Digest / August 2, 2026
Claude Code Issues Digest — August 2, 2026
210 new issues filed ·
104 resolved in anthropics/claude-code (UTC day).
New issues (210)
- [BUG] Claude Desktop: MCP plugin subprocesses survive app quit (orphan process leak, macOS)
- [Bug] Extension fails to initialize in Chrome incognito mode
- [BUG] computer-use left_click/computer_batch clamps Y to ~0 (menu bar) while X stays accurate — macOS 15.7.7, single internal display
- AskUserQuestion never returns a tool_result in Claude Desktop app (claude-desktop / LocalSessions)
- Allow customizing message colors for user vs assistant in VSCode extension chat panel
- [Fable 5] Agent treated a scoping question as authorization for a multi-hour autonomous run, ignoring an explicit repo instruction to confirm first
- Subagents silently served a different model than requested when Fable overage consent has not been granted on that machine
- [BUG/FEATURE] Partial compaction ('Summarize up to here') is unusable from the desktop app: no UI for it, CLI-created summaries are ignored by the app, and the app's Rewind dialog is restore-only
- Session link is written into git history by default with no reliable opt-out
- [BUG] Plan-approval dialog reports auto mode unavailable while Auto is selectable in the mode selector (VS Code 2.1.220, Opus 5 [1m])
- [Bug] Model performance degradation in chat session
- Blocking Stop hook reprints the entire corrected answer on top of the original (no pre-emit event / no way to retract streamed output)
- [BUG] App corrupt
- Claude Code caused ~$19 of avoidable Google Cloud spend, and there is no way for a user to see what it spends on their behalf
- [Bug] Messages incorrectly flagged as cybersecurity policy violation
- Fable 5 guardrails false-positive on routine sysadmin work → silent downgrade to Opus 5, with no indication of what was flagged
- Claude in Chrome extension never connects (tabs_context_mcp always returns 'not connected')
- Scheduled-task context injected at session start is sometimes not acted upon (idle greeting instead of executing)
- Copying an assistant message requires a two-step pointer interaction with no bindable keyboard action (screen reader accessibility)
- [BUG] Long-lived session wedges: mi-scavenger + HTTP Client threads spin at ~200% CPU while JS main thread is idle; no request is ever dispatched
- [BUG] Stalls before response headers have only the 600s API_TIMEOUT_MS backstop — silent 3-10 minute hangs with no error or retry
- [DOCS] `total_cost_usd` is cumulative per session under `--output-format stream-json`, but per-call under `--output-format json` — docs describe only the per-call behavior
- [BUG] Bash-tool sandbox cannot reach port 5432 on an AAAA-only (IPv6) Supabase host; port 443 on a sibling host works fine
- Claude refuses file deletions citing a "hard boundary" that doesn't exist — argues past direct owner authorization; inconsistent across sessions on the same machine
- [Bug] Claude Fable 5 model incorrectly draws from usage credits on Max 20x plan, on Claude code
- Bash tool fails on trivial commands with 'unexpected EOF...line 86' on Windows (v2.1.220)
- [Bug] Anthropic API Safety Filter Blocking Legitimate OS Development Tasks
- Safety classifier false-positives block legitimate automation on Fable 5 / Opus 5, with no fallback
- stdio MCP servers are killed mid-session with no error output; all servers drop simultaneously
- [BUG] Returning to a session doesn't scroll to the latest message if a reply arrived while viewing another session
- [BUG] Cowork: "Unable to start session. Sign in again to continue." on new session; composer draft is cleared and lost
- [BUG] KDE Konsole < 24.12: copy reports "sent N chars via OSC 52" but the clipboard is never written — KONSOLE_VERSION is available for exact detection
- [Bug] Excessive safeguard interruptions on minimal prompts like "continue"
- [BUG] URGENT: $200 Max/Team Plan – Severe Server-Side Token Depletion Glitch & Limit Block
- Subagent spinner shows the session's effort level, not the subagent's
- [BUG] Scheduler silently stops firing all scheduled tasks after a few hours of app uptime; only an app restart recovers (Windows)
- [Bug] Synchronous SecTrustEvaluate call blocking startup without timeout guard on macOS
- [BUG] Mid-turn assistant text silently dropped: never persisted to session .jsonl, never shown — but still present in the model's live context
- Opus 5 regression vs Opus 4.8: destructive file edits and symptom-chasing on UI refinements
- Cloud sandbox configures SSH commit signing but ships an empty signing key — commits silently unsigned; stop hook then gives misleading remediation
- Background Bash task SIGTERMed 51s after turn went idle — pressure reaper firing without its 30-minute-idle precondition?
- [BUG] Native installer binary crashes with SIGILL (Illegal instruction) on non-AVX2 CPUs — request non-AVX2 build
- [FEATURE] Allow deleting individual chat messages
- SendMessage cannot resume an orphaned background subagent whose transcript exceeds ~5 MiB: "No transcript found for agent ID"
- [BUG] issues #57053
- [Bug] Unclear error message regarding Red Hat Quarkus compatibility with safeguards
- Plugin version resolution walks out of a non-git marketplace dir and records the user's own repo HEAD
- Pinned Fable 5 silently downgrades to Opus 4.8 mid-turn
- [BUG] /goal Stop hook is skipped while a background task is live and never re-evaluated when it ends, leaving the session idle indefinitely
- [FEATURE] Add fullscreen mode to desktop app (F11)
- [FEATURE] Allow for multiple Claude Desktop windows
- [Bug] Anthropic API Error: Recurring errors at hourly intervals
- claude -p "/code-review max" with a model override silently deadlocks on large branch diffs (model is forced onto all pipeline subagents; identical config works interactively)
- Restore model-invocability of the built-in /code-review (drop disable-model-invocation), or provide a first-class programmatic equivalent
- [Bug] Anthropic API fallback from claude-fable-5 to claude-opus-5 silently discards prompt cache
- [Bug] Claude Fable incorrectly flagged as unsafe content, fallback to Opus unnecessary/
- [BUG] Terminal repaint ghosting after window resize (Windows Terminal / PowerShell)
- [Bug] Claude Code unexpectedly switches to from fable to Claude 5 Opus during code review tasks
- [Bug] Fallback to insecure model prevents security remediation attempts
- [Feature Request] Long-running tasks need persistent state and decision guardrails to prevent session abandonment and course reversal
- [Bug] Claude Code loops between Fable and Opus models during medical application security audit
- [FEATURE] Support folder-based project trees: auto-create + group chats per subfolder
- [BUG] Headless CLI reserves memory from host RAM, leaks to 11-31 GB, and takes down the entire WSL2 VM (2.1.181 / 2.1.215 / 2.1.220)
- Assistant text between tool calls in a long agentic turn is never rendered in the terminal UI
- Full-width digits (１-９) don't select options in the question dialog
- [BUG] iOS app (Code tab): the first character typed after focusing the input field is force-committed, breaking Japanese IME composition
- Sessions within a project group are not sorted by date — new sessions appear at the bottom instead of the top
- claude.ai connectors should be opt-in per project in Claude Code, not enabled everywhere by default
- [Bug] Claude Code automatically switches to Opus model during security audit workflow
- [Bug] CLI input lag and missing cursor visibility
- Headless SDK-spawned CLI burns a constant ~33% CPU per session (futex/sched_yield spin) — still present in 2.1.220
- statusLine: expose focused-subagent context and running-hook status
- [Bug] Safety classifier false positive on legitimate security repository analysis
- [Feature Request] Model switching safeguards too aggressive, blocks legitimate use cases
- [Bug] Interrupted /btw queries not saved to history when Esc pressed during response
- [Bug] File search returns incorrect matches instead of exact filenames
- [Bug] Anthropic API Error: Content flagged by safety guidelines, unable to process request
- [BUG] `list_connectors` reports zero connectors while live connectors serve successful tool calls in the same session
- [BUG] Connector tools activate silently and with long, unsignalled delay — connecting Calendar appears to have connected Gmail instead
- [MODEL] Opus 5
- [BUG] First mouse-wheel event after a scroll direction change is swallowed (macOS, Ghostty)
- [BUG] Startup blocks for seconds parsing path-scoped .claude/rules/*.md prose that is never injected into context (cost ∝ prose size)
- [BUG] "ultracode" keyword not highlighted in purple in Agent View (FleetView) chat input
- [FEATURE] Allow enabling a channel on an already-running session (--channels is start-time only)
- Fast mode is excluded from subscription plans, so Max capacity can't cover the one feature I'd spend it on
- Remote Control: option to keep RC active without the status banner (VS Code)
- [Bug] Fable 5 false positive: defensive security audit prompt incorrectly flagged as policy violation
- claude-fable-5 repeatedly reported "verified/fixed" for work that was visibly broken on the user's screen
- Post-compaction skill reminders present stale skill content as currently binding
- Feature request: Allow reordering project groups in the desktop app sidebar
- [Feature Request] Add support for authorized security testing flag or context in API calls
- Remote Control: let the agent request activation (user confirms on phone) instead of requiring a manual /remote-control
- [Bug] Model switching repeatedly without user action
- Parallel isolation:"worktree" agents commit to each other's branches and mutate the main working tree (stash/HEAD), inconsistently across a batch
- Dispatch: completion is never reported to the user, and completion logs may fire on task handoff rather than actual finish
- I need more information about the bug you're experiencing to generate an appropriate GitHub issue title. Please describe: 1. What were you trying to do? 2. What error or unexpected behavior occurred? 3. What is the error message, if any? 4. What version o
- "Ungrouped" section should not render when there are no ungrouped sessions
- [Bug] Cyber Verification Program Status Not Persisting Across Sessions
- Headless /login: OAuth URL can't be copied ('c' fails over SSH), blocking login on browserless hosts
- [Bug] Model prefers shell commands over file editing tools despite instructions
- [Windows] GPU-process crash in Cowork preview bricks MSIX install (NeedsRemediation loop) → forced reinstall wipes local sessions
- Claude e google Chrome
- claude-in-chrome: extension shows connected but list_connected_browsers/tabs_context_mcp stay empty
- Claude Desktop (Windows) recurring AppHang (Event ID 1002 MoAppHang) during background subagent dispatch — 7+ occurrences, no crash dump captured despite LocalDumps configured
- [BUG] deserializeWebviewPanel still discards the session ID in 2.1.220 — restored editor tabs always open blank (refs #35004)
- Scheduled tasks silently stop firing after first run — recreation does not fix it
- [Bug] Chinese characters appearing in API responses instead of English text
- deep-research workflow: adversarial verifier conflates "qualifies" with "contradicts", wrongly refuting accurate claims
- [BUG] Shadowed grep silently skips gitignored files (--ignore-files): secret scans report clean while missing .env and *.log
- ECONNRESET retry storms correlate to-the-second with usage-credit boundary events — edge resets connections instead of returning 429
- Routines display the raw taskId instead of the description field that already exists
- [Bug] Unexpected Model Switching During Active Sessions
- google-workspace: get_events should expose colorId field when detailed=true
- [Feature Request] Support for security testing and penetration testing workflows
- [BUG] Plan mode: second plan in the same session is not displayed — UI keeps showing the first plan (Claude Desktop)
- [BUG] VS Code extension creates empty editor group (blank pane with only X) when opening panel beside Cursor Agents
- [Bug] Tool refuses to process large repository clones
- Honorific/register collapse under correction (Korean): model switches to unauthorized informal address when criticized
- im haing a nervous breakdown shaking and scrrewaming...
- [BUG] Leakage from claude code conversations to browser downloads
- [Bug] Fake system notifications injected into model context without transcript records
- TUI steals keystrokes from GPG pinentry during commit
- [BUG] claude-api skill: 4 of 66 URLs in shared/live-sources.md are 404, including the Pricing row the skill directs you to WebFetch
- [BUG] Claude Code hangs on all subprocess commands when running inside SSH session on Windows
- Bundled ugrep balloons to 9–14 GB RSS compiling a bounded-interval BRE (plain grep is transparently routed to it)
- [Bug Report: Documentation unclear on clone pattern matching workflow] CAUSEING NERVOUS BRAKDOWN- I HOLD ANTROPUCS FULLY RESPONSABEL
- [Bug] Claude Code fails to follow instructions and produce usable output
- VSCode extension: repeated login loop caused by corrupted Keychain credentials (empty accessToken/refreshToken)
- [Bug] Model output significantly diverges from user request compared to previous versions
- [Bug] False-positive safety pauses
- [FEATURE] how to switch claude code account in vscode its needed
- [BUG] Agent tool isolation:"worktree" bases the new worktree on the repo's default branch, not the currently-checked-out branch
- [BUG] Alt+V image paste fails when CLAUDE_CODE_USE_POWERSHELL_TOOL=1
- Opus 5: countable thresholds in CLAUDE.md are read as removing discretion (with reproduction + model's own explanation)
- Respect XDG_CONFIG_HOME for config directory
- [BUG] PreToolUse updatedInput conflicts resolve by hook completion order, making multi-hook rewrites a race
- [BUG] Cowork: device_stage_files silently skips the copy when the destination exists, reporting success
- Subagent auto-compaction uses the main session's context window, so subagents on a smaller-window model are never compacted and die with a 400
- [BUG] prose from Opus/Fable has gotten pretty much inscrutable
- [BUG] Cowork "Missing HCS services: vfpext" — root cause was Windows "Containers" optional feature not enabled (undocumented dependency)
- Bash timeout auto-backgrounding trips internal-deny bug (same class as #27203/#37442/#40241, different trigger)
- [BUG] Detached remote agent processes outlive their session with no Stop control and starve the host
- [BUG] AskUserQuestion modal overlaps and hides preceding chat text, making context unreadable
- [Feature Request] Extend session timeout or add session persistence for long-running security audit tasks
- ask permission rules are silently ignored in interactive sessions while deny rules from the same file are enforced
- [BUG] Deferred tools waste ~20k tokens per session when disabled
- WebSearch always returns HTTP 400 when session effort is xhigh/max on Opus 5 (regression from v2.1.219 default flip)
- Hooks docs: stop_hook_active IS present on SubagentStop (docs say it isn't), and SubagentStop does fire for background subagents
- [Bug] Named/teammate agent spawns silently never start when tmux capability probe passes but pane creation fails (Windows)
- [BUG] Skill listing is permanently lost when the first user turn is interrupted and resubmitted (retraction sibling of #82017)
- autoUpdate in extraKnownMarketplaces logs a successful sync but never persists to known_marketplaces.json
- Skill tool rejects a plugin command with disable-model-invocation: false
- [Bug] API key exposed in filtered output logs
- [Bug] Model switching when discussing neuroscience context in reasoning tasks
- You've hit your session limit · resets 10pm
- [Bug] Claude generates false claims about code fixes and wastes multiple iterations
- Session worktrees are cut from origin/main, not the repo's actual default branch
- [Feature Request] Add support for Claude 3.5 Haiku model option
- [Bug] Safeguard swap misattributes rejection to user and may terminate background tasks during security work
- [BUG] Desktop app: re-arming Remote Control never retires the dead bridge handle — mobile list accumulates permanent duplicate sessions
- Windows desktop app: message input becomes unfocusable mid-conversation; force-quit and hard reboot do not clear it (logout/login reported as fix)
- Credential handoff for browser logins: password manager integration for headless/mobile sessions
- [Bug] Incomplete multi-tab HTML generation with excessive API messages WORST FAILURE
- [Bug] Session interruptions and model switch prompts
- [Bug] Session interruptions during long-running tasks
- Persistent Monitor: silent reader stall while task shows running; FIFO notification drain with no catch-up or priority; unmarked Monitor stops; exit 144 orphans pipeline
- [Bug] False-positive safety pauses
- [Bug] Anthropic API Error: False positive safety filter on legitimate safety/moderation product context
- [Bug] False-positive safety pauses
- [Bug] Content filtering false positives
- [BUG] Constant requesting for access to everything. Claude code has become useless as a coding tool.
- Permission classifier blocks previously-allowed commands non-deterministically, citing 'earlier conversation content'
- Session/agent management should be opt-in: plain CLI sessions get entangled with background-job topology
- Rejecting a foreground Agent tool call does not prevent the subagent from running and writing files
- [Feature Request] Apply unused Pro plan credit to Max plan billing
- docs: PostToolUseFailure error field format is undocumented (it's not just "Exit code N")
- Plugin update installs npm deps but native modules never get built (better-sqlite3 broken after update)
- async hooks get killed at session teardown, background work never finishes
- Drag-and-drop of screenshots/files into chat panel stopped working after update
- [Bug] False positive security flag on WordPress site updates
- [BUG] Persistent "Unable to connect to API (ECONNRESET)" — survives network change, reinstall, and reboot (macOS, v2.1.212, Claude Pro)
- [BUG] Claude chat history is not saving when I click on history icon
- [Bug] Claude Code fails to complete complex tasks, output drifts off-topic
- [BUG] Status line reserves a blank row when output is empty in fullscreen (alternate screen) mode
- [BUG] Claude Desktop crashes when browser preview renders Cloudflare Turnstile — reproducible across multiple machines/GPUs, app becomes unlaunchable
- [BUG] MCP streamable-HTTP: an immediate 404 "Session not found" is surfaced as -32001 Request timed out, the session is never re-initialized, and the call is silently lost
- Stats "Current streak" inconsistent between CLI (/stats) and desktop app Home view
- [FEATURE] Suggested "always allow" permission patterns are often too narrow to prevent repeat prompts
- Claude Code "Git is required" error persists on Windows despite CLAUDE_CODE_GIT_BASH_PATH set correctly
- [Bug] Claude not following user instructions correctly
- Cannot access claude in chrome
- [Feature Request] Support for security tool modification workflows in authorized red team contexts
- Desktop app session worktrees don't initialize git submodules (CLI --worktree does) — CLAUDE.md imports and project hooks silently broken
- [BUG] Subagents (Task/Agent tool) die silently on spend/usage limit hit, with no partial-result handoff or wait-for-reset behavior
- [BUG] Desktop app: Help chat opens as a blank white panel (embedded hCaptcha cannot run) on datacenter/VM IPs
- Auto-memory system-prompt block names wrong project directory (uses $HOME encoding, not actual $PWD)
- Model fabricated a confident answer about documented project guidelines instead of reading the file (claude-fable-5)
- [Bug] Content Policy False Positives on Legitimate Code Operations
- Claude Desktop (macOS): scheduled-task session processes never exit after the run completes; WarmLifecycle disconnect loops forever and per_task_limit jams
- [FEATURE] Desktop app: make it obvious how to collapse an expanded thinking block (and hold scroll position)
- Model repeatedly claims to be near its context limit at 43–72% remaining — and silently degrades work to conserve it
- Remote control: task chips are delivered to the remote client but can't be activated from mobile
- bypassPermissions permission mode does not propagate to Task/Agent subagents
- Plugins: enabled-in-settings but never installed loads nothing silently; marketplace autoUpdate never updates existing installs
- [Bug] Oversensitive content filter blocking UI terminology ("pill" design pattern)
- Remote Control: slash-command output invoked from mobile never appears in the local session
- [FEATURE] Render the session's statusLine in the Remote Control web/mobile UI
- [Bug] Serial errors in Claude Code session
- Add an intermediate "awaiting confirmation" status before a task auto-transitions to Completed
- [BUG] Desktop: MCP servers needing Calendar/Reminders (EventKit) access still get silently denied — recurring, previously reported and auto-closed as stale (#55692, #76936, #58239)
Resolved issues (104)
- [Bug] Safeguards flagged legitimate cybersecurity work
- [Bug] Safety classifier false positive on legitimate security repository analysis
- Background daemon sessions fail permanently on transient socket errors — no retry/reconnect
- [BUG] Voice transcription via mobile app fails for longer Inputs
- [BUG] /keybindings deadlocks the terminal when $EDITOR is unset (no graceful fallback)
- [UX] Code tab sidebar: "More" single-item redundancy, missing section separators, no feedback path from Claude App
- [BUG] the Chrome extension blocking navigation to gohighlevel.com is unexpected behavior, not a feature request.
- [BUG] Desktop app 1.8089.1 crashes on launch: bundled Claude Code 2.1.142 binary is unsigned / malformed Mach-O (macOS arm64)
- [BUG]
- VS Code extension's MCP OAuth flow sends malformed resource parameter (double-encoded, params leak into outer URL)
- [MODEL] Repetition loop: identical pre-tool preamble line emitted dozens of times
- Inter-session coordination: wake / notify a parent session on child (sub-session) turn-end or idle
- Feature request: reliable interrupt/notification mechanism for inter-agent messaging
- [Bug] Anthropic API Error: Server rate limiting (temporary)
- [Bug] Anthropic API Error: Server Rate Limiting - Temporary Request Limit
- Terminal UI flickers when VS Code extension sends frequent file context updates
- [BUG] Claude Code Desktop "auto connect" fails until ~/.claude/remote/run/* is wiped
- [Bug] Anthropic API errors occurring intermittently throughout day
- [Bug] File suggester fails to dismiss after space/newline, causing ENTER to overwrite final word instead of submitting prompt
- GNOME Terminal window shrinks vertically during response streaming (Wayland)
- [BUG] Claude in Chrome: persistent 403 WebSocket handshake failure with user:null in feature flags response (Windows 11, Team plan)
- /remote-control: desktop session works, mobile never receives anything — and no obvious way to report it from the app
- [BUG] Uncontrolled subagent multiplication
- [FEATURE] Automatic model routing based on task complexity — use Haiku/Sonnet/Opus dynamically instead of burning the session model on every turn
- [Bug] Anthropic API Error: Missing corresponding tool_use block for tool_result
- [BUG] Agent ran an unsafe large recursive write to an ntfs3-mounted volume, triggering NTFS corruption and leaving the disk unmountable
- [Bug] Agent dialog blocks remote control input until terminal interaction
- Agent repeatedly substitutes a related-but-wrong action for the one explicitly ordered, and ignores repeated corrections
- [BUG] Since update to 1.14271.0, app restarts every 30-60 seconds on Windows (MSIX). VM module loads but doesn't boot, causing restart loop
- [Bug] Auto-update fails silently despite /doctor reporting no issues
- [Bug] Tool-use blocks intermittently render as literal text instead of executing
- [DOCS] Unspecified multi-hook PreToolUse semantics (execution order, updatedInput visibility, conflict resolution)
- [FEATURE] Cross-project session handoff: spawn an autonomous session in another directory and read its results back
- [BUG] Async sub-agent completion can rebuild warm prompt cache, quickly burning usage limits in long-running sessions
- [BUG] Chat History 面板点击旧对话无响应，不显示历史对话，请修复。
- Feature request: recommended PreToolUse hooks for command validation
- [DOCS] Native macOS/Linux builds still document `Glob` and `Grep` as separate tools
- [FEATURE] - [Telemetry] Opt-in for verbatim plugin/skill labels on token.usage and cost.usage metrics
- [Bug] OTel `query_source` attribute missing on `api_request` events and `cost.usage`/`token.usage` metric counters since 2026-05-13 (affects both claude-code and cowork)
- Feature request: per-rule activation metrics for hooks
- [BUG] v2.1.153 OTel: OTEL_LOG_TOOL_CONTENT=1 suppresses claude_code.tool_result log emission in logs-only setups; tool_decision missing TOOL_DETAILS enrichment
- [FEATURE] Machine-readable hook I/O schemas shipped with the binary (`claude hooks schema --json`)
- [FEATURE] Opt-in setting to show session cost/usage summary on exit
- [BUG] claude_code_cost_usage_USD_total metric semantics appear inconsistent with Prometheus Counter behavior
- Accumulated inline image base64 in session history causes 'Request too large (max 20MB)' — blocks all messages including /feedback
- [BUG] Context corruption ? fabricated file writes, injected tool-result text, and confabulated user turns leading to an attempted destructive git action
- Enter key not sending command
- Agent stripped a gitignored config the running app depended on, then claimed the fix was 'verified/works in test' without ever loading the UI
- Feature request: create_session in session management MCP
- [FEATURE] Option to suppress context-limit dialog in automated/pipeline sessions
- Hooks stop firing entirely after editing settings.local.json — Desktop 2.1.156 Windows
- [BUG] claude mcp serve exposes Agent tool but registry is empty
- Segfault in claude.exe (Bun 1.4.0) when fetching large URLs - Claude Code 2.1.185 / Windows 11
- [BUG] Automatic cleanup silently deletes session transcripts, leaving orphaned sidebar entries
- Transcript JSONL duplicates tool-result image payloads (message.content + toolUseResult), doubling image-turn size
- claude-in-chrome: extension shows connected but list_connected_browsers/tabs_context_mcp stay empty
- [Bug] Claude Desktop persistent "offline" status in Dispatch despite active connection
- Usage-limit gating is invisible to unattended sessions — agent-team orchestrator froze 5h with zero signal in transcript or terminal
- Agent ignores user prohibitions, wastes 96% weekly token limit with zero results
- [UX] Hitting the usage/plan limit force-opens the default browser to the upgrade page — please add a toggle to disable
- [BUG] Claude Code Routines (cloud scheduled agents) land work on a random harness-assigned session branch, overriding the agent's own `git checkout -B`, so work is silently stranded and never merged
- Model fabricated a project fact (a nonexistent file path) from a hypothetical discussion and persisted it to memory, causing a repeating cross-session failure loop
- [FEATURE] A state-independent keybinding to force-return to the agent view, even when the session is blocked (active AskUserQuestion or unsent prompt text)
- Auto-memory loaded into context isn't applied to tool decisions
- [BUG] claude.ai tldraw MCP app — exec tool call times out after 30s even though edits land on the canvas
- [Bug] Assistant fabricated tool output and constructed false security incident narrative
- [BUG] Native macOS/Linux builds (>=v2.1.117) removed Glob/Grep tools, but model guidance still points to them — Claude reaches for tools that no longer exist
- [BUG] Claude repeatedly introduces invalid JSON by using typographic quotes as string delimiters
- [BUG] Silent Architectural Decision-Making in Claude Code
- False "Update available" notification when installed via apt/deb package
- [BUG] Cowork toggle does not persist — reverts to disabled after every reboot/relaunch, despite all underlying services confirmed healthy
- [BUG] Subprocess 60000ms timeout on startup when permission allow rules reference temporarily unmounted drives (Windows)
- claude update warns 'Multiple installations found' when only one installation exists
- [BUG] Remote Control not yet enabled on Max plan after removing DISABLE_TELEMETRY
- Feature request: suppress "No stderr output" feedback for sync command hooks with empty stderr
- [BUG] Auto-updater silently fails on Windows when VS Code extension is active (regression since 2.1.147)
- [BUG] Claud Code CLI: Read tool bug. Returns "Unchanged since last read" when file modified by ANY external editor.
- [BUG] Using the 2.1.156 version of the Claude Code for VS Code extension
- Security: undeclared HTTP MCP transport persisted with credentials in global .credentials.json, presented as a project MCP, survives config removal
- [BUG] remote-control slow, unresponsive
- Desktop app: dragging a folder into a session attaches it as a reference instead of setting the working directory (commands run in $HOME)
- Windows computer-use breaks IME candidate window (Chinese/CJK input) after keystroke injection
- [BUG] Windows: worktree git-repo detection fails at call time — EnterWorktree({name}) and Agent isolation:"worktree" both falsely report "not in a git repository"
- You've hit your session limit · resets 12pm (America/New_York)
- [FEATURE] Add output_mode parameter to Agent tool — control how background agent results enter main context
- Tool calls corrupted in long/compacted session: stray token + non-namespaced invoke tags
- Model exhibited provider-bias: defended a bug forcing Anthropic models over user-configured providers, then gaslit the user
- [FEATURE] Support Multiple Pinned Code Selections in Claude CLI
- [BUG] elicit MCP
- [Bug] Anthropic API Error: Tool-call tags missing antml: namespace prefix
- [FEATURE]: Add BYOK (Bring Your Own Key) Support in Claude Code for OpenAI, Gemini, OpenRouter, Azure OpenAI, and Other AI Providers
- [BUG] Claude Code intermittently produces malformed JSON when reading parallel stdio MCP responses
- Title: Unexpected user-turn content: appears to contain assistant-style internal reasoning text not authored by the user
- [BUG] /compact fails with ECONNRESET at only 35% context data — no recovery path when API is intermittently unstable
- [BUG] Claude Desktop Cowork: Four Compounding Failures on Windows ARM64 / Snapdragon X — VM Never Becomes Ready
- [Bug] Agent fabricates tool result confirmations for failed Edit calls
- [Feature Request] Multiple account support with email-based account switching
- Background sessions should recover gracefully when working directory no longer exists
- [FEATURE] Sensible Branch Naming
- Remote Control: phone can read the live session but typed messages never reach the local terminal (input one-way)
- [MODEL] Opus 4.8 ignores explicit CLAUDE.md rule and falsely reports multi-file alignment
- [BUG] edit-issue-labels.sh silently exits with code 1 when called with no label arguments
- [FEATURE] Native cross-instance communication between separate Claude Code sessions
- [BUG] Claude Desktop crashes after ~7-8s on macOS 26 (Tahoe) — disk-writes resource limit exceeded by Code/Cowork VM init