Digest / August 28, 2026
Claude Code Issues Digest — August 28, 2026
249 new issues filed ·
155 resolved in anthropics/claude-code (UTC day).
New issues (249)
- [BUG][Windows] npm-global auto-update leaks a ~240 MB claude.exe.old per failure and can leave no claude.exe at all (update_apply_exe_locked)
- [Bug] Review Pressure Causes Truth Degradation in Autonomous Workflows
- API edge RSTs /v1/messages when time-to-first-header exceeds ~15s — large cold prompts (>~90K tokens) fail deterministically through all retries
- apiKeyHelper written to fix an unrelated MCP key, silently overriding claude.ai auth
- [Bug] Unexpected flag triggered without explicit command invocation
- [BUG] Claude Code on the web: Claude Design handoff session can push a branch but has no GitHub MCP tools, so it cannot open the PR for the branch it just pushed
- You've hit your session limit · resets 1:40pm (Pacific/Auckland)
- [BUG] SSH Remote: "NO_HOME" error on QNAP NAS despite correct $HOME, /etc/passwd, and SFTP pwd
- [BUG] computer-use takeover minimizes all open windows on a dual-monitor Windows machine, every time, survives a full restart; reopened windows then auto-close within 2-3 seconds.
- Feature request: lightweight 👍/👎 acknowledgment on assistant messages (not just model feedback)
- Stale Remote Control pairings (800+) truncate cross-session reachability scan
- Doesn't work anymore on VS Code version1.135
- [BUG] CCD desktop app rejects MCP tools with draft-07 outputSchema that work fine in CLI
- [Feature Request] Display model name in terminal CLI inline agent status view
- [Feature Request] Add graceful restart mechanism for agent teams without losing connections
- Scheduled tasks fail silently when the app is closed; no signal distinguishing "running on schedule" from "only running while app is open"
- [BUG] JSON file upload to claude code has broken claude code
- GPU process crashes recurring since v2.1.237, still present in v2.1.246 (Windows, hybrid NVIDIA/Intel graphics laptop)
- Scheduled-task run sessions flood the session list - group, auto-archive, or bulk-archive them
- [Feature Request] Add language/dialect consistency mode to prevent unintended English variant drift
- [BUG] In the desktop application's code feature, the Rewind function does not undo modified code.
- Claude in Chrome (v1.0.85): phantom blank tabs "http://native/", "http://host/", "http://wrapper/", "http://script/" open on every Chrome startup
- Cloud/remote sessions push unsigned commits, blocking repos that require signed commits
- [BUG] Nested subagent (depth-2) completion notifications route to root session and are discarded on 2.1.250 — regression of #75043 item 2
- claude-in-chrome: one-way channel — extension live-renders the same session that gets [] from list_connected_browsers
- [BUG] Authentication using IAM Identity Center credentials is failing
- Custom (user-created) skill from a plugin intermittently disappears from slash-command availability
- [Bug] Database operation performed without confirmation on unverified environment
- [Feature Request] Add hint text to auto-generated feedback box indicating `/feedback` command
- [Bug] Anthropic API Error: Unsafe content flagged on multi-tool workflows with external integrations
- [FEATURE] artifact-design: add a "Glossary sidebar" pattern (collapsible, category-filterable, main-column-only scroll)
- Multi-agent session: background-task orphaning, cross-session message holds, and subagent drift (field report)
- Trusted Devices fails open on individual accounts: revoking devices doesn't re-verify an active session, and declining "Sign in again to verify your device" has no effect either
- [FEATURE] Let Pro and Max accounts require device verification before Remote Control attachment
- [BUG] macOS: opening the agents view (left arrow) fires a "Data Access Blocked" notification every time, attributed to "2.1.250" instead of Claude Code
- Runaway repetition: a single response repeats one line ~21,000 times and consumes the whole 64k output budget (effort high/max + thinking rarely used)
- [BUG] Windows/MSIX: update installs successfully, then Claude launch fails with AppModel-Runtime 215/208 (0x80070020)
- Cowork plugin page shows "Version 1" instead of the version from plugin.json
- Plugin-provided MCP connectors (.mcp.json) cannot be selected when Claude Desktop is signed in via AWS Bedrock
- [BUG] Desktop app OOM on /compact with a much smaller transcript (~40MB) than previously reported crash threshold
- Terminal pane unavailable for a session started in an external terminal and resumed in the desktop app
- [Bug] Anthropic API Error: Server Rate Limiting - Request Throttling
- [Bug] Anthropic API Error: Server rate limiting requests
- [Bug] Agent behavior not complying with agents.md specifications
- [Bug] Claude ignores user observation of persistent bug, continues explaining theoretical fixes instead of investigating running system
- [Bug] Anthropic API Error: Message flagged by safety filters
- [Bug] Reduced context window or intelligence degradation in Claude Code
- [Bug] False positive [cyber] safeguard triggered on benign developer-tooling questions
- Plugin install fails with "EXDEV: cross-device link not permitted" (Windows desktop app)
- [Bug] Plan-mode reminders fired unprompted without EnterPlanMode call (Session: 17e11dbd-649e-47a5-854d-86bc86f363ee)
- [BUG] Cowork "RPC pipe closed" on Windows 11 — shipped claude.exe fails Authenticode check (HashMismatch), reinstall path disabled
- [FEATURE] Let the user open the Browser pane directly — toggle is inert until Claude creates a browser session
- [Feature Request] Provide incentives for users contributing chat data to model improvement
- VSCode extension: image paste requires Shift+Cmd+V — plain Cmd+V should work like other GUI apps
- [FEATURE] Full two-way voice mode (speech in and out) for Cowork
- [BUG] Cowork remote sandboxes omit user.email / user.account_uuid / organization.id from OTel; local desktop sends them
- [BUG] Shared Claude Code artifact returns 404 for a viewer with explicit access
- [FEATURE] "Copy link to session" in Desktop: a native claude:// link that opens a specific chat
- [Bug] `timeout` command not found on macOS causing exit 127 errors
- [BUG] validate-hook-schema.sh fails on plugin hook manifests and non-tool hooks (Fix included)
- Worktree-isolated sessions: benign read-only command shapes refused as "too complex to verify"; suggested remediation cannot apply to loops
- [BUG] Sidebar fails to load when active editor is a SAP ADT virtual document
- [Bug] Anthropic API Error: Content Policy Violation with Claude 3.5 Opus
- UserPromptSubmit hooks: two silent failure modes — mid-turn submissions never fire the hook, and additionalContext intermittently never reaches the context
- [Bug] Anthropic API Error: Safeguard flags blocking authorized security testing for CVP user
- [FEATURE] No way to verify which org/account a `claude setup-token` token is scoped to, after generation
- fullscreen sticky prompt header stopped rendering in 2.1.247 (regression: memoized isSticky() never re-evaluates)
- [Feature Request] Add approval requirement for merge operations in auto mode
- [Security] No sanctioned channel to hand Claude a secret — a map of 18 open requests and the minimal primitive that unblocks them
- [Feature Request] Add internal search functionality for conversation history
- [BUG] Desktop sidebar: "Show N more" is one-way — no "Show less" and the reveal count never decreases
- Let Channels / Remote Control target an existing Claude Desktop conversation (not just Telegram/Discord/iMessage)
- [FEATURE] Browser pane: allow a persistent grant for interaction on an approved private-network host, not just for reading
- Chrome extension not detected when no 'Default' Chrome profile exists (Linux)
- Worktree-isolated background sessions: command shape verifier refuses safe read-only commands at scale (0/100 protective in sampled refusals)
- [Bug] Cyber Safety System False Positives on Harmless Input
- [BUG]
- BUG] Cowork task cannot be linked to computer — "This task can't be linked to this computer right now" (server-side pairing state stuck)
- [Bug] Unclear error message: [bio] blocking Linux embedded firmware development
- [Bug] Tool result timeout causes model to assume instead of defer to user
- I don't see any image attached to your message. Could you please share the bug report or image you'd like me to create a GitHub issue title for? Once you provide the details, I'll generate a concise, technical issue title following the specified format.
- [BUG] /btw side question ends silently after a short preamble when the model attempts a tool call — no answer, no notice, and the truncated turn poisons subsequent side questions
- Windows native CLI 2.1.250 launches embedded Bun instead of Claude Code
- [Bug] Anthropic API Error: Cyber Verification Program approval not recognized for organization
- Feature request: optional UI tooltips / learning mode for non-obvious interactions (split view, rewind, pickers)
- [BUG] ralph-wiggum stop hook: all four state-file error handlers are unreachable
- [Bug] Generated code fixes lack sufficient context and introduce new bugs
- /feedback review screen defaults "Send transcript: yes" with no visible control to change it
- [Bug] Claude Code generated untested service version causing failed deployment and wasted session tokens
- [FEATURE] Feature request: opt-in global memory shared across projects
- Auto-compact should defer to end of response/turn, not trigger mid-response
- Desktop model picker shows two identical "Opus 5" rows (1M-context label dropped)
- Desktop model picker shows two identical "Opus 5" rows (1M-context label dropped)
- Malware Read-reminder still causes refusals (2026-08-28) — and in unattended cloud agent seats the refusal is unrecoverable
- [Bug] Safeguards blocking legitimate code generation without security violations
- [Bug] Unexpected model downgrade to Opus 4.8 for non-restricted use case
- Desktop app: per-workspace plugin store syncs on schedule but never downloads (silent, every indicator green)
- [Bug] Claude Code making autonomous decisions without user approval
- [BUG] Desktop app: clicking a task notification opens the Code session but the Chat/Code tab selector stays on Chat
- [BUG] --resume sessions register the agent-team hub under a stale placeholder session id — teammate replies are consumed but never delivered
- Feature request: Localize plan-approval dialog UI text (Accept this plan? dialog)
- Desktop app: pinned session keeps flashing activity indicator with no active work
- [Bug] CLI misinterprets user commands instead of executing literal requests
- Markdown file links with spaces or umlauts in the path are not clickable (file not found)
- [Bug] False positive cybersecurity check blocking legitimate operations
- No shared identifier between a session's identity (ListAgents/SendMessage name) and its title (what the UI actually shows)
- [Bug] Claude Code fails to authenticate to localhost after extended sessions
- [BUG]
- [Bug] Remote Control auto-enables without user consent on new sessions
- [BUG] total_tokens system message is appended to context after EVERY request
- [BUG] - Profile with annotation @ in Claude Code
- Voice dictation deletes/rewrites previous word when switching language mid-sentence (PT/EN code-switching)
- [FEATURE] /rename + Tab key => insert current session name
- Model volunteered advice to merge a PR while automated review comments were still unresolved
- Feature request: per-agent auto-compact window (separate boundaries for coordinator session vs subagents)
- Usage limit can stop a session mid-operation and leave a half-applied state
- Feature request: let a session start a new named session in the desktop app (spawn a peer, not a subagent)
- Several issues with Caude code harness: Too many conflicting , text based governance artifacts and an unreliable hook mechanism
- WSL remote: CLI download livelocks on slow connections (180s idle timeout, no fetch progress, concurrent installs delete each other's temp files)
- I need more context to help you. Your message doesn't include a bug report or error message for Claude Code. Please share: - The error message or unexpected behavior you're experiencing - Steps to reproduce the issue - What you were trying to do when the
- [BUG] Claude Desktop (Windows 11, MSIX 1.37937.3.0): clicking a link inside the app kills it instantly - reproducible 0x80000003 assert at Claude.exe+0x6E89D89, zero third-party DLLs in the process
- [BUG] Claude Desktop Linux arm64: Code tab fails with "No path to Claude code executable" — downloader opens the version directory itself (EINVAL)
- [BUG] sandbox.network.strictAllowlist is silently discarded from project settings — no warning at startup, in --debug, /status, /sandbox, or doctor
- [BUG] Desktop: Auto-fix CI checkbox silently no-ops for PRs not created through the app
- [BUG] Run button echoes a corrupted command: input line overflow wraps to column 0 and overwrites the start instead of wrapping to a new row
- [Feature Request] Restore prompt navigation/jump-to-prompt functionality in CLI
- [BUG] [Claude Desktop / Linux] Cowork & Dispatch hang on infinite spinner — app opens its own storage directories with O_DIRECT, ext4 returns EINVAL
- [Bug] Overly restrictive safety guards blocking legitimate coding tasks
- [BUG]
- MCP servers (context7, playwright, pdf-viewer) stuck in respawn loop, old processes never cleaned up -- 58+ orphaned node.exe over ~13h (Windows)
- [BUG] "an image in the conversation could not be processed and was removed" is NOT removed from the transcript - poisons the prompt cache, every later turn billed as cache-creation (~12x), drains a session in ~3 turns [root cause + fix identified]
- [Bug] Claude defaults to shell commands instead of file writing tools despite instructions
- [Bug] Anthropic API Error: Request blocked for device unlock assistance
- [Bug] Claude Code deviating from specified requirements
- [Bug] Tool reports completion prematurely, leaving incomplete file state
- [Bug] ESC key does not close `/status` and `/mcp` popup windows, blocking session interaction
- [Bug] Documentation and skill writing requests consuming tokens without completion
- MCP OAuth: redirect_uri uses localhost:<port> while metadata publishes portless localhost and 127.0.0.1 (still failing in 2.1.246)
- [BUG]Session retention cleanup deletes transcripts well before documented 30-day default (VS Code extension)
- [Bug] Model retries failing MCP tool instead of switching strategy after repeated failures
- [BUG] macOS: Desktop-hosted Claude Code is spawned via Helpers/disclaimer, so all TCC-protected folders return EPERM — unfixable by Full Disk Access, and no FDA was ever granted
- Embedded browser pane repeatedly crashes/freezes due to GPU process crashes (AMD Radeon 780M)
- [Bug] Session recovery failure preventing website restoration
- [BUG] Inscrutable copy shown in the UI
- Worktree-isolation guard: string-executor check matches every argument position, case-insensitively
- [BUG] Small blank-line gap appears above the input box in VS Code integrated terminal (Windows), persists across updates
- [FEATURE] Advisory notes channel for background watchers — system-attributed context, not chat messages
- [Bug] Legitimate requests blocked by overly strict content filter for term "cyber"
- [FEATURE] Task-scoped rules: load rule files by task type, not only by path
- [FEATURE] Effect-based approval for Bash-mediated writes: run against staging, approve the computed diff
- [Bug] Atlassian MCP: 403 Forbidden on transitionJiraIssue and addCommentToJiraIssue
- No conecta chrome con claude
- Chrome extension reconnect fails intermittently even after fixing native-messaging host conflict (macOS)
- Add a way to disable whimsical/playful spinner status words
- [BUG] Remote Control sessions are archived on teardown and never unarchived, making them permanently invisible on mobile after a desktop auto-update
- [Bug] Prompt cache not persisting across API plan switches within session
- Windows desktop app crashes with "Another program is currently using this file" under high concurrent-session memory pressure
- [Bug] Claude Code performance degradation with account CVP participation status
- [Bug] CVP Account Coding Sessions Blocked During Website Editing
- [Bug] Incomplete user messages appearing in feed
- [Bug] Unexpected high API usage consumption rate
- Store per-project memory and settings inside the project directory
- Artifacts deleted server-side with no user action, no retention setting, and no notice (personal plan)
- [Bug] CLI commands /model and /resume have increased latency (40+ seconds) since v2.1.230
- [FEATURE] Cache-eligible file reads: allow declared knowledge paths to be treated as stable context anchors
- Diff panel keeps showing an edit after a later edit in the same session reverts it
- [BUG] Ctrl+C does not interrupt `claude auth login`'s code-paste prompt (raw mode swallows SIGINT)
- [Feature Request] Add session isolation and permission gating to prevent cross-session state leakage
- RECENT UI DOWNGRADE ISSUE
- [FEATURE] Sidebar: order the chats/tasks under an expanded pinned Project by last activity (or add a sort control)
- Built-in /resume silently shadows user custom command (~/.claude/commands/resume.md) — undocumented precedence, no disambiguation
- Sessions started with --channels write no transcript to ~/.claude/projects/
- Model emits cwd-relative file links for files it edited OUTSIDE cwd (git worktree sessions). In full worktrees these resolve silently to a stale copy at a different commit.
- Auto-compact fires at ~1M tokens on large-context models (~24% of window) — silently, with no user notification
- AskUserQuestion reported an option the user did not select; assistant acted on it and started Docker containers
- [BUG] VS Code extension ignores project and local settings scope for model, and lets user scope override both scopes for effortLevel (Desktop app unaffected)
- [BUG] CLAUDE.md rules violation (no /compact was used whatsoever!!!)
- [BUG] Windows MSIX: staged update leaves old instance running from WindowsApps\Deleted for 15h — 0x12C cleanup loop crashes it repeatedly, then servicing force-terminates it mid-session
- [BUG] Claude Desktop installation fails - Administrator access required (Cowork) on Windows 11 Pro Build 26100
- [BUG] macOS: Local Network permission is lost on every auto-update (versioned claude.app bundle path); 20+ stale permission rows accumulate
- [BUG] Unauthorized "Gift Max 5X" invoice generated — never purchased (EBMIFEW5-0002)
- [FEATURE] claude-in-firefox
- Claude Browser action confirmations ignore permissions.allow and defaultMode: bypassPermissions
- WebFetch passes raw PDF bytes to its summarizer with no text extraction and no "unreadable content" signal (still in 2.1.250)
- Cross-session SendMessage: messages not delivered, or echoed back to sender mislabeled as inbound
- Windows desktop app crashes when Browser pane opens a Cloudflare-challenged site (preview_start)
- [Bug] In-session effort level changes not persisting when CLAUDE_CODE_EFFORT_LEVEL env var set
- [BUG] Plugin update/reinstall never refreshes installed cache for private git marketplaces — panel shows new version, but old code still runs
- [BUG] Shell snapshot silently truncated at ~7.2KB on Windows (Desktop) - every Bash call fails with unexpected EOF
- [Bug] Inappropriate content flag on "Cyber" terminology
- Desktop app session list can't adopt existing local CLI sessions — sidebar left empty after Max→Teams account migration
- [BUG] Windows desktop: "Open in new window" is permanently dead for a session after its pop-out window closes abnormally — stale entries in session-popouts.json
- Desktop app collapses assistant text written between tool calls; substantive answers silently hidden under 'Ran N commands'
- [BUG] File names used by Claude wreak havoc with XFence security
- [Feature Request] Make 0 == No, 9 == never - Reorder permission prompt options to prevent accidental acceptance
- Session title desync: iOS app shows an unrelated title for a renamed session
- [Feature Request] Keep the status line (usage percentage) visible while the AskUserQuestion picker is open
- [Bug] Anthropic API Error: Content Flagged - Request Blocked
- I need more information to generate an appropriate GitHub issue title. "chk log" is too vague. Could you provide: - What error or problem occurred? - What were you trying to do? - What is the full error message or log output? - Is this a bug or feature r
- Worktree isolation verifies complexity, not targets: a simple Bash command writing an absolute path outside the worktree passes unchecked
- [BUG] Per-device session visibility — sidebar leaks session titles across machines on one account
- Improve Claude CLI Session Visibility and Model Switching
- Feature request: UI localization / i18n support (non-English languages)
- FileChanged hook not triggered by real file deletion (unlink) outside session cwd on Windows, despite correct settings.json config and watchPaths from SessionStart
- [Bug] Whitespace indentation breaks copy-paste of commands and URLs
- [BUG] Windows: .claude.json accumulates project keys differing only in case, making the file unparseable and silently hiding MCP servers
- [BUG] Windows/PowerShell 5.1: bare -- in 'claude mcp add' is stripped by the shell, producing a misleading "missing required argument 'commandOrUrl'"
- [BUG]
- [BUG] command autocomplete only triggers at position 0, while @ triggers mid-prompt
- Prompt too long
- [Bug] Subagent polling wake-ups consume excessive tokens for no-op status checkssu
- [Bug] Overactive content policy safeguards triggering on benign requests
- [Feature Request] Use alternative keybinding instead of Ctrl+C for agent termination to avoid conflict with terminal text clearing
- Plugin PreToolUse hook matcher removed from cache/hooks.json still fires after full restart
- [Bug] Token consumption mismatch: Incorrect model billed for API calls
- Warning 30 seconds before context compaction
- [BUG] Subdirectory `CLAUDE.md` loads only via the `Read` tool — every other file access path silently skips it
- [BUG] Auto Mode's Bash-first instruction silently disables nested CLAUDE.md and path-scoped rules
- Feature request: optional timestamps on assistant responses in long-running sessions
- [BUG] Desktop app memory usage grows to 3+ GB / 75-80% of system RAM, and returns immediately after a full restart
- [DOCS] agent-teams Limitations: /rewind does preserve in-process teammates, contradicting the documented claim
- [Bug] Unauthorized financial transaction occurred
- SendUserFile: file card reports success but recipient cannot open the file (reproducible)
- Worktree provisioning writes Git LFS hooks to a literal dev/null/ directory instead of .git/hooks/
- [FEATURE] Compress API request bodies with gzip by default
- [Bug] False positive security flag on project files
- [Bug] Inline LaTeX/MathJax rendering not supported, only block-level math
- [BUG] Desktop app: typing a slash command for a user-level skill (~/.claude/skills) does nothing, while the Skill tool loads the same skill fine
- Windows: Browser pane GPU-process crash on Cloudflare Turnstile page kills entire app; session resume re-triggers it (crash loop)
- [BUG] Files sent via SendUserFile are not rendered in the VS Code extension panel (work on mobile / web)
- Reply to SendMessage from a nested subagent is delivered to the root session instead of the requesting parent
- Desktop app: prompt suggestions (Tab to accept) silently stopped — host gate off since ~Aug 14-20 while user setting is on
- [Feature Request] Hide /clear command entries from session history
- Unable to delete open sessions on VS Code
- [FEATURE]
- Artifact publish stuck in a loop refusing content as "identical to a previously refused version"
- Claude-in-Chrome extension shows disconnected on macOS Claude app despite both reporting connected
- Auto-update banner persists after restart when already on latest version
- VS Code extension: 'Unsupported content type: advisor_tool_result' in webview
- Claude in Chrome: navigate fails with "Navigation to this domain is not allowed" for every domain, despite all permissions being fully open
- v2.1.243 segfaults before main() — bundled allocator interposes glibc free(), crashes in __newlocale (native install, Linux x86-64)
- [Feature Request] Persistent session support for multi-turn conversations
- [BUG] Blocked `sleep` recommends an unbounded `until`-loop, which becomes an unkillable background task
- [Windows desktop] GPU process crash while Browser pane loads a Cloudflare-challenged site; app hangs as zombie, relaunch impossible until MSIX Repair
- HTTP MCP server registers as Connected with zero tools when it issues no Mcp-Session-Id
- Browser/preview pane hard-crashes the entire Claude Desktop app on Windows 11 — conhost.exe AppHang, hang type `Quiesce` (correlates with KB5120998)
- Surface plan usage limits to the agent for budget-aware planning
- [Bug] Unclear Safeguard Penalty During Adversarial Review Plugin Refactoring
- Cross-session messaging (ListAgents/SendMessage) permanently disabled after update, survives reboot/reinstall
Resolved issues (155)
- [BUG] With each release, the installer package increases in size by about 10–20 MB.
- [BUG] Subagents cannot reply to inter-agent messages: incoming from= is the agent type ('general-purpose'), which is unroutable
- [BUG] Cloud routines / Claude Code on the web have no non-interactive way to trust a project-scoped `.mcp.json` MCP server - daily routine broke after the untrusted-workspace gate
- [BUG] Cloud routines fail with "An error occurred while executing Claude Code" every morning since July 4
- [BUG] Code on the web: `apt-get update` in setup scripts fails with exit 100 — base image ships the retired ondrej/php PPA (upstream 'Label' change)
- [BUG] Routine detail page renders blank (app shell loads, content area empty)
- Plugin install fails with "EXDEV: cross-device link not permitted" (Windows desktop app)
- [BUG] Custom network egress policy blocks 3rd custom domain (exact and wildcard) while earlier entries on the same list work
- Fable 5 → API-only: please reconsider a subscription path (👍 + comment here if it affects you)
- [Feature Request] Effort level setting should not interrupt agent autonomy or require additional input
- [Bug] Character encoding corruption in tmux terminal integration on Ubuntu 24 LTS
- [BUG] ERROR "Install Git, Git for Windows is required to run local sessions..."
- [BUG] Dispatch entry point completely absent from Claude Desktop sidebar on Windows — mobile shows Dispatch and fails pairing with "Pairing failed"
- [BUG] Desktop sign-in magic link is consumed by the mobile app, so no verification code is ever shown
- [BUG] Claude Desktop installer (Claude Setup.exe) hangs indefinitely on Windows 11 — EventID 1002 Application Hang
- [BUG] Claude Desktop allows user to submit text not present on the screen - Pasted URL creates an auto-generated hyperlink, and modifying the visible URL doesn't modify the underlying hyperlink
- Windows app: per-app volume is lost on every update (60 stale mixer entries), and a wrong level is invisible in-app
- [BUG] Claudeコード内のブラウザバグについて
- [BUG] visualize:read_me and visualize:show_widget return 400 in Claude Desktop (Windows) and don’t register in tool context
- Desktop (Windows/MSIX): transient auth 503 → in-app Turnstile challenge → GPU crash; package left in Modified, NeedsRemediation
- [BUG] "Couldn't update saved configurations" in Configure Third-Party Inference
- Subject: Confirmation of authorized security research — Louis Vuitton public bug bounty (YesWeHack)
- Jak skontaktować się z SAS w Polsce? [SAS polska]
- [Windows] Desktop app spellcheck doesn't support Vietnamese (works fine on macOS)
- [Bug] Database Error: Could not find table 'public.v_user_status' in schema cache
- /mo[Bug] Consecutive API failures with Claude Opus and Claude 3.5 Sonnet models
- [Bug] Inconsistent task completion compared to competing models
- [Bug] Claude unable to process cybersecurity assessment data and findings consolidation
- [Bug] SafeGuard incorrectly triggered on benign content
- [Bug] Agent fails to follow user-specified rules or constraints
- [Bug] Fable 5 Safeguard Triggered Incorrectly
- [Bug] False positive security risk flagging for non-cybersecurity code
- [Bug] Unable to continue conversation - API request failures across multiple sessions
- [Bug] Tool execution deletes files without user confirmation
- [Bug] Agent generates incorrectly formatted commands
- [Bug] Safety guardrails incorrectly triggered during API testing workflow
- [Bug] Agent communication timeout: agents idle instead of exchanging messages
- [Bug] Model misinterprets user intent as criticism, prioritizes defensiveness over task execution
- I'd be happy to help you create a GitHub issue for this API error, but I need more details to generate an accurate technical title. Could you please provide: 1. **What is the exact error message** you're seeing when trying to add a skill? 2. **What step t
- [Feature Request] Add learning plan generation capability
- [MODEL] Quite hard when you are working with a pentester because safeguards are always thrown while you just want to give him information or this kind of things when you want to make a plan
- [Bug] Reduced reliability and consistency compared to previous model versions
- [Bug] Claude Code repeating own wording in generated code
- [Bug] Security flagging incorrectly triggered for code review operations
- [Bug] False Positive Detection in Code Analysis
- [Bug] Opus 5 model performance issues
- I appreciate you providing the request ID, but I don't have any bug report or issue description to generate a title from. To create an appropriate GitHub issue title, I need you to provide: - A description of the problem or unexpected behavior - Any err
- [Bug] False positive error during OS troubleshooting on tablet
- Claude Chrome extension doesn't connect
- [Bug] API Error: Fable 5's safeguards flagged this message
- [Bug] Anthropic API Error: False Positive Detection
- I'd be happy to help, but I need more information to generate a proper GitHub issue title. Could you please provide: 1. **What error or unexpected behavior are you experiencing?** (e.g., crash, wrong output, API error message) 2. **What is the exact error
- [Bug] Generated code output lacks clarity and readability
- [Bug] Excessive message flagging during normal operation
- [Bug] Anthropic API Error: Overly broad safeguards blocking legitimate security audit requests
- [Bug] Resume session fails with "undefined is not an object (evaluating 'e.includes')"
- Cloud web session (claude.ai/code) hangs indefinitely after initialization completes, zero response to any message
- [BUG] 2.1.19x native CLI: OAuth login fails CERT_HAS_EXPIRED on claude.com / platform.claude.com (no proxy) — bundled Bun 1.4.0 regression vs 1.3.14
- [Bug] Safeguard classifier triggers false positives based on conversation history instead of current message content
- [Bug] Session transcripts leak secrets (GitHub PAT, API tokens) to persisted logs
- [BUG] Claude offers/accepts pasting secrets (API keys, tokens, passwords) directly into chat with no upfront warning
- [BUG] Fable 5 fallbacks to opus for all questions since you are using memory for classifier and memory from sessions show I do cell biology research
- Model output degeneration: ~8.7k chars of unrelated training-data-like text (blog boilerplate + </document_contents> tags + eval-style instruction) appended to a normal response
- [Bug] Claude Fable model incorrectly triggering safety guardrails on benign CRM code
- [Bug] Safety guardrails blocking legitimate oncology research workflows
- [Bug] Anthropic API Error: Safety policy violation with Claude Fable 5
- [Bug] Claude 3.5 Sonnet safeguard triggering unexpectedly, forcing fallback to Opus model
- [BUG] Desktop Linux: sidebar hover flyout closes when moving cursor from hamburger icon toward "Chat" tab, but only when starting from "Code"
- [Bug] Anthropic API Error: Overly aggressive cybersecurity safeguard triggering on benign requests
- [Bug] Fable 5 safeguard blocking model access, reverting to 4.8
- Bug: /insights narrative analysis fails silently on large...
- Fable5 session keep throwing hitting session limit & does not support changing model
- [Bug] Persistent API Response Timeout: Retry Loop Since June 22
- Garbled [Dynamic Context] block leaks shredded system-prompt/skill content into tool output after auto-compaction
- Intermittent malformed or silently dropped tool calls in long sessions (1M context, post-compaction)
- [Bug] Model auto-switches from Fable to Opus mid-task on safety flag
- [BUG] Sumarização destrói comportamento em redesenho em curso - projeto comprometido
- [BUG] Auto-opened URLs (e.g. /login) are wrapped in literal quotes on Windows, breaking Edge
- [Bug] False positive safety policy violation on Fable 5 usage
- [Bug] Excessive safeguard filtering on standard file format parsing tasks
- TUI layout collapses to near-blank screen under tmux/byobu in 2.1.201 (regression from 2.1.200)
- [BUG] Claude Code Fable Model Is completely non Sense.
- [BUG] Cowork silently corrupts files in OneDrive folders with Files-On-Demand
- [BUG] Fix with Claude Code does not appear
- [BUG] Claude code hangs
- [FEATURE] Desktop app: show which folder a session is running in, not just repo and branch
- [BUG] Persistent 401 Authentication Error
- Sessions rerouted from pinned claude-fable-5[1m] to Opus 4.8
- [FEATURE] Quick task progress recap after hitting usage limit
- [Feature Request] Refine safeguard filtering to distinguish defensive security auditing from malicious activity
- Fable 5 safeguard false positive: defensive security triage on own repo triggered silent model switch to Opus 4.8
- Background security reviewer guesses wrong absolute paths (first Read always fails) and intermittently fails findings schema
- [FEATURE] Optional per-session activity summary the agent can read for self-monitoring
- Claude Code provides no in-session determinism mechanism, forcing automation users onto the metered Agent SDK path
- Telegram plugin: surface reply_to_message context on inbound channel messages
- [FEATURE] Desktop app: copy button at the bottom of long code blocks (avoid scrolling back to top)
- VS Code extension: option to collapse edit diffs to +/- line counts (like desktop app)
- Show last-updated timestamp for sessions in chat history sidebar
- [Feature Request] Improve model selection detection to avoid unintended switches to Opus 4.8
- [Feature] Live mid-run steering of a running Workflow: hot-reload the already-persisted script + signal in-flight agents to re-read settings (change model/effort/prompt, inject agents) without a stop→resume cycle
- [Feature] Per-agent and per-phase goal in the Workflow tool (scope the /goal self-verification loop below the session level)
- [Feature Request] do not switch away from Fable when working on trying to add Support for open-weight models with custom attention mechanisms
- [FEATURE] Desktop Code tab: user-defined nested folder hierarchy in the session sidebar (Obsidian-style)
- [FEATURE] Desktop Code tab: tab-based session UI within one window (Obsidian/browser-style tabs)
- [FEATURE] Desktop Code tab: visit-history Back/Forward session navigation + remappable shortcuts
- VS Code拡張: 長文の初回プロンプトが画面を占有し回答がほぼ見えない
- [Windows] Bring back persistent terminal sessions to eliminate flashing console windows
- [Feature Request] Improve safeguard calibration for scientific research in geochemistry and Earth history simulation
- Add a shell-execution mode that never touches the model or context
- [Feature] Per-call `baseRef` on the Agent/Task tool (isolation:"worktree") to override the session-wide worktree.baseRef setting
- Cross-session message bodies arrive HTML-entity-encoded: Desktop escapes and stamps encoded=1, the CLI never decodes
- [Bug] ESC key does not close `/status` and `/mcp` popup windows, blocking session interaction
- Plugin PreToolUse hook matcher removed from cache/hooks.json still fires after full restart
- [Bug] Inline LaTeX/MathJax rendering not supported, only block-level math
- Auto-update banner persists after restart when already on latest version
- [BUG] Title bar context menu freezes app on GNOME/X11 (infinite _NET_CLIENT_LIST_STACKING loop)
- [BUG]
- [BUG] GPU/Viz process crash takes down entire app when Browser-pane preview loads a page that trips the private-host XHR block
- [Bug] Claude model producing incorrect results, workaround required switching models
- [Bug] Claude Code ignores user instructions and fails to follow processes consistently
- Terminated (dead) teammates linger in the teammates tray until the session ends
- [FEATURE] TOTP-based per-session authentication option for headless/SSH environments
- [BUG] Any CLAUDE.md file (even empty) hides the Claude Code header
- Daily token badge undercounts usage: claude-fable-5 excluded from headline sum but counted in model breakdown %
- Claude Code fabricated/misattributed user message about ending session
- Feature idea: agent-mediated feedback with curated aggregation visible back to users
- [Bug] Fable 5 false-positive on defensive-security threat modeling in privacy tool
- [BUG]
- [BUG] Sidebar session items on claude.ai/code are <div>s, not links — can't open in new tab or copy link
- [Bug] Safeguard falsely triggered during game design discussion, model downgrade irreversible
- grace_daily_limit_reached triggers inconsistently for claude --print, independent of account usage % shown in UI
- [Bug] Unclear "Fable" flag or parameter issue
- [Bug] Claude fails to generate code output, only produces questions
- Auto mode classifier denies cleanup of side effects from an action approved earlier in the same session
- C:/Program Files/Git/remote-control disabled server-side — Max account, was working until ~June 30
- [FEATURE] Path anchor prefix for the current project's metadata directory in sandbox filesystem rules
- Workflow StructuredOutput: placeholder 'test' payload accepted as final result after repeated schema-validation failures (built-in deep-research synthesize)
- Thinking blocks broken in VSCode extension v2.1.198 (regression from v2.1.197) [BUG]
- macOS: Full Disk Access / TCC prompts reappear on every auto-update (per-version binary path)
- [BUG] Usage Limit Jumps from ~80% to 100% (with proof)
- [BUG] Remote VM session: tool call in flight at laptop sleep phantom-denies after exactly 600s — "user doesn't want to take this action" + toolUseResult "Error: undefined", despite bypassPermissions
- [Bug] Inappropriate content filter flag on benign scientific discussion
- [Bug] Scheduled agent run fails silently with Git permission error at execution time instead of validation
- [Bug] Ambiguous prompt interpretation when Claude asks for next steps with numbered options
- [BUG] Recurring false-positive Usage Policy refusal triggered by screenshot/vision content, not prompt text
- Feature request: Enter on empty input sends "continue"
- [MODEL] Claude executes a previously suggested action without explicit user approval (regression since ~v2.1.193)
- Remote Control: background subagent activity and tool calls are invisible in the app UI
- Workflow tool script sandbox does not expose globalThis.crypto
- Session creation fails for directories with trailing whitespace in path (trust mismatch)
- Message queued during agent work is silently dropped when app backgrounds / phone locks (Android)
- VS Code extension: panel history lists pruned/unloadable sessions and disagrees with Spark (Activity Bar) history
- Feature request: event-triggered skills (skill frontmatter `on:` triggers or hook action type `skill`)
- [BUG] Cannot connect Claude CoWork to private Github repo
- [BUG] Opus 4.8 in Claude Code declares work "verified" / "done" without running the canonical build — false-green regression vs. Opus 4.7