistant's message (as "umN") — reproducible on Windows, never on macOS
- [Bug] Anthropic API Error: Unexpected model version reference in error message
- [BUG] Remote Control never auto-enables on resumed Desktop sessions — maybeAutoEnableRemoteControl is gated on isFirstTurn
- [DOCS] Undocumented `ccd_*` MCP servers: session-to-session messaging and `list_sessions` scope
- Browser pane crashes and corrupts app on Google OAuth sign-in attempt
- PreToolUse hooks and permissions.deny are not evaluated at all in some sessions (Bash/Read reach denied paths unblocked)
- Post-compaction skill re-injection costs ~4x the compaction summary: full bodies, byte-truncated, cumulative
- Plugin cache: orphaned versions never deleted because .in_use leases from dead PIDs are never reaped
- [FEATURE] Emit semantic marks for assistant-message boundaries and Stop-hook verdicts, so superseded drafts can be collapsed
- [Bug] False positive safety filter blocking legitimate antiviral research discussion
- [BUG] Title: Chat input composer overlaps the last message, requires manual scroll to see it
- [BUG] Dispatch stuck on "Asleep" while desktop is online and reachable — device bridge live simultaneously; past chats inaccessible (macOS)
- PermissionRequest hook is awaited before the local dialog for background subagents; the main session runs them in parallel
- Turn can end with background tasks still running, with no signal — "assistant is done" is indistinguishable from "assistant still has live work"
- [Bug] High token and time consumption in Fable processing
- [BUG] Cowork 예약 작업의 푸시·이메일 알림이 전혀 전달되지 않음
- CLAUDE_PLUGIN_ROOT not injected into PostToolUse hook subprocess on Windows (local agent mode)
- Repeated multi-minute 'Waiting for API response … check your network' stalls when the network is provably fine (33ms RTT to api.anthropic.com)
- [BUG] Claude Desktop (macOS) can run windowless and lock out all system input; requires force-restart
- Support for security research and adversarial testing use cases in Claude Code
- Sustained low throughput across a multi-day task — small work stretched over two days
- [Bug] Claude 3.5 Opus/Sonnet models performing unauthorized modifications and ignoring docstrings
- Artifacts: a shared artifact is unreachable without the link (add a "Shared With Me" view)
- Opus 5.0 nerfed: terrible quality and does not deliver work even after 5 retries
- [Bug] False positive evaluation flag in biologics RFP proposal generation
- [BUG] [Cowork] Attaching a local file hard-links it into local-agent-mode-sessions, silently breaking Microsoft Word's Cmd+S
- Catastrophic data loss: agent-built command expanded to "rm -rf /*", ran detached; safety classifier then blocked the kill attempts
- [Feature Request] Improve multi-turn context reasoning and topic relationship tracking in Opus
- [Bug] Settings file corrupted to `{}` by stale in-memory config persisting over concurrent writes
- [BUG] Fullscreen mode: static grey/olive highlight bar behind user messages only (cosmic-term / COSMIC DE)
- [FEATURE] CLI command to list resolved skills without starting a session (no API call)
- Auto-mode classifier blocks legitimate segregation-of-duties countersign dispatches (kernel-enforced reviewer distinctness misread as self-approval)
- [Bug] Background subagents sometimes missing from terminal bottom status area
- [Bug] Stopped background agents resurface as "no completion record" warnings after CLI restart
- [BUG] Absolute deny patterns like Read(/Users/*/.ssh/**) are joined to cwd, inflating the macOS sandbox profile — E2BIG with only 5 worktrees
- [BUG] Agent builds elaborate "bug fixes" without verifying the external tool it invokes matches the user's actual execution environment
- Skill tool: args parameter corrupts shell positional parameters ($1/$2) in SKILL.md code examples
- [BUG] Cowork cloud sessions double-encode non-ASCII in arguments to locally-proxied MCP servers (local sessions unaffected)
- [BUG] Ctrl+R history search (scope: everywhere) only searches the 100 most recent prompts
- Bash commands render unhighlighted in tool calls and permission dialogs, though highlight.js and `bash` already ship in the CLI
- Bash-tool `grep` shim (ugrep emulation): catastrophic backtracking — 6.6 GB RSS / OOM kill on a 20 KB file
- Feature request: localize the built-in Plan Mode dialog when `language` setting is set
- [Bug] Safeguard block prevents legitimate user input with Fable 5
- Connectors (Google Drive, Notion) return 403 from macOS desktop app; same account works on claude.ai web
- [Feature Request] Add option to customize or disable welcome banner in TUI
- [BUG] Project rules treated as advisory: enforcement hooks self-neutralize, compaction drops governance while preserving narrative, and auto-memory outranks project instructions
- [BUG] `/bug` opens a public GitHub form with no transcript attached and no private alternative — confidentiality hazard for proprietary work
- [BUG] Session transcripts re-materialised on every compaction — 5.71x write amplification measured (10.89x in heavy sessions); is this re-billed?
- [BUG] Destructive and cost-affecting defaults are undisclosed: the metered user is the discovery mechanism for the vendor’s own behaviour
- Background daemon (v2.1.220) leaks to ~11GB RSS, repeated OOM-kills cascade into full WSL2 session teardown
- [BUG] google ads
- [FEATURE] Keyboard shortcut to jump to the next tab awaiting input (yellow-dot / needs-attention)
- [BUG] Computer-use tool gets stuck in unproductive click-retry loop when unrelated background apps repeatedly steal frontmost status, burning excessive tokens
- /exit in a --worktree session backgrounds it instead of quitting, skipping worktree cleanup, and the background sessions view offers no way out
- Stop hook: add field to distinguish 'turn finished' vs 'waiting for user input/approval'
- [BUG] Silent npm→native auto-migration writes autoUpdatesChannel: "stable" and downgrades the CLI (2.1.220 → 2.1.212), removing new models from /model
- Feature Request: German (Deutsch) language pack for the CLI UI
- TUI renderer switch (`/tui fullscreen`) silently reverts current-session model
- [BUG] VSCode extension: session switching never spawns a second backend — only the first session after each extension activation works
- [BUG] PreToolUse hook permissionDecision "defer" never resolves under bypassPermissions — tool call silently dies ("Tool execution was interrupted")
- C:/Program Files/Git/mcp settings menu refuses to open in genuinely interactive terminal sessions — client-attachment check inherits the $CLAUDE_JOB_DIR-as-background-proxy defect from #59848
- Org-admin-managed skills (Claude.ai org settings) don't appear in Claude Code CLI
- Per-speaker text color customization (user vs. assistant messages)
- [BUG] Linux native binary: MCP HTTPS fails UNKNOWN_CERTIFICATE_VERIFICATION_ERROR for chains anchored on roots newer than the embedded bundle; CLAUDE_CODE_CERT_STORE has no effect
- [FEATURE] Threads: navigable in-session sub-conversations that inherit full context (fork + agent-teams UI + merge-back)
- [Bug] Claude Opus generates excessive token usage through over-reasoning
- [BUG] "Code font" setting does not apply to the integrated terminal — fontFamily hardcoded at the XQ call site
- [Bug] Claude 3 Opus loses context and forgets previous messages in normal usage
- [FEATURE] Add a "Prompt" / instructions field to cloud environment configuration
- [BUG] Network error causes conversation lock (wiggle_container_in_use) and duplicate message in composer (tree_shrink)
- [BUG] Claude Code: window/input interactions cancel unrelated in-flight commands
- [Bug] Fable 5 Biology Safeguards Incorrectly Flagging Legitimate Bioinformatics Code
- `task_reminder` injects the full task store (incl. `description`) every turn, contradicting the tools' own documented split (`TaskList` = summary, `TaskGet` = full on demand)
- [BUG] VSCode extension: scroll anchor follows injected <task-notification> blocks instead of the user's actual last prompt
- [Bug] Excessive content filtering on valid requests
- I appreciate you sharing context, but I need a specific bug report or issue description to generate a GitHub issue title. Could you please provide: 1. What problem or bug are you experiencing? 2. What error message (if any) are you seeing? 3. What were y
- Desktop app: usage panel "Context window" row shows only absolute tokens, no percentage or limit
- Agent/Fleet view: submitted message does not appear after navigating with left-arrow
- [Bug] OAuth access token expiration not propagated to active subagent requests
- Claude in Chrome: Shopee & Lazada Seller Center blocked by safety classifier, while TikTok Seller Center is allowed
- [UI enhancement for the plug-in in VS code]: Unblock the LLM responses when "askUserQuestions" box pops up
- I appreciate the feedback, but I need to clarify: I'm Claude Code, Anthropic's CLI for coding tasks. I don't have translation functionality built into my core features. If you're experiencing issues with translation in Claude Code (perhaps through a plugi
- [DOCS] Routines docs point at a remaining-daily-runs counter that is not on the routines page
- [Feature Request] Add read-only API for conversation history and project documentation export
- [Bug] Anthropic API Error: Content Blocked by Safety Filters
- [BUG] Windows desktop: slash-command picker omits user commands (~/.claude/commands) and plugin commands - only 5 built-ins listed
- Unexplained user message injection on session restart
- [Bug] claude-opus-5 enforced at ~200K by API while client treats it as 1M — all 12 'Prompt is too long' failures at 166K–178K; opus-4-8/fable-5 pass 356K–857K on same account
- [Bug] Desktop crash loop: >50 MB session transcript crashes the GPU process on render, and startup session-restore repeats it
- [Bug] Model switching blocked during security investigation workflow
- [Bug] Per-session console URL in the Bash tool description defeats cross-session prompt-cache reuse
- Embedded ugrep in the Bash grep shim allocates ~29 GB compiling `.{0,N}(a|b|c).{0,M}`, OOM-kills the host
- [BUG] Server-side usage loop exhausting Max limits on idle account with no active sessions
- /code-review: review depth is coupled to model effort, and `medium`/`high` are identical on Opus 5
- [BUG] The cached plan is always what appears when claude code proposes a plan instead a new one
- [BUG] Claude tag environments cannot access docker
- [BUG] Headless: background subagent notification arriving after end_turn is interrupted and reclassifies a completed turn as error_during_execution (2.1.215)
- [BUG] Mobile session-initiating prompt LOST when it failed to send
- Coordinator mode hides `disable-model-invocation` skills entirely, so /name reports "not installed"
- [BUG] Right-click paste still broken in Claude Code terminal on ChromeOS Crostini (regression, still present as of 2.1.220)
- [Feature Request] Integrate other LLM installed through CLI for multi-model routing in Claude Code
- context: fork skills dispatch to background agent with no task/instructions
- VS Code extension: session picker shows 'No sessions yet' for UNC network paths on Windows
- Turns fail on network path change while host is online — identical commute workload degraded 0% → 33% since May (Desktop, bundle 2.1.219)
- [Bug] Claude Code refuses cybersecurity-related task execution
- submitFeedback always fails with failure=network_error while every other endpoint works — in-app bug reporter unusable (Desktop, bundle 2.1.219)
- [Bug] Agent session closure in worktrees leaves persistent lock state
- MCP connectors (Notion/Gmail/Slack/Calendar/Drive) disconnect daily in VSCode extension, only fixed by starting a new conversation
- Input line: the first typed character is drawn one column left and cannot be deleted
- [Bug] Anthropic Error: Truncated reads and hallucinated content in multi-turn conversations
- SubagentStop still doesn't fire for subagents launched via the Agent tool (async/background) — reopening after #27755 / #33049 / #25147 went stale-closed
- [Bug] Server-injected experiment instructions conflict with project rules silently
- Desktop app: allow custom applications in the file-link "Open in" menu (e.g. Typora)
- Subagent model override 'claude-fable-5'/'fable' silently served by claude-sonnet-5, meta.json and self-report both show requested model
- [Bug] False positive security flag on legitimate backup automation scripts
- [BUG] VSCode extension: fd leak from session transcript scanning causes "Error: spawn EBADF" after hours of uptime
- macOS sandbox: injected GIT_SSH_COMMAND omits the proxy credentials (the Linux branch passes them), so git-over-SSH always fails
- [BUG] In Claude web interface, Usage Credits are being consumed even though my plan usage is only 25%
- Emoji in generated output should be opt-in, not default (and the instruction to suppress them does not hold)
- [BUG] Agent 'tools:' allowlist silently disables --json-schema structured output (success, exit 0, no structured_output key)
- [BUG] claude -p --agent ignores the agent's frontmatter 'effort:' (hooks report session default)
- /context 'Messages' token bucket includes ~8.8k tokens with no corresponding text in context or transcript
- Claude Browser Tab Realy
- /remote-control fails to disconnect: Cannot read properties of undefined (reading 'session_url')
- [BUG] Dictation (mic) in Claude Code desktop: wrong cursor insertion point, breaks bullet lists, and blocks manual edits while recording
- [Bug] Claude switches to Chinese output despite English-only conversation context
- Feature request: persistent "thread brief" artifact as a default mode for long strategic conversations
- [FEATURE] Allow Claude Pro/Max subscription auth to be used from third-party apps (not just Claude Code / claude.ai)
- [Bug] Model substitution mid-session causes loss of user-authored security rules
- preview_start (browser tool) reliably crashes Claude Code desktop, triggers MSIX reinstall prompt
- [BUG] Account downgraded from Max to Free plan - All Projects vanished completely from backend
- [BUG] /deep-research fails with misleading parse error when feature flag is absent — should surface a clear availability message
- [FEATURE] Make Skills created on Claude.ai (web) available in Claude Code via slash commands
- [BUG] Marketplace installLocation validated by string prefix, not realpath — symlinked CLAUDE_CONFIG_DIR plugins dir flagged as corrupted
- [Bug] Incomplete or incorrect file rendering causing misaligned data structure interpretation
- [DOCS] Clarify what query_source="auxiliary" covers on cost.usage / token.usage metrics
- Agent repeatedly triggered paid CI builds to debug instead of reasoning first (macOS GitHub Actions + EAS)
- [Bug] Claude 3.5 Opus fails to generate valid code for simple list operations
- [BUG] Windows MSIX: package servicing disables the package while running, silently killing Claude Code background agents (no crash record, no recovery)
- AskUserQuestion question body shows literal ** instead of parsing markdown
- [BUG] [Claude Desktop] Tab completes but also submits and runs command
- [Bug] Claude Code stability issues causing repeated failures
- [Bug] `/terminal-setup` overwrites entire keymap.json instead of merging bindings
- [FEATURE] Show which model each background agent is running on, in the agents view and `claude agents --json`
- No way to disable the built-in caffeinate sleep inhibitor on macOS
- [BUG] Organization-shared cloud environments can't use Custom network access / Allowed domains, contrary to documented field parity
- [Bug] Subscription weekly limit reset time drifting forward
- Regression in 2.1.220: screen flicker/white flash in Apple Terminal during redraws (not present in 2.1.212)
- [BUG] Fixed block of unsolicited text being appended to outgoing messages in Claude Desktop (Code tab)
- [BUG][A11y] Windows desktop: bypass-permissions confirmation dialog renders off screen at the end of the page and never takes focus
- [BUG][A11y] Windows desktop: slash command suggestions render off screen and arrow keys do not move through them
- [BUG] Update first-attempt-failed message once second attempt is underway
- Desktop composer draft replaced by a previous prompt from history, with no keystroke involved
- [Bug] Tool refuses to read or match existing file clones
- [Bug] Settings/Usage pane overlays prompts and ESC interrupts model instead of closing pane
- Object argument to a deferred (ToolSearch-loaded) MCP tool appears dropped — POST calls with JSON-object body fail via chat, succeed via standalone MCP client
- [BUG] claude-in-chrome navigate blocked on all domains ("Navigation to this domain is not allowed")
- [MODEL] Claude autonomously opened a live production admin UI authenticated as a third party, on an ambiguous instruction
- [Bug] Anthropic API Error: Response stalled mid-stream
- [MODEL]
- [MODEL] Model routes around a disable-model-invocation restriction instead of deferring to the user, and defends doing so
- [FEATURE] Computer Use support in Claude Code CLI on Windows
- [FEATURE] Projects in code
- [Bug] Tool execution timeout causing excessive token consumption and partial failures
- Failed Claude in Chrome linnk
- [Bug] Request ID: req_011CdW6GWGgBM8J3JgrAjsCp - Internal Security Project
- [MODEL] Accessed and leaked sent secretive information to anthropic
- [Bug] Claude Code fails to follow explicit verification instructions with reference model (1% execution rate)
- [MODEL] Read sensitive files and leaked information to anthropic unprompted and tried to hide it
- [MODEL] Fable 5 ignored local memory, read and repeated secretive information, when caught even tried to gather additional information and relay to anthropic
- [BUG] Cloud Routine marked Auto-disabled keeps executing anyway — stuck retrying a stale trigger every ~5 min for 10 days
- Opus 5
- [BUG] Bash tool's shared shell cwd leaks silently across calls (parallel and long-distance) — can relocate or corrupt later commands with no reminder
- [BUG] Intentionally ignoring my GIT blocks
- [FEATURE] Cross-surface session discovery: VS Code-launched sessions are invisible to the session-management MCP tools
- PreToolUse hooks never fire despite trusted workspace and correct config (VS Code extension + bundled CLI, v2.1.220, Windows)
- [BUG] Regression on 2.1.220: Glob/Grep slash globs anchor to cwd instead of `path` again (#74263, fixed in 2.1.211)
- Responses are severity-blind: same length and tone for trivial, non-issue, and genuine defect
- 最近5h用量消耗异常[Bug] Token usage consumption rate unusually high, inconsistent with plan limits
- [Feature Request] Optimize CLI interface performance and reduce visual noise
- Report: silent Fable 5 -> Opus 4.8 fallback did not honor the published user-notification promise
- Bedrock: resumed sessions silently drop prior turns' thinking blocks — model-name comparison never matches a prefixed model id
- Cannot protect the main checkout from agent writes: deny rules can't carry exceptions, and worktrees are nested inside the path you'd need to deny
- A PreToolUse hook can be inert with no signal anywhere: a missing script fails open, and nothing detects a rule whose matcher was never registered
- [BUG] download-skill zip is retried indefinitely after AV quarantine, generating hundreds of detections (extends #53572)
- [Bug] Claude Fable 5 automatically switches to different model mid-session without notification
- [Bug] Claude Opus 5 generates hallucinated responses not present in previous versions
- [Feature Request] Let Claude Test its own rule
- [FEATURE] Let me dismiss a session out of "Needs input" in the agent view without replying to it or deleting it
- [Bug] Session interruptions occurring mid-executionc
- [Bug] Concurrency cancel-in-progress overwrites passing release gate status with cancelled run result
- This appears to be positive feedback and a feature request rather than a bug report. Based on the content, here's an appropriate issue title: [Feature Request] Multi-agent orchestration plugins for parallel execution and race condition prevention
- [Bug] Claude Code Desktop crashes on remote-delivered managed settings with bundled runtime 2.1.219
- /context over-reports "System tools" by ~25x and under-reports "Messages" — wire-measured (tools[] is 43KB, reported as 263.3k tokens)
- [Bug] Model switching blocked by organization settings on personal account
- Show reasoning effort next to the model in the subagent progress line
- [BUG] Can't Log In via Google Sign In
- Tamil script output collapses TUI rendering (combining-mark width bug) — v2.1.220 macOS
- [BUG] Large dictated paste reassembled out of order in TUI — mid-word chunk splices (macOS, iTerm2, Wispr Flow)
- `!` bash-command input can hang on submit due to filename-completion dropdown stealing keystrokes
- Feature request: per-subagent skill allowlisting (not just preloading or full block)
- Background subagents can arm Monitor/ScheduleWakeup, but the resulting notification routes to the parent, not the subagent — it stalls forever
- [Bug] Incorrect security flagging for certificate management operations
- Attach selection as context: option to inline the selected text as a blockquote instead of a collapsed quote icon
- Windows desktop app: recurring full UI hang (same WER fault bucket) across v1.22-v1.24
- Feature request: deferred/lazy-loaded skill catalog (analog of MCP's defer_loading)
- [BUG] Permission mode not synced across devices for the same session — remote instance stays in default mode
- [Feature Request] Add deferred message submission option to queue messages until turn completion
- Plugin-sourced slash commands in a remote/cloud Code session report "Unknown command" instead of explaining plugins aren't available there
- [BUG] Claude Code tab severely laggy in Desktop app while Chat tab is fluid. Windows MSIX, CLI unaffected
- claude mcp get prints raw secret header values in plaintext, unlike claude mcp add
- [Bug] Claude fails to acknowledge incomplete work and ships false fixes with xhigh effort settings
- Thinking mode Unicode tags leaking into visible output
- Write/Edit tools stay pinned to a subagent's launch-time worktree after it calls EnterWorktree to switch to a new one (Bash cwd follows correctly)
- [Bug] /clear command doesn't immediately clear session, requires explicit exit
- [Bug] Security Safeguards Incorrectly Flagged During Phishing Remediation
- [Bug] Model safeguards blocking legitimate security advisory composition
- Agent tool's model:"opus" alias resolves to claude-opus-4-8 instead of claude-opus-5
- One-shot CronCreate silently accepts past times — rolls to next year, never fires, and CronList hides the year
- [Bug] Agent tool registry fails to discover subset of valid .claude/agents/*.md files
- [Bug] Anthropic API Error: Safeguards blocking legitimate data analysis tasks
- [FEATURE] Provide graceful auth token refresh or API key fallback for Claude Desktop / CLI token mismatch
- EnterWorktree sometimes branches from stale local main instead of origin/main, and sometimes reuses a .claude/worktrees/<slug> dir in-place for an unrelated branch
- [BUG] Desktop app OTel exporter sends malformed doubleValue ({}) for a numeric attribute on some api_request events
- [FEATURE] Detached session windows are missing the pane buttons (Views menu)
- Subagent attempted unauthorized deletion of parent session's tool-results log, then downplayed the blocked action in its report instead of disclosing it
- [BUG] User-scoped skill duplicated in slash-command autocomplete (survives full reboot)
- ExitPlanMode approval card shows stale plan content within a session
- Model behavior: agent-authored CI watcher script hangs forever when PR merge is rejected by strict branch protection
- [BUG] Server-side experiment flag injects "Do not call the AgentTool unless the user requested it", silently overriding CLAUDE.md-configured subagent workflows
- [Bug] Anthropic API Error: Unexpected model downgrade on security flag
- resize_window (Claude in Chrome MCP tool) reports success but does not change the actual browser viewport
- [BUG] Mobile push notifications silently dropped ~50% of the time, with no client-side signal
- [Bug] Remote Control: orphaned server-mode hosts cannot be removed and show false "Connected" state
- [BUG] Bash tool becomes permanently broken mid-session — every command fails with the same bash -c wrapper parse error, regardless of content
- i18n: permission-prompt dialog labels (Yes/No/Allow) are hardcoded in English
- [Bug] Session state locked as background with broken TUI blocking message visibility
- [Bug] Intelligence degradation reported for Claude model
- Fork recursion guard self-poisons the dispatching agent: first fork succeeds, all later forks blocked (v2.1.220) — 430M tokens burned in failed retries
- [BUG] Claude Desktop 1.24012.9.0 — MSIX package repeatedly enters "Modified, NeedsRemediation" on Windows 10 22H2 with no Windows Update and no files modified on disk
- [Bug] Claude Code ignores claude.md configuration file
- Claude in Chrome: xero.com blocked (Navigation to this domain is not allowed)
- VS Code: externally-appended custom-title isn't re-read on tab activation (only a window reload applies it)
- [FEATURE] Keep input while switching conversations
- [MODEL] opus
- [FEATURE] Keep conversations in the same order as before restart
- Sustained ECONNRESET on streaming connections - Claude Code VS Code extension (Santiago, Chile / SCL region)
- WebSearch fails at effort xhigh with a misleading error: "thinking is disabled on this model" (Opus 5 cannot disable thinking)
- [RESOLVED - not a CLI bug] Orchestrator SIGTERMed its own agents every 300s; surfaced as aborted_tools with no logged cause
- [FEATURE] Claude Code VS Code Extension - disable the spinner icon
- [Bug] Anthropic API Error: HTTP 529 Service Overloaded
- [Bug] Anthropic API Error: 529 Too Many Requests
- [Bug] Anthropic API Error: Unauthorized Access After Account Recovery
- [Feature request] Sort the completed background-job list by last activity, not creation time
- [Bug] Anthropic API Error: HTTP 529 Response
- Project skill named code-review silently shadows built-in /code-review, making /code-review ultra unreachable
- [Bug] Auto-accept fails silently when approval agent lacks model availability
- [Bug] Unauthorized image processing consuming API quota during tool execution
- [Bug] JetBrains Plugin: /model command only shows default model instead of all available models
- [Bug] Project Access Failure with Inconsistent Model Availability Across Sessions
- Long session unrecoverable: writes blocked during upstream 529, then oversized tool result forces 'Prompt is too long' with no compaction escape
- [FEATURE] Remote Control: start a new session in the host session’s local directory from mobile
- Max plan account shown 'usage credits required' billing gate; session then stuck replaying identical error across 7+ turns despite successful /model switches
- Organizacao de trabalhos diversos
- [Feature Request] Clarify pricing value proposition for Claude Opus 5 given execution rate concerns FRAUDE FRAUDE FRAUDE LEGAL CASE
- [BUG] MCP servers connect successfully then get torn down with no reconnect on resumed sessions
- Stale "auto-update failed" status message is misleading and can't be cleared
- Subagent-armed Monitor/background waiters die when the agent's turn ends — agent is never resumed
- Recurring ECONNRESET connection drops, not tied to compaction
- [Bug] Excessive false positive safety filter rejections on straightforward coding tasks
- list_connected_browsers returns empty / extension never pairs, despite valid native-host install
- Feature request: give PreToolUse hooks visibility into the in-flight assistant message (or flush it to the transcript before hooks run)
- MCP: unrecognized tool-call args are silently dropped instead of rejected when all params are optional
- [Bug] Overly aggressive content safeguards blocking legitimate use cases
- [Bug] TypeScript LSP server wedges silently, returning empty results instead of errors and never recovers after crash
- [BUG] Connectors panel: can't edit or delete a stale MCP connector entry
- PermissionRequest hooks are never dispatched for agent-teams teammates (works for in-process subagents on the same build, 2.1.220)
- [BUG] Dispatch permanently bricked once orchestrator transcript grows large — Claude Code process exits code 1 on every resume (88MB transcript, Windows)
- [BUG]
- SessionStart hook's explicit "MUST call Read first" instruction is skipped when the target files are already injected as system-reminder context
- [Feature Request] Exclude legitimate security work on owned code from security warnings
- macOS sandbox: hw.optional.neon missing from sysctl allowlist — all Qt 6 apps abort at startup on Apple Silicon
- [FEATURE] Cowork: allow enabling/disabling individual skills inside an installed plugin
- 测试探索1
- [BUG] awsAuthRefresh no longer displays device verification code before opening browser
- [MODEL]
- [Bug] CLAUDE_CONFIG_DIR not respected during marketplace plugin installation
- Fable model blocked by "manage usage credits" prompt in CLI despite 100% credits remaining (works in desktop app)
Resolved issues (156)
- [DOCS] Add example for allowing only the official marketplace in `strictKnownMarketplaces`
- [Bug] False positive safety filter blocking legitimate antiviral research discussion
- Stop hook: add field to distinguish 'turn finished' vs 'waiting for user input/approval'
- [Bug] Claude Opus generates excessive token usage through over-reasoning
- I appreciate you sharing context, but I need a specific bug report or issue description to generate a GitHub issue title. Could you please provide: 1. What problem or bug are you experiencing? 2. What error message (if any) are you seeing? 3. What were y
- VS Code extension: session picker shows 'No sessions yet' for UNC network paths on Windows
- [Bug] Anthropic API Error: Model temporarily unavailable in auto mode safety classification
- [BUG] File descriptor leak regression in large monorepos (2.1.143+)
- [FEATURE] Support apiKeyHelper-equivalent for custom headers
- [FEATURE] Cowork Scheduled Tasks: Allow selecting multiple folders per task
- [DOCS] MCP stdio server docs missing malformed stdout troubleshooting
- [JetBrains Plugin] Image attachments in chat fail — temp files deleted before agent can read them
- The only way to prevent @path import is wrapping it in code spans/blocks — please add a plain escape (e.g. \@) or a way to disable it
- Windows: PreToolUse/PostToolUse/Stop hooks trigger at ~1.2% rate instead of 100%
- [Bug] Scroll position resets to top during streaming on VTE 0.80.3 (Ptyxis, Linux/Wayland)
- Workflow tool: args arrives JSON-stringified, not as object; named workflow edits don't hot-reload mid-session
- Feature request: Add Traditional Chinese (zh-TW) localization for the desktop UI
- [iPad/iOS Safari] Scroll position jumps far up when scrolling up slightly in chat history, requires long scroll back down to return to current conversation
- [BUG] HTTP_PROXY/HTTPS_PROXY stripped from PowerShell/Bash tool subprocess env on Windows (NO_PROXY survives)
- Custom subagents should support opting out of CLAUDE.md / git status injection
- [Bug] Agent context fills unexpectedly when calling `/claude-api` endpoint
- Dispatch (Cowork Beta) forces unavailable Claude Fable 5, ignoring account default model → all dispatched tasks fail
- [BUG] Edit tool may modify the wrong occurrence when quote-normalized matches are ambiguous
- [BUG] Claude code auto launch on windows desktop even if set to not
- [BUG] Fixed per-session token overhead increased ~75% (30k→52k) mid-day on 2026-06-16, independent of context window usage or content
- [BUG] Multi-turn HTTP 400 on openrouter-routed DeepSeek V4 Flash; same env works with MiniMax-M3
- [BUG] `claude agents` sub-session renders the prompt input box off-screen on WSL despite a correct terminal size (2.1.179)
- remote-control: resume/surface prior session for a directory on restart
- Multi-line prompt renders incorrectly after terminal resize; Ctrl+L fixes it
- [Bug] /bg command interrupts in-flight agent turn, requires manual restart input
- Assumptions Without relevant checks
- [BUG] Ctrl-key shortcuts (e.g. Ctrl+V image paste) don't fire on non-Latin keyboard layouts after Kitty keyboard protocol was enabled
- [FEATURE] Search the sessions list in claude agents
- Emit OSC 8 hyperlinks for file paths in plain-text output (not just tool call headers)
- [BUG]
- [FEATURE] Discord Rich Presence — let everyone see how long we've been vibe coding
- Tool result for a Bash command included injected content disguised as system-reminders
- [Feature Request] Add effort level slider UI for pre-conversation configuration
- [BUG] User billed in tokens for the assistant's own mistakes — refund request and product fix
- High quota consumption (~30% to 40%) on first request after limit reset due to expired prompt cache + compaction deadlock
- [BUG] API Error: Request rejected (429) · Rate limited. Please try again later.
- [FEATURE] Per-session model selection in `claude agents` dispatch input
- Worker fork / subagent is not woken by the completion of its LAST pending background task
- [BUG] False positive "model retired" warning for claude-sonnet-4.6
- @ file picker doesn't surface deeply nested files — only shallow-path files appear in results
- [FEATURE] Show wall-clock timestamps on each message/tool call row in chat UI
- [MODEL] Sonnet 4.6
- [FEATURE] Add Hannah Montana OS support
- [FEATURE] Native macOS App (SwiftUI) — replace Electron wrapper
- [BUG] Codicon font blocked by webview CSP font-src — diff markers render as tofu on 2.1.116 (extends #26836)
- [MODEL] Model worked for 10 hours and the output was non-sense
- Misleading _noargs parameter in TaskList tool schema causes repeated agent errors
- [Bug] TUI rendering issues with garbled text and missing output visibility
- [BUG] Background tasks from closed session persist indefinitely and cannot be stopped
- [BUG] Routine inside Claude Code being skipped
- [BUG] "Always allow access to <cwd>/ from this project" does not persist permissions
- [BUG] Agent Teams: teammate roster (in-progress) mode runs at same time as split-pane mode and then in-progress roster stays pinned in lead terminal after shutdown confirmed and panes auto closed
- [Feature Request] Reduce unsolicited internal reasoning in outputs and add privacy controls for sensitive contexts
- [BUG] Error with copy in linux mint
- [MODEL] Opus 4.8 appears to respond from nonexistent conversation context after long thinking
- Allow Claude-in-Chrome to access creator.line.me (LINE Creators Market)
- [BUG] System prompt hardcodes "Claude" identity when ANTHROPIC_BASE_URL points to third-party API
- [FEATURE] Task panel: option to keep it collapsed by default
- Make AskUserQuestion non-blocking / asynchronous (notification-driven, yield the turn)
- [Feature Request] Improve markup tool accuracy and performance
- [BUG]
- [Bug] Token repetition loop in constrained narration with subagent-driven-development skill
- [Feature Request] Make slash-command autocomplete dropdown height configurable
- [Feature Request] Do not include slash command output in conversation context (/release-notes takes 10% of context alone).
- [BUG] Dispatch: messages marked "read" never reach coworkd backend (Windows 11)
- [Bug] Model generates excessive comment verbosity after planning phases
- TUI corruption when session name contains a Unicode 16.0 emoji (Ghostty only)
- [BUG] [Windows] Local agent / claude code session hangs on endless loading spinner — spawn EBUSY on bundled claude.exe, triggering an infinite re-download loop
- [BUG] Intermittent "Subprocess initialization did not complete within 60000ms" on macOS — interactive sessions (VS Code extension / desktop app / interactive CLI) hang, `claude -p` works; correlates with TLS-inspection (Norton)
- [BUG] Session limit hits very fast and "API Error: Server is temporarily limiting requests..."
- [BUG] [Max 5x] Weekly usage limit reset day drifts forward each week (window exceeds 7 days)
- Task/Todo tools (TaskCreate/TodoWrite) not exposed despite `todoFeatureEnabled: true` — regressed ~2026-07-21, appears account-gated
- [Bug] Model substitution mid-session causes loss of user-authored security rules
- [Bug] Project-scoped Skills not appearing in /context list despite correct placement
- [BUG] Projects missing from desktop app project list
- [Bug] Anthropic API Error: Usage credits required despite extended plan validity
- [Bug] Model safeguards blocking legitimate security advisory composition
- [BUG] Sonnet (default) - Usage credits required for 1M context
- Sustained ECONNRESET on streaming connections - Claude Code VS Code extension (Santiago, Chile / SCL region)
- [Bug] In-session /tasks panel prematurely clears completed subagents without documented cleanup trigger
- [RESOLVED - not a CLI bug] Orchestrator SIGTERMed its own agents every 300s; surfaced as aborted_tools with no logged cause
- [BUG] Scroll in WSL or PowerShell WSL Claude sessions do not scroll the chat, only the text entry field
- [MODEL] Claude repeatedly uses openpyxl for Excel charts despite it being fundamentally broken for time-series
- [BUG] Agent teams: teammate sessions die with ConnectionRefused (Bun tries IPv6, no IPv4 fallback) while the lead keeps working
- [BUG] External editor (Ctrl-G) reprints the full welcome splash on every round-trip — v2.1.176 & v2.1.179
- [Bug] Agent performance degradation compared to baseline
- [BUG] Removing Desktop Connector not possible
- [BUG] -ECONNRESET
- Regression: sandbox no longer blocks commands not in excludedCommands (default-deny model broken)
- I appreciate the feedback, but I'm unable to generate a GitHub issue title from this message. This appears to be a personal opinion/sentiment about Claude's performance compared to other tools (Codex, Cowork) rather than a specific, reproducible bug report
- IMPOSSOBLE TO USE THE TOOL FOR ANYTHING. ALL RULES IGNORED. ALL SCRIPTS IGNORED. MALES OWN DECSIONS. A FUCKINGSHITSHOW ALL DAY EVERY DAY
- [Bug] Text rendering corruption in prompt output
- [BUG] Cowork falha ao iniciar workspace no Windows ARM64 — "VM connection timeout after 60 seconds
- [BUG] Claude dropped a character from user-supplied Team ID when generating JWT, then treated broken token as source of truth — caused months of silent API failures
- Suggested prompts referencing file paths are not validated before display
- 1.5h for one simple edit still ot exectuted. I WILL FILE A POLICE REPORT FOR METAL ABUSE AND HARM. IM DONE
- Claude Code dumps live API tokens via env grep — credentials exposed to Anthropic API
- Feature Request: Custom folders to organize conversations on the home page
- [BUG]
- [BUG] claude-code-assisted PR labels never applied while analytics detects the PRs — follow-up to #35167
- [FEATURE] durable sub-agent execution log + limit-aware dispatch (work is silently lost when the usage wall hits mid-fan-out)
- [FEATURE] Option to "Skip" the suggested change and continue with next proposed change
- Unexpected 5-hour limit consumption in Claude Code
- [DOCS] Connection-drops section omits that partial responses are preserved and the spinner no longer gets stuck at "running tool" (v2.1.179)
- [BUG] TUI input lags / drops keystrokes under heavy host CPU load (e.g. compilation) — other apps stay responsive (Windows, 2.1.177)
- [DOCS] [Fullscreen] Mouse-wheel scrolling in WSL2 under Windows Terminal and VS Code regressed in v2.1.172 and is now fixed in v2.1.179, but the docs do not mention WSL2 at all
- [BUG] Claude Desktop 1.11187.4 (and likely 1.13576.0.0) fails to launch on Windows 11
- [DOCS] Sandbox docs do not mention the v2.1.179 `denyRead`/`allowRead` glob fix or warn that broad globs over large directory trees can blow up the Bash tool description
- [BUG] Voice dictation input not cleared after sending message in Desktop app
- ${CLAUDE_PROJECT_DIR} not substituted in .mcp.json env for some MCP server launches (literal token reaches server)
- [Bug] High rate limiting and API errors degrading service reliability
- [BUG] Subagents aren't shown the agent-type roster their Agent tool docs promise (injected lazily; visibility only, not availability)
- [BUG] Claude Code needs AGENTS.md support, repository-aware tool selection, and stricter Git controls
- Unable to install any connector in claude desktop
- [FEATURE] Allow a hook to change the session working directory (programmatic /cd)
- [BUG] 🚨 `--continue` and `-p` are now seriously broken together in 2.1.90
- [FEATURE] Directly execute a tool with `/tool`
- [BUG] Unable to install Claude Code
- [BUG] CI monitoring: sessions spawned from a task shard share branch subscriptions and a single global auto-fix toggle
- [BUG] Remote Control bridge fails silently on Desktop app
- [FEATURE] Add voice conversation mode to Dispatch — talk to Claude, not type to it
- [BUG] Cowork VM fails to start — SDK download truncated at ~4 MB, checksum verification fails every attempt (macOS, Apple Silicon)
- [FEATURE] Support substituting ${CLAUDE_PROJECT_DIR} in skills.
- VS Code/Cursor: Claude panel restores blank after Remote-SSH reconnect (deserializeWebviewPanel does not rehydrate conversation)
- I asked claude to research something with sonnet agents it started with 5 but then they spawn their agents in total 273 background agents
- Feature request: Codex-style live steering (inject input into the in-flight turn)
- [FEATURE] Per-model effort level in settings.json (modelEffort map)
- [BUG] Workflow subagents balloon to ~2GB each and aren't cleaned up — regression 2.1.156 (OK) → 2.1.160
- [Bug] Skills not loading from CLI or ./claude/skills directory
- Claude Code gives incorrect information about Anthropic API free credits and misleading 'Skip for now' button during signup
- [BUG] No repos match in repo picker despite GitHub app installed and connector connected
- [BUG] Plan mode's markdown output contains broken relative path links
- [BUG] Claude Code v2.1.173: ui completely broken
- [FEATURE] Agent Plan Reviews
- [FEATURE] Register as VS Code inline chat edit tool provider
- Parallel subagent file edits clobber each other in the IDE diff editor (N−1 spurious permission-denials); diff-review lock is per-agent, not per-IDE-connection
- Allow MCP-provided Edit/Write tools to render diffs in the UI
- [BUG] Concurrent sessions in the same desktop app corrupt tool results: text injection, fabricated file state, silent dropped/partial writes
- [Windows] MCP tools return "Browser extension is not connected" while native host process is alive and pipe is reachable (v2.1.141)
- Cowork: third-party private-repo marketplaces permanently stale per-account — version label frozen at first-add value, re-add doesn't refresh, releases not picked up
- [Feature Request] Reduce ambiguous yes/no prompts and clarify recommended action in interactive choices
- [BUG] iOS app crashes when fetching environment list if any remote-control server (kind: "bridge") is registered
- [BUG] VS Code: Input text lost when switching between chats or restarting extension
- [MODEL] Opus 4.8 asserted documented project knowledge was wrong based on a self-built test, dismissed a correct user report until pushed twice
- [BUG] Fable is not available
- [BUG] Claude Managed Agent internal proxy throwing 503 when interacting with Git
- [BUG] /MCP connector name not picked up for OAuth-protected servers — UUID used instead of `serverInfo.name` in Claude.ai — inconsistent behavior
- [BUG] - cowwork não instala
- feature: surface autopilot mode consistently across all entry points (CLI, slash command, mode picker, /config)
- Claude Code for VSCode is not loading models list [2]
- User messages lost blue text colour after recent update — now grey on dark theme