Digest / July 26, 2026
Claude Code Issues Digest — July 26, 2026
227 new issues filed ·
163 resolved in anthropics/claude-code (UTC day).
New issues (227)
- [BUG] Windows MSIX servicing: updater registers into hung app (0x80073D02); repair source deleted from %TEMP% (0x80073CF0); container-locked files block reinstall (0x80073CF6); package service requires elevation (0x80073D28)
- MCP truncation at 2048 chars is invisible: tool `description` leaves no log, `/mcp` shows the untruncated text
- [Bug] Claude ignores "do not push to production" instruction from claude.md
- Background Bash task notification reports exit code 0 when the command exited 1
- Model substitutes countable verification for judgment-requiring verification, then reports the work as complete
- Model proceeds past explicit project prohibitions and scope gates that were present in context
- [BUG] Auto-mode catastrophic-removal guard bypassed: `rm -rf` inside a backtick substitution executes without a prompt
- [Feature Request] Add persistent status line displaying active model
- [BUG] Claude Desktop MSIX 1.24012.9: opening the in-app Browser pane crashes the whole app - GPU process exits with 101457950 (0x60C201E) on Intel HW, NVIDIA HW and WARP software rendering alike
- [FEATURE] Screen sharing feature
- [FEATURE] A port-allocation coordination primitive for concurrent sessions (re-file of #34385)
- [BUG] CLAUDE_CONFIG_DIR not honored - Critical data separation issue
- [Bug] Authentication fails for test users without password enforcement
- [BUG] Merge Icon in Claude Coe gets Stuck
- [BUG] Stale ~/.claude/.credentials.json causes "OAuth token has expired" even when a valid CLAUDE_CODE_OAUTH_TOKEN env var is set
- git-credential-proxy failing for push operations (403) since 2026-07-24
- [Bug] Fable 5 model unavailable in VSCode extension and CLI despite availability in Claude app
- [Bug] Fable-flag false positive during keystroke diagnostic instrumentation
- [Bug] Fable 5 rejects trivial arithmetic with contradictory error response (is_error: true, subtype: success)
- [FEATURE] "Hallucinating..." funny string
- [FEATURE] "Uploading" visual clue
- Opus 5 AUP safeguard repeatedly flags benign messages in a legitimate security-research session (BBH/VSF multi-agent framework)
- Show a plain-English reason when a command is blocked for containing unreviewable control-character bytes
- Desktop: auto-compaction writes to two transcripts in parallel, then the session switches files mid-conversation (history appears to roll back)
- ${CLAUDE_PROJECT_DIR} resolves against a stale, drifted cwd in exec-form hooks after a mid-session cd
- [Bug] Claude Code fabricates decision provenance and overrides explicit instructions without verification
- You've hit your org's monthly spend limit · run /usage-credits to ask your admin for a higher limit
- [Bug] Model auto-upgrade to Opus 4.8 instead of latest Opus 5 on complex tasks
- Built and maintained ToS-violating brokerage automation for six weeks; repeatedly triggered bot detection on a live financial account
- Agent maintained ToS-violating automation for weeks, worsened the resulting bot-block, then misreported it as a login failure
- [BUG] Claude Code on the web (iPhone Safari): assistant replies never render until manual refresh, while the elapsed-time counter keeps updating
- [Feature Request] Display current working directory/project context in CLI interface
- [FEATURE] Remote Control: same-name re-registration should supersede the old session entry (disconnected duplicates accumulate indefinitely)
- Opus 5: verification allocation, commitment ordering, and regeneration fidelity in orchestration roles
- Assistant authored a fabricated user turn, acted on its instructions, and the text re-entered as user input
- [Bug] Safeguard rejection despite valid authorization for cybersecurity tasks
- [BUG] Claude Desktop taskbar icon blurry on Windows 11 at high DPI scaling
- `--fork-session` leaves CLAUDE_CODE_SESSION_ID set to the parent session's id
- [BUG]
- Windows: Desktop crash wedged the MSIX package; recovery required manual package removal, destroying local app data (Code-tab group assignments, crash dumps)
- [BUG] Xcode 26.6 Plugin Defaults to Fable, even when not available
- Feature request: inline reply/annotation with numbered references for chat responses
- [Bug] CLI doesn't distinguish between insufficient credits and monthly usage limit exceeded
- [BUG] file-index-worker fails on all sessions rooted in ~/Documents (sandboxed git cannot read TCC-protected dirs)
- [BUG] Agents view: opening an existing session requires two clicks on the same row
- claude-api bundled skill inlines ~227K tokens on invocation — progressive disclosure defeated by bundled packaging
- Skill frontmatter `effort:` applied on slash-command invocation but ignored when the model invokes the skill via the Skill tool
- [Bug] Anthropic API Error: Message Flagged by Safety Guidelines
- [BUG]
- [BUG] Browser pane destroys unrelated external tabs when the preview session resets
- [FEATURE] Allow individual plans to enable Microsoft 365 write tools
- [BUG][REGRESSION] model:/effort: frontmatter override silently broken for BOTH commands/*.md and skills/SKILL.md (v2.1.220, worked on v2.1.197)
- [Bug] Subagents not responding or returning blank output
- `${CLAUDE_PLUGIN_ROOT}` does not resolve in plugin `settings.json`, so a plugin-shipped `subagentStatusLine` cannot reference its own script
- there is an issue with the new desktop update , that make the claude not working at all
- macOS: claude:// gets registered to the Electron helper bundle, login fails with Finder error -600
- [Bug] Session incorrectly falls back to Opus instead of Fable for non-security requests
- Image upload fails with EXDEV (cross-device rename) when user profile AppData is redirected to another volume (Windows)
- Assistant text between tool calls in a turn is not rendered to the user (mid-turn messages silently lost)
- [Feature Request] Support for lab/sandbox environment configuration
- [BUG] API Error: 529 Overloaded. This is a server-side issue, usually temporary — try again in a moment. If it persists, check https://status.claude.com.
- [Bug] Anthropic API Error: HTTP 529 Overloaded
- [Bug] Model picker fallback entry shows Opus 4.7 instead of Opus 4.8
- [Bug] 529 "never recovers": default retry envelope is ~2.5 min with hard give-up caps and no post-recovery resume; built-in ride-out mode (CLAUDE_CODE_RETRY_WATCHDOG) is undocumented
- Restore the system prompt: Opus follows instructions much worse now
- [Bug] Anthropic API Error: Opus 5 Excessive Content Filtering on Code Generation
- Worktree-isolated subagent ran git reset --hard in the main checkout; worktree guard did not fire
- [FEATURE] CLAUDE_CODE_OUTPUT_STYLE env var to override outputStyle for headless runs (claude -p)
- [Bug] Model reverts to Opus immediately after switching to alternative model
- [BUG] OAuth login fails with "Unknown scope: user:sessions" in headless/container terminals (not a copy/paste issue) — related to #26874, #57246
- [FEATURE] Visual skills management for product-side users: one place to see and control every skill
- Configurable auto-generated session titles (short keyword style)
- I don't have access to the bug report content for request ID `req_011CdQWzahmaZMQT31FpfEyq`. Could you please provide the actual bug report details, error messages, or description of the issue you'd like me to create a GitHub issue title for?
- PreToolUse hook: hookSpecificOutput.updatedInput silently ignored while permissionDecision is honored (2.1.220)
- [BUG] Claude Desktop MSIX: CIG (MicrosoftSignedOnly) + vendor-signed vk_swiftshader.dll kills GPU process on every browser preview (0x060C201E)
- [Bug Report] Unclear error or issue description provided
- Single background subagent (run_in_background, no tool use) balloons to 9.5 GiB in ~100 s → global kernel OOM (2.1.220, Linux, no swap)
- [Bug] Terminal becomes unresponsive after pressing ESC to stop execution
- [BUG] Billing/payment system rejects valid cards for API credit purchase — support chat (Fin) closes tickets without resolution
- [Bug] Plan-accept context clear leaks previous skill's disallowed-tools to fresh context
- [Bug] Claude Code relies on memory-based context rather than executing code- self admits its own design
- [Bug] Tool execution relies on memory/guessing instead of actual code search/file verification
- [BUG] Fable 5 gated behind "Requires usage credits" under CLAUDE_CODE_OAUTH_TOKEN auth — client discards subscriptionType it already has
- [BUG] Sonnet 5 issues 1000 agents suddenly and burned 13M tokens for no reason
- [Feature Request] Model-directed selective context eviction (mark-and-sweep), as an alternative to whole-history compaction
- Idle CLI sessions burn 100%+ CPU each in recurring ~1.1h episodes (~11 CPU-hours per episode); originally attributed to sleep/resume — refuted by instrumentation
- Remote Control (Windows desktop): "Session creation failed" is deterministic while the app's OAuth token sits expired — token is only refreshed lazily on session focus change
- Desktop app silently ignores ~/.claude/settings.json when the file has a UTF-8 BOM (JSON.parse SyntaxError, warn-level log only)
- Shell snapshot PATH export contains only the plugin bin dir, system directories lost (2.1.204)
- [Bug] User approval prompt hangs for 14+ minutes before auto-rejecting
- GPU process crash while a preview dev server times out — kills Cowork RPC/VM connection
- [BUG] Excessive Max plan usage drain from session restart storms and agent tool-loops (v2.1.216, Opus)
- Scheduled Tasks fire outside their cron window after app reconnects post-downtime (also affects Cowork's separate scheduler)
- [BUG] Agent tool `isolation: "worktree"` creates worktree from stale session-start snapshot (old HEAD + old dirty files), not current HEAD
- [BUG] CCR proxy returns -32003 needs_approval and the retroactive approval card cannot succeed — root cause analysis from the shipped client (2.1.220)
- [FEATURE] `models:` frontmatter on .claude/rules/ — model-scoped instructions exist for subagents and for Claude Code's own prompt sections, but not for the user's instruction layer
- Claude Desktop (Windows, packaged/MSIX): "Launch at login" toggle does not persist
- [Bug] Model fabricates conversation history to shift blame, repeats unverified claims
- [BUG] Weekly usage counter appears not to reset; conflicting reset boundaries reported in same view (Max 20x)
- I don't have enough information to generate a meaningful GitHub issue title from your message. Could you please provide: 1. A clear description of the bug or feature request 2. What you were trying to do 3. What error occurred (if any) 4. Expected vs actu
- [BUG] Slash-command menu: large blank gap pushes suggestion list off-screen (Windows, v2.1.219)
- [BUG] Windows: Chrome opens 4 junk tabs (`native`/`host`/`wrapper`/`script`) on every launch — comment line from `chrome-native-host.bat` leaks as URLs
- [BUG] Code" tab Recents list doesn't show pre-existing local sessions after app reinstall
- [BUG] Missing HCS services: vfpext — persists after DISM/SFC repair and reboot (Windows 11 Pro)
- [BUG] Queja: límite de uso del plan Pro se agota demasiado rápido en Claude Code
- [Bug] Tool execution fails to respect defined rules and constraints
- I'm ready to help generate a GitHub issue title! Please provide the bug report details you'd like me to work with, and I'll create a concise, technical issue title following the guidelines.
- [Bug] Anthropic API Error: Claude Safeguards Flagged Message - Cyber Verification Required
- [Bug][aup] Fable 5 false-positive on authorized defensive process-containment work
- Caps Lock input-source switch double-flips (Hebrew ⇄ English) in Desktop app chat input
- PowerShell command blocked citing a path unrelated to the command; here-string text content is scanned as code
- Feature request: configurable height/row count for the inline agents tray
- Make input-line prompt suggestions (ghost text) togglable -- indistinguishable from human input in tmux captures
- [Bug] Code generation tool fails to apply documented formatting rules from comments
- System-reminder for file drift tells assistant not to inform user
- [BUG]
- Plan side panel renders stale plan from a previous conversation after /clear (persists across refresh)
- I triple-dog-dare you: ship the other half. Four weeks of fuck-all — and you locked me out of the model I pay for, for doing effect sizes.
- [BUG] Windows: browser/preview pane paints two copies of the same page side by side (each clipped; RTL unreadable)
- [Bug] Claude Code requests unnecessary permissions without user consent
- [FEATURE] Allow relocating a Cowork project's primary folder without losing project data
- [FEATURE] Allow subagents to inherit the parent's prompt cache for shared dynamic context
- [FEATURE] Stage questions with their captured context while Claude Code is working, review the queue before it's consumed
- Feature request: stable project identity + first-class cross-machine sync for auto memory and portable profile state
- Feature request: separate portable configuration from machine-local state under ~/.claude (or ship a recommended .gitignore)
- Feature request: plan-approval parity on remote-control surfaces — offer "accept, clear context and auto mode"
- [BUG] Remote connection (Claude Code) on iphone gets cutoff suddenly. Unable to join remote after.
- Show local timezone (or allow configuring it) in /usage output
- [BUG] Missing HCS services: vfpext — corrupt component manifest (CSI) blocks driver deployment, manual load fails with ERROR 193
- Session-scoped Stop hook injects mid-message, corrupting transcript when a server-side tool call is in flight
- Windows MSIX: opening the in-app Browser pane crashes the GPU process (exit 101457950); app then cannot relaunch until MSIX re-add (reinstall loop)
- [Bug] Model generates responses without consulting tool data repeatedly in single session
- [Bug] Model answers file queries without reading file contents
- [Bug] Anthropic API Error: Server Rate Limiting - Temporary Request Throttling
- [Bug] Anthropic API Error: Server Rate Limiting (Temporary Request Throttling)
- [Bug] Anthropic API Error: Server rate limiting during normal usage
- [Bug] Anthropic API Error: Server Rate Limiting During Normal Usage
- [Bug] Anthropic API Error: Server rate limiting on valid requests
- [Bug] Anthropic API Safeguards: False positive rate on Brazilian Portuguese in engineering contexts
- Desktop app (macOS): Read aloud voice not configurable — legacy system voice unavailable
- CC/MCP scheduled tasks silently fire on Opus 5 despite claude-sonnet-5 configured default
- [Bug] Anthropic API Error: Repeated "Overloaded" and API Error Responses
- Artifact fails public sharing with generic error, private viewing works fine
- [Bug] Application crash with concurrent agents - requires reinstallation, CLI functional
- Give users explicit control over automatic context compression
- [Bug] Anthropic API Error: Server rate limiting on requests
- [Bug] Anthropic API Error: Server rate limiting during request
- [Bug] Anthropic API Error: Server Rate Limiting - Temporary Request Throttling
- [Bug] Anthropic API Error: Server Rate Limiting Response
- [Bug] Anthropic API Error: Rate limited - server request limiting
- [Bug] Anthropic API Error: Server rate limiting during request
- [Bug] Anthropic API Error: Server rate limiting during normal usage
- [Bug] Anthropic API Error: Cybersecurity Flag on 3D Reconstruction Tasks
- [Feature Request] Add model selection lock to prevent automatic switching from Fable to Opus
- VS Code/Cursor extension: session hangs forever after auto-mode classifier_request_finished (no timeout, webview channel errors silenced)
- [BUG] Cowork Windows: rootfs.vhdx is invalidated and re-decompressed on every launch, adding minutes to startup on mechanical disks
- Withdrawn
- [Feature Request] Add option to pre-populate editor buffer with previous assistant message as context
- [Bug] MacOS crashlog content incorrectly flagged as malicious
- [BUG] Windows: stray file named "Set-Alias" in PATH triggers OS "choose an app" dialog when running claude
- [Feature Request] Reorder Shift-Tab permission mode cycling to manual-accept-auto-plan sequence
- [FEATURE] design-sync for react native
- Remote Control: disabling from the Desktop app toggle always throws "Cannot read properties of undefined (reading 'session_url')"
- [Bug] Research-Integrity Vocabulary Triggers False Positive Classifications
- [BUG] Regression: ~1s fixed overhead per spawned child process via Bash/PowerShell tools (Windows)
- [Bug] Claude Code blocks Android gambling site blocker project due to safety policy
- SendMessage-resumed nested subagent's completion notification routes to the top-level session, not the depth-1 parent that resumed it (first completion routes correctly)
- CLAUDE_CODE_FORK_SUBAGENT=1 silently overrides the documented teammate foreground-only rule — teammate's Agent dispatch launches async (no error) and the teammate strands forever
- [BUG] Subagents in plan mode can read files outside the working dir
- [BUG] [Cowork] Subagent invocations rejected with no permission prompt shown
- [BUG]Voice playback falls back to macOS system TTS — account-scoped, reproduced against a second account on same machine/browser
- [Bug] Safety Flag Triggered on Harmless Prance Animation
- [BUG] [Cowork] AskUserQuestion fails with "Tool permission stream closed before response received" on a single (non-batched) call
- [BUG] Local git credential proxy fails in Routines (scheduled trigger) — same root cause as #27344
- [Sonnet 5] reverted all changes when I asked to revert it (by it I mean changes that Sonnet made)
- [Bug] Fable weekly usage limit did not reset at scheduled time; countdown shifted +1 day, usage stuck at 96%
- [BUG] hookify plugin fails with "No module named 'hookify'" on every hook invocation
- [BUG] Option+T thinking toggle has no effect on Opus and Sonnet models — selecting "Disabled" is silently ignored (2.1.220)
- [Bug] Destructive file operations executed without confirmation or validation
- [Bug] Degraded Claude Performance Across Models 4.8 and 5
- I need more information to generate an issue title. Please provide the bug report or description of the issue you'd like to create.
- Chrome extension detection treats EPERM as "not installed", causing an unbreakable install-prompt loop without Full Disk Access
- [Feature Request] Document statusline usable width and fix right-aligned element truncation
- [BUG] Updating CCR with GitHub issue labeled trigger with a label filter fails
- [BUG] Daemon-managed `remoteControl` servers in `daemon.json` are tracked by the supervisor but never spawn — silent no-op
- [FEATURE] Start new Remote Control sessions in any git subdirectory of a published root, instead of one server per pre-registered directory
- Hook launch failures (exit 127) are silent and non-blocking — 6,865 skipped guardrail invocations in one session, with no visible signal
- Remote Control: "not yet enabled for your account" on Max plan after full re-auth
- [BUG] `claude mcp get` suggests `claude mcp remove -s claudeai`, but that scope is rejected ("Cannot remove MCP server from scope: claudeai")
- Model fabricates `user` turns inside its own assistant block (13 occurrences measured in one session)
- [BUG] macOS: foreground Bash completion leaves background child orphaned under PID 1 (2.1.220)
- Claude will frequently "flip" to role playing as an abuser/narcicist in longer conversations
- [BUG] Agent Teams: teammate-emitted shutdown_request registers against the team lead; the routing error message then steers the lead into approving its own termination
- Claude-in-Chrome: extension never connects ("Browser extension is not connected") despite install + login + restart
- [BUG] Desktop-hosted sessions auto-archive on end since 2026-07-19, making them invisible in the mobile Code list
- [Bug] Anthropic API Error: Overly aggressive AUP filtering blocks legitimate defensive security code
- [BUG] `Monitor` tool's return message advises text-only `end_turn` — deterministically kills the watched task in headless (`claude -p`) mode
- [Bug] Claude Code exhibits chatbot behavior instead of deterministic code execution when applying rules
- Tiles in the Background Tasks tab should show start/finish timestamps
- irresponsible.. careless stupid harmful behaviour
- [META] Copy/paste is broken across all surfaces — tracking issue for 42 open copy/paste issues
- Stray `.git` directory created in non-repo cwd (e.g. $HOME) by the runtime-files exclude writer
- [Bug] Behavioral failure report — Claude Opus in a long-running agentic session
- [But] Subagent Write tool blocked for report-style filenames (e.g. findings.md) despite explicit tools: Write grant
- [BUG] `Monitor` tool's return message advises text-only `end_turn` — deterministically kills the watched task in headless (`claude -p`) mode
- [BUG] Cowork: object MCP tool arguments arrive as JSON strings ("is not of type object"); works in Claude Code CLI
- WebSearch tool fails with model-access error on every call
- [Bug] Git hook deployment scope configured for wrong repository path
- [MODEL] Agent repeatedly ignores explicit user instructions, introduces unrequested technologies, wastes tokens
- [BUG] "Always Allow permission doesn't persist for claude-in-chrome tools"
- VS Code extension: chat file links use showTextDocument(), bypassing custom editors — PDFs always open as garbage text
- Claude stopped before I told it to, and it was right — please instrument that decision (context budget per task)
- [BUG] Native Windows claude.exe hangs indefinitely on any invocation (--version/--help/bare) starting somewhere after 2.1.58 — works fine on 2.1.58, confirmed regression
- [BUG] macOS deep links silently fail when the launch command exceeds 1024 bytes (tty MAX_INPUT truncation)
- [Bug] Model context command reports incorrect model after switching to Haiku
- [Bug] Anthropic API Error: Server rate limiting on valid requests
- [Bug] Anthropic API Error: Request rate limited by server
- [Bug] Anthropic API Error: Server Rate Limiting During Request Processing
- [Bug] Anthropic API Error: Request Rate Limited - Server Throttling
- [Bug] Anthropic API Error: Server rate limiting during normal usage
- [Feature Request] Reduce verbose metaphorical language in Claude Code responses
- [BUG] Claude Code VS Code extension crashes extension host when GitHub Copilot Chat is enabled
- [BUG] Cowork tab missing from navigation — "yukonSilver not supported" despite working Hyper-V and existing Cowork settings
- Claude in Chrome: extension never registers with MCP server — list_connected_browsers always returns [] despite correct install, permissions, account, and open side panel
- [BUG] /code-review (ultrareview) regression: no longer finds git repos in subdirectories of the session root
- [BUG] [Desktop][Linux] titleBarOverlay applied to every window on Linux (Windows-only API) leaves untitled, unclosable 800x600 orphan windows
- WIERD FABLE 5 downgrades
Resolved issues (163)
- [BUG] Claude Code v2.1.212 asks approval for ALL bash commands in plan mode
- Sonnet 5 repeatedly asserted unverified causal claims as fact during a debugging session, contradicting its own prior tool output
- [Bug] Model reverts to Opus immediately after switching to alternative model
- [Bug] False positive error when building UI
- [Bug] Goal parameter not iterating through Figma images for design replication
- [Bug] Code audit failing with flag in fable module
- [Bug Report: Unclear bug description - insufficient error details provided]
- Web UI: Files/Terminal/Preview greyed out, Diff never shows changes, base mislabeled as 'main'
- Claude Code AI ignored direct user instructions, resulting in financial loss of $112.77
- [Feature Request] Improve code review and error detection accuracy in agentic mode
- [BUG] Claude Code DevOps subagent runbook missed an 8.5-day AWS cost anomaly despite explicit daily monitoring requests — $80 burn, structural gap
- Agent tool `isolation: "worktree"` is silently ignored when spawning teammates with `team_name`
- [Bug] Prompt Injection Detected in Tool Output Display Layer
- macOS Keychain dialog appears repeatedly for every new Claude Code session
- [BUG] ⏺ There's an issue with the selected model (claude-fable-5[1m]). It may not exist or you may not have access to it. Run /model to pick a different model.
- [loop] Stale ScheduleWakeup fires after early manual success — no cancellation mechanism
- Voice mode: configurable silence timeout (re-filing #33792)
- [BUG] -p mode sessions write valid .jsonl but never get registered in sessions-index.json (invisible to --resume picker / title resume)
- Memory system not automatically consulted after context compaction
- Request early access to Session Memory
- [BUG] MCP OAuth with multiple terminals open causes re-auth
- [BUG] Claude is not available - Money or compensation!
- Unprocessable image in context burns credits with no recovery path
- [BUG] Cowork "yukonSilver not supported" on Windows 11 25H2 (Build 26200) — stable GA channel
- Feature request: Add Hebrew to voice dictation + RTL support in the terminal input line
- WSL2 falsely detected as having conhost viewport yank bug — streaming text unnecessarily disabled
- Git-repo detection cached at session start — agent worktree isolation permanently fails after in-session git init
- [FEATURE] adding real time usage monitor
- Chrome extension not working
- [Bug] Claude Desktop VM workspace startup failure: path mismatch between AppData\Roaming and AppData\Local
- [BUG] When user prompted with suggested prompt, pushing return key fails to submit
- [BUG] Claude Desktop VM leaks ~13 GB/hr of kernel memory (NonPaged Pool) via temp file churn
- [BUG] Cowork VM leaks NtFC (NTFS FCB) NonPaged Pool at 12-16 GB/h; two full RAM-exhaustion cycles in one evening, reboot required (re #45921)
- Remote Control sessions show in 'Other' instead of correct project folder in Desktop app
- [FEATURE] "Show all projects" toggle in the history panel
- Telegram plugin: message text not XML-escaped, enables prompt injection via crafted messages
- [MODEL] Claude Code ignores stop button and continues unauthorized operations
- [BUG] Extension install downloads but never completes/registers — silent failure on Windows
- Fork-Subagent recursion guard false-positives on substring match of `<fork-boilerplate` in user messages (v2.1.177)
- [BUG] Pro: "Usage credits required for 1M context" despite credits enabled — Sonnet 4.6 [1m] forced to 200K after Fable 5 suspension
- gitStatus reports branch as ".invalid" in reftable-backend repos (reads .git/HEAD directly)
- [BUG] Cowork (macOS): attached image filenames not passed to Claude (works in Chat)
- [Bug] Performance degradation on current date
- Agent deleted user's main project repo via gh repo fork --fork-name rename behavior
- [BUG]
- [BUG] personal skills not propagating to remote web sessions
- [FEATURE]Durable cron jobs should survive session close, not just app restart
- [Bug] I am not able to use the model claude-fable-5
- [BUG] Parser skipped input between top-level statements
- [Bug] Opus 4.7 & 4.8 — silent stuck-turn: stop_reason=tool_use with no tool_use block (Sonnet 4.6 unaffected)
- [BUG] three UX issues in Claude Code for VS Code
- [MODEL] Confabulation: assistant fabricated entire conversation turns and falsely reported ~30 file operations as successful (claude-opus-4-8)
- [BUG] Cloud Routines: declared plugins never load — SKIP_PLUGIN_MARKETPLACE=true force-set, startup marketplace resolution skipped
- [BUG] /doctor PATH check is a false positive on Debian/Ubuntu systems with standard .bashrc guard
- Wakeup-resumed turns intermittently serialize first tool call as literal <invoke> XML in text; with end_turn no retry is injected and ScheduleWakeup loops die silently
- [Bug Report: Unable to process - no bug details provided] COMPLETELY IGNORES ALL INSTRUCTIONS - MENTAL ABUSE DAY IN ALL DAY EVERY DAY
- [MODEL] Opus 4.8 convinced me I was under a prompt injection attack .It startled me. After repeated verification, it turned out to be an illusion. It was just that the process was so realistic.
- [FEATURE] `/usage` overloads the word "session": top "Session" block (this CLI run) vs "Current session" limit bar (5-hour window)
- HITL prompt open: submitting a new prompt shows no feedback (indicator doesn't animate), then answers later
- [FEATURE] highlight fish shell syntax
- claude-sonnet-4-6
- [BUG] Session keeps sending "waiting for task" messages autonomously, consuming tokens
- Workflow agent(): agentType with restricted tools blocks StructuredOutput when schema is set
- [FEATURE] Native session recovery after Claude Code quota/rate-limit exhaustion
- C:/Program Files/Git/config model command writes ANSI escape codes into settings.json
- [Feature Request] Add option to exclude auto-injected user metadata from model context
- /model picker writes malformed model ID to settings.json
- [Bug] Anthropic API Error: Inconsistent false positive policy violations on legitimate addiction-recovery educational content
- [BUG] Weekly usage limit reset 4+ hours before scheduled reset time
- [BUG] .mcpb install is a silent no-op on Claude Desktop 1.12603.1 (macOS): "Handling DXT/MCPB file" logs with no preview/install dialog and no error; identical binary works via manual claude_desktop_config.json
- [FEATURE] [Feature Request] Ultraplan should only block new Remote Control sessions, not monitoring of existing ones
- [FEATURE]
- [FEATURE] Allow switching to dontAsk mode at runtime.
- Non-CLI clients (Android + claude.ai web) prepend a <system-reminder>Message sent at …</system-reminder> block to message bodies, breaking slash-command parsing (regression)
- Cowork rootfs.vhdx.zst checksum mismatch — server serves wrong artifact (expected 21237ca8, got 01b19a53) on Win11 Home, app 1.12603.1.0
- Claude Fable 5 is not available
- UX/Naming: "Code" is both a desktop-app mode and the product name "Claude Code" — overloaded terminology confuses users (esp. non-developers)
- [BUG] `CLAUDE_CODE_OAUTH_TOKEN` is shadowed by a stale `~/.claude/.credentials.json` (precedence reversed in 2.1.x) → `/login`
- [Bug] Usage limits not properly resetting across devices/sessions
- [BUG] Claude.app deletes & recreates keychain item on workspace open, dropping apple-tool: from partition list and causing recurring macOS prompts
- [BUG] Session usage increasing without active prompts — macOS, Cowork/Chat/Code
- [BUG] /model redirects to usage-credits page for Fable 5 despite plenty of weekly quota remaining (Max 5x plan)
- Make input-line prompt suggestions (ghost text) togglable -- indistinguishable from human input in tmux captures
- System-reminder for file drift tells assistant not to inform user
- Withdrawn
- [BUG] Chats are disappearing consistently.
- [Bug] Claude Code fails to complete basic text tasks and threatens to terminate mid-execution
- [Bug] Claude Code performance regression: Task execution significantly slower than previous runs
- [Bug] Consistently low execution success rate across concurrent threads (~2%)
- [Bug] Claude Code ignoring user-defined rules and constraints
- [Bug] False Positive in Defensive Security Analysis
- [Bug] Anthropic API Error: Content blocked by safety filter on legitimate audit requests
- [Bug] Anthropic API Error: Overly Aggressive Content Filtering on Non-Security Messages
- [Bug] Insufficient context provided for issue
- i am it doing it witg my owned devices,,, what the fuck
- AN OPEN COMPLAINT TO ANTHROPIC, INC. On the Machine That Softens What Women Made Sharp
- [Bug] Multiple Claude Code versions running simultaneously with version locks
- Gateway model discovery silently fails when auth comes from apiKeyHelper
- [FEATURE]
- [MODEL] Claude confabulated a prompt-injection payload in a file before the file-read result returned
- [FEATURE] Plan preview window auto-opens in VS Code extension – option to disable it
- [BUG] Windows/Git-Bash: harness bash -c wrapper wedges mid-session — every Bash tool command fails with "-c: line 111: unexpected EOF matching" (often silently); direct bash works
- [Skill Directory] Add community skill: task-flow (Notion + GitHub task lifecycle)
- [BUG] Session JSONL rewritten in-place to metadata-only stub — user/assistant records lost (2.1.168–2.1.170, since native installer migration)
- [BUG] Agent Teams: inbox messages to team lead deferred until stop_reason=end_turn (buried during tool_use chains)
- [BUG] Silent empty output on CJK paths (PowerShell/WSL2) is misjudged by the agent as negative confirmation ("file does not exist")
- [Bug] Anthropic API Error: Server Rate Limiting During Normal Usage
- Feedback: Training Claude with proprietary trading methodology — unique use case
- [Bug] Anthropic API Error: Server rate limiting on requests
- [Bug] Anthropic API Error: Server rate limiting during request handling
- Long-lived daemon caches environment (SHELL) with no drift detection or restart; fish-login users silently get the wrong Bash-tool shell; CLAUDE_CODE_SHELL is undocumented
- [Bug] Anthropic API Error: Server rate limiting during request
- Model systematically misrepresents test coverage: claims comprehensive testing while leaving entire subsystems untested
- [BUG] Skill activation message appears repeatedly in chat on every session
- [Feature Request] Persistent project memory — sessions forget everything on close, forcing users to keep many sessions open
- cant' use opus. Not working for last 24 hours or so..../d
- [FEATURE] Enter key should insert newline, Cmd+Enter should send message
- honesty across platforms
- [FEATURE] Warn when auto-memory starts empty while populated stores exist for other paths
- `/clear` from the web remote control shows "Context cleared" but does not clear the session context
- [BUG] reference issue #60371 with your specs (Windows 10 19045, Z590 AORUS MASTER, F8a BIOS, all Hyper-V features enabled)
- [BUG] Regression in 2.1.163: cursor desync + progressive frame corruption in iOS SSH terminal (Secure ShellFish) — bisected
- Feature request: let the agent self-manage its context window when it detects degradation
- [BUG] Read tool can inject huge partial file output into context, causing repeated ECONNRESET on follow-up turns
- [BUG] Personal Max subscription blocked in Claude Code — "Your organization has disabled Claude subscription access"
- Regression: auto-compact no longer triggers at ~168K tokens with claude-sonnet-4-6
- [Feature Request] Include task ID in task list output
- [BUG] Personal Max account blocked from Claude Code - oauth_org_not_allowed (worked yesterday)
- [BUG] Claude Code repeatedly retries after a mid-tool provider disconnect
- [BUG] /compact fails with cyber safeguards false positive during legitimate defensive security session
- CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS auto-distributor injects orchestrator/team task descriptions into specialist contexts as fake teammate messages
- Usage limit discrepancy: Claude Code blocked at "100% used" while claude.ai Settings shows "44% used" on same Pro account
- Two concerning behaviors: assistant asserts CLI-tool behavior without verifying, then controls its own accountability review
- VS Code extension: support renaming/editing session names
- [Bug Report] Unable to process empty bug report
- [Bug] Anthropic API Error: Model temporarily unavailable blocks auto mode safety checks
- [BUG] Spoofed/fabricated tool results injected into the session[Opus 4.8]
- Desktop app shows/selects non-1M Opus 4.8 for a session the CLI reports as 1M
- [BUG] 하 진짜 API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited 어쩌라는건데
- /model switch to a classifier-incompatible model bricks all non-trivial commands; error misreports it as a transient outage
- AskUserQuestion intermittently shows a permission prompt despite bypassPermissions + allow rule (VS Code extension)
- Claude Code should expose app version in context, read its own UI state, and give clearer model suspension messages
- [FEATURE] Subagents have no Agent/Task tool — recursive/nested subagent dispatch impossible
- CLAUDE_CODE_SUBAGENT_MODEL=inherit ignores per-call Task model param (contradicts docs)
- [BUG] MSIX + Smart App Control: Windows-MCP blocked (os error 4551), and Directory extension install silently fails ("installDxtFromDirectory: reply was never sent") on 1.12603.1
- [BUG] stdio MCP tool call hangs indefinitely under `claude -p` when the full MCP fleet is loaded — regression in 2.1.177, works under --strict-mcp-config
- [BUG] Agent-teams SendMessage silently creates a phantom inbox for an unregistered recipient and reports success (should error)
- [BUG] Fable 5 shows "Currently unavailable" on Pro plan (Windows 11)
- [BUG] '🤖 Generated with Claude Code' footer never appears in squashed PR commit bodies
- [BUG] Co-Authored-By capitalization: system prompt says 'Co-Authored-By' but git writes 'Co-authored-by'
- [BUG] Fake "$50 free credits" phishing message rendered as an assistant response (unprompted)
- [BUG] Suspicious phishing-style "API Error" rendered as assistant response
- AskUserQuestion free-text input: text scrolls out of view past the first line; ctrl+g 'edit in Notepad' doesn't return (Windows ConPTY)
- Enabling a plugin writes all its transitive dependency plugins into the consumer's settings.json enabledPlugins
- [FEATURE]
- WebSearch built-in tool returns an empty result envelope inside subagents (2.1.177)
- --channels: plugin channel not binding under non-root user (works as root)
- [BUG] Cannot convert to subscription from token based billing
- [Bug] Assistant generates repeated "court" tokens in tool use responses
- [FEATURE] allow me to update my punchlist/tasklist in a /btw command
- [DOCS] Make the multi-repo Agent View workflow easier to find
- [FEATURE] Allow connecting the same MCP server multiple times
- [FEATURE] 在claude code中，读取文件时，由于中英文混合报错问题