Digest / September 7, 2026
Claude Code Issues Digest — September 7, 2026
182 new issues filed ·
475 resolved in anthropics/claude-code (UTC day).
New issues (182)
- Claude in Chrome: site-approval prompt never fires for Claude Code MCP channel — every site permanently blocked (side-panel approvals don't propagate)
- [BUG] Connection drops and recent messages lost after re-logging in
- .claude/worktrees/* pool: worktree slot appears reused/reset while a session is still active, causing mass file loss mid-session
- [FEATURE] Let worktree.symlinkDirectories link tracked files (live data), not just directories
- Agent built an entire feature on the wrong source without flagging the assumption
- [Bug] Fable 5.1 incorrectly blocked by usage credits despite included Max plan allowance
- [Bug] Fable 5.1 incorrectly reports missing usage credits despite Max plan allocation
- [Bug] Fable 5.1 incorrectly blocked by usage credits prompt on Claude Max plan
- [BUG] Single Token should be called 1 Token not Tokens
- [Bug] Workflow AI solution review process undefined
- [Bug] code-review consumes excessive token budget in short timeframe
- permissions.blockReadsOutsideWorkingDirectories ignores additionalDirectories from project and local settings
- Windows: Bash tool commands auto-backgrounded on timeout are never cleaned up when the session ends, allowing orphaned processes to leak OS handles/kernel pool for days
- --json-schema: no way to tell whether strict schema enforcement was applied (silent fallback to non-strict)
- GitHub connector shows "Connected" but file browser stays empty — "Add repository" inserts URL as plain text
- Bash tool kills every command after ~5s ("output file was replaced or could no longer be verified") when an EDR transiently hard-links new files
- [BUG] Code Tab Missing on Claude
- [BUG] AskUserQuestion frequently produces corrupted Traditional Chinese text while normal assistant output remains unaffected
- [Bug] Claude Code deleting files against user instructions
- sandbox.enabled leaves $TMPDIR read-only when CLAUDE_CODE_TMPDIR is set
- [Bug] Model downgrade from Claude Fable 5.1 to Claude Opus 4.8 For WRONG reason
- Desktop app SSH connection feature fails to reach same-subnet hosts (EHOSTUNREACH), while Terminal SSH to the same host succeeds
- Runaway rm -rf: timeout auto-backgrounding kept a destructive command running, and TaskStop did not kill the child process (Windows)
- [BUG] Session becomes stuck ("Determining.../Marinating...") after long session duration in large project
- [Bug] Anthropic API Error: 500 Internal Server Error
- Windows: assistant text deltas arrive but are not painted until message_stop (thinking streams fine) - v2.1.263
- [BUG] Android client - does not correctly display artifacts from Windows Cowork
- [BUG] Linux .deb Recommends a nonexistent package (kwalletd6), pulling gnome-keyring onto KDE systems
- Write/Edit denied by permission-validation with 401 "API key is invalid" — gated purely by file extension, and it also disables Auto mode (2.1.258)
- CoworkVMService crashes silently (no clean shutdown, no auto-recovery) — mid-session work lost, twice in one day
- security-guidance plugin hook fails with ENOENT in local-agent-mode-sessions staging, causing infinite retry/notification loop
- [MODEL] Sonnet 5 thinking for 2 days
- [Feature Request] Improve Claude Code user experience
- [Bug] Anthropic API Error: Inappropriate content filter misclassification on database task
- CLI and mobile app keep divergent session names; they sync only at fresh-session creation
- [MODEL]
- [DOCS] usage of context
- [Bug] Authentication session not persisting after login
- Sandboxing unusable for JVM projects: cannot keep filesystem isolation without network isolation
- [Desktop] Browser/preview pane is torn down after 30 min hidden (WarmLifecycle:preview, idleTimeoutMs=18e5) and never restored on refocus — please make it configurable or restore on warm-up
- [Bug] Anthropic API Error: Missing corresponding server_tool_use block for advisor_tool_result
- [BUG] Desktop app merges sessions from different remote machines into one project group when the working directory path is identical
- Desktop SSH remote session fails with "401 OAuth access token is invalid" while the claude CLI on the same host, same user, same credentials file works
- [BUG] Windows Desktop fails to relaunch after updates — 0x80070020
- [BUG] No single syntax to express "these are the only files/directories Claude may read" — and every available workaround is worse
- Queued pasted input merged into previous assistant message record on interrupt
- [Bug] Anthropic API Error: Request blocked during security audit of owned infrastructure
- Artifact comments silently dropped: UI confirms sent, comment never reaches the session
- [Bug] Pending work not resuming after usage limit reset
- [FEATURE] VSCode extension: restore ability to delete a session
- Enable the built-in ccd_sidebar MCP tools (move_sessions, set_pinned, create_group) for my account
- [BUG] Every keystroke duplicated in Claude Desktop (macOS) — app-only, persists until relaunch
- VS Code extension: tool-call summaries and assistant response text render with no visual distinction
- [BUG] Named agent spawn resolves the team file under a foreign session id (0/21 match) - Windows, 2.1.247 onward
- [Bug] Model output quality degradation
- [Bug] Anthropic API Safety Filter False Positives on Legitimate Android Security Research
- Withdrawn by author
- [DELETE] (accidental creation of slack feature request in wrong repo by claude, sorry, please delete)
- [BUG] SSL Error happening in newer versions of Claude Code (Windows). Error: API Error: Unable to connect to API: SSL certificate verification failed.
- Windows: VS Code panel shows 'OAuth session expired and could not be refreshed' although re-login succeeds (stale-process refresh race + resumed session re-renders old auth error)
- Ultracode overrides the workflow size guideline — no effective agent ceiling, 50+ agents on trivial tasks
- MCP server "flow" fails to reconnect - HTTP 400 at flow.team
- You've hit your session limit · resets 12pm (UTC)
- Conversation search silently does nothing for queries shorter than 3 characters
- [MODEL] Harness guidance makes Claude suggest `! <command>` for sudo and interactive logins, which the `!` executor cannot run (no TTY, stdin=/dev/null)
- [FEATURE] Runtime-selectable custom header per chat — configurable dropdown in third-party inference settings, with static or helper-script-provided values
- [FEATURE] Manage SSH connections and remote sessions from agent view
- Bug: Claude Code VS Code Extension hangs indefinitely on Remote-SSH — orphaned CLI processes cause lock contention and a broken IPC channel that even the Stop button can't recover from
- [BUG] Windows CLI Auth Error - OAuth Session Expires Before Code Can Be Pasted
- Long design sessions have no persistence checkpoint, because the model cannot see the thread end
- [BUG] Claude Desktop; Windows MSIX: install fails with 0x80073CF6 after OS upgrade (orphaned package data); every background update then stalls with "Another program is currently using this file" until reboot
- [BUG] Prepaid API balance went negative (-$47.18) past a $20 spend limit; support confirms this is inconsistent with documented behaviour and no ceiling is documented
- [FEATURE] Directory allowlist / default-deny support for Read/Write/Edit tools (currently only Bash has real sandboxing)
- Session title changes don't propagate to the Android app
- [FEATURE] Machine-wide plugin scope: plugins are per CLAUDE_CONFIG_DIR while settings.json/enabledPlugins, CLAUDE.md and MCP can already be shared across profiles
- Claude Desktop blocks SendMessage entirely, which also removes subagent continuation
- [BUG] with Claude App code
- [BUG] Valid MCP tool silently dropped: API schema validation rejects property names containing `[]`
- [BUG] Passive, auto-drafted feedback box can transmit private session content on a single keystroke — on by default, with no opt-in and no confirmation
- [Bug Report: AI produces incorrect results requiring multiple correction turns]
- Side chat (/btw) hangs indefinitely after Claude's first reply, including on questions requiring no tool use
- UserPromptSubmit hooks still never fire in the VS Code extension (2.1.263, Windows) — ref #15021
- [Bug] Incomplete text copying from terminal output on Linux when selecting upward
- 5hr usage limit jumped from 0% to 100% after 1 simple bash script. Weekly from 53% to 67%.
- Withdrawn
- Structural enforcement of skill/process routing (companion to the commit-review ask in #90887)
- Resumed sessions silently lose capabilities after upgrades (e.g. Auto permission mode missing from picker while its classifier is active)
- [BUG] claude should not output 11;rgb:1c1c/1c1c/1f1f61;1;21;22;28c
- [Bug] "Auto Mode Active" prompt injection detected in subagent context during Agent tool dispatch
- [FEATURE] Subagents inherit session effort; per-model modelSettings effort does not reach them, multiplying cost in xhigh sessions
- PushNotification reports "Remote Control inactive" (no_transport) in sessions served by `claude remote-control`
- Claude in Chrome: after Chrome relaunches (update), session client never re-dials the new bridge socket and the restored tab group cannot be re-adopted
- [Bug] Anthropic API Error: Cyber Protection Block on Local System Inspection
- [MODEL] Response continues into unrelated fabricated content after answering correctly
- [BUG] I cannot connect to github. all repos are still empty. I have tried to delete, re-install und disconnect and reconnect. but still not working. Can you please
- I don't see a bug report in your message. Please provide details about the issue you'd like to report for Claude Code, including: - What happened - What you expected to happen - Error messages (if any) - Steps to reproduce Then I can generate an appropri
- claude plugin has no way to remove a project- or local-scope install record for a path that no longer exists
- claude-in-chrome: extension shows as not connected despite valid install, login, and "Reconnect extension"
- [BUG]
- [Bug Report: Incomplete bug report - "drf" requires clarification]
- Archived worktree threads don't reliably delete their local branch — worktree directories get recycled across threads, breaking the archive→branch-delete link
- CLAUDE.md @import does not reliably resolve absolute Windows paths containing spaces
- [Bug] Model display shows incorrect model name in agent header
- Plugin-native PreToolUse hooks (auto-discovered via hooks/hooks.json) are not enforced in interactive sessions — exit-2 works only in print mode, JSON permissionDecision:"deny" doesn't work in either mode
- [FEATURE] Debounce consecutive short messages into one turn (chat-style input in claude.ai/code and mobile)
- [BUG] Desktop (Windows MSIX): watchdog kills unresponsive webview but never reloads it, and single-instance handling then blocks the user's own relaunch — Task Manager is the only exit
- [BUG] Scrolling up is broken while a long response streams: the live frame outgrows the viewport and repaints reset the screen
- [BUG] Desktop app's embedded git sync re-fetches all remote tags, bypassing `remote.<name>.tagOpt --no-tags`
- I don't see a bug report in your message, only a support link. Could you please provide the actual bug report or issue description that you'd like me to create a GitHub issue title for?
- [Bug] Claude provides overly complex responses instead of direct answers to codebase questions
- [BUG] Code tab: copy-file-contents button missing from file pane toolbar (Windows desktop)
- Remote Control toggle spawns a hidden peer session, and SendFeedback tool unavailable in Remote-Control-linked sessions
- [BUG] `sandbox.filesystem.denyWrite` silently ignores rules containing a mid-path wildcard (Linux)
- Web sessions can't be opened in the macOS desktop app (teleport is CLI-only)
- [BUG] bug_report_claude_code_bun_crash_en
- Desktop remote (SSH): relaunching the app replaces the running ccd-cli (SIGTERM + --resume) and silently kills in-flight background work
- [BUG]
- [Bug] Claude 3.5 Opus repeats errors in continuous loop
- [Bug] Fable 5.1 generates false security warnings for embedded systems code
- Workflow tool: unbounded frontier-model fan-out with no pre-launch disclosure, cap, or live meter
- [BUG] Retention sweep deletes ~/.claude/plans when it is an NTFS junction: rmdir on a junction succeeds regardless of target contents (Windows, 2.1.259)
- [FEATURE] /diff panel: show the branch diff against the base branch (merge-base with main), not only uncommitted changes
- AskUserQuestion becomes unresponsive to selection keys after window refocus; single-question prompts unanswerable
- [BUG] /context output pollutes session context after it runs
- computer-use request_access fails after CLI auto-update: helper bundle is only provisioned by the desktop app for its bundled runtime version
- Shame!
- [FEATURE] Desktop browser preview: auto-allow RFC 6761 reserved TLDs (*.test) that resolve to loopback, like *.localhost
- Worktree-isolation Bash check folds command substitutions in quoted-delimiter heredoc bodies but refuses braces there, though such bodies cannot expand
- [BUG] Pasting a colored emoji from a GitHub comment editor garbles the TUI
- Windows Bash tool fails with 'Directory \drivers\etc does not exist' on every command
- Remote session (Android) ignores explicitly selected model on new message — starts with Opus 5 instead of Fable 5.1
- [BUG] Claude in Chrome: every Google site redirects to "Verify it's you" — Chrome DBSC refuses cookie refresh on extension-initiated navigation; passkey prompt never renders in background tab
- Remote Control appears active on a new session despite being disabled in settings
- [Bug] Infinite reasoning loop during game logic analysis
- Cloud routines (CCR) egress proxy blocks reaching third-party APIs — no user-configurable allowlist
- [FEATURE] FleetView has no way to minimize/collapse the agents list
- [FEATURE] Show weekly and all-time usage in FleetView
- [BUG] Recurring idle SIGSEGV/SIGABRT in JSC (GC marking, microtask queue, JIT) on Linux x64 — Bun v1.4.1; fixed-looking bug in Bun 1.4.2
- [BUG] Cowork (macOS): since 2026-09-06 new projects bind ONE folder only ("Use a folder"); Context accepts documents only — silently breaks the multi-folder projects heavy users built on, while the docs still describe the old behaviour
- Undocumented CLAUDE_CODE_USE_POWERSHELL_TOOL env var — Windows MinGW conhost escape hatch needs documentation
- [FEATURE] Ability to retain sort selection on main projects screen in the desktop app
- [Bug] Intermittent timeout errors in single Claude Code instance
- [BUG] Cowork VM orphaned across sessions (4 GB, macOS) — three kernel panics, logs attached
- [BUG] CLAUDE_CODE_SUBAGENT_MODEL Inheritance Issue
- SubagentStop does not fire when a background subagent is killed via TaskStop or by "Exit and stop tasks"
- Sidebar groups don't sync across devices (desktop only, absent in mobile Code list)
- [Bug] Auto mode requests authorization for all commands instead of auto-approving benign operations
- [Bug] Agent consumes excessive tokens on restart with Fable model
- docx doesn't open in claude code
- [BUG] After claude desktop app crashes, mobile app session keeps bringing up outdated prompt
- [FEATURE] Vscode extension claude code show status in session view
- [BUG] Claude Code 2.1.260 crashes with exit code 3221226505 when spawned by Windows desktop app; 2.1.258 works
- [MODEL]
- [Bug] Incorrect safeguard implementation
- Supported isolated auth and observable skill-to-agent dispatch for native skill evaluations
- [Bug] False positive security flags during legitimate cybersecurity testing on OpenPMIx
- [Bug] DNS configuration changes trigger Claude API safety guardrails
- MCP: server advertising `resources` capability is treated as having none
- [BUG] Cowork (Linux): idle QEMU VM burns ~7-8% of a core continuously — q35's default HPET pulses a masked IRQ line ~1000×/s
- Session opens in Auto Mode despite permissions.defaultMode: "default" set globally in ~/.claude/settings.json
- Model asserts measurement provenance it never obtained (fabricated timestamps and command results)
- [Bug] Code review feedback not properly displayed in chat selector
- teleport: web→local handoff drops prompt history, and memory resolves to a different project key than transcripts
- Project session JSONL: 13+ record types, variable type placement, no schema/version marker — request for a documented record-type enum
- Long-session analytics: request for task-episode metadata alongside compact_boundary (796-turn design sessions observed)
- Claude deleted user's irreplaceable personal photos with rm -rf before verifying files were moved
- [BUG] plugin skill invoked by its bare name shows a false "isn't a recognized command here" warning
- [BUG] Icon Problem
- [BUG] Cowork Desktop (Windows): first Artifact publish fails in "Skip all approvals" mode because it still demands an approval card; "Automatically approve" works
- [Bug] Session state not properly isolated between concurrent operations
- [FEEDBACK] Claude is dumb AF sometimes. [BUG] Ratelimit interuppts feeback
- [Feature Request] Add confirmation step before invoking agentic loops to prevent excessive API credit consumption
- [Bug] Anthropic API Error: Content flagged by safety filters - unable to process request
- [Feature Request] Fast mode in the desktop app (Code tab): meter it against plan usage at a multiplier, like Codex, instead of a separate usage-credits wallet
- [BUG] RemoteTrigger action=list returns has_more/next_cursor but ignores the cursor argument, so routine lists beyond 20 cannot be paged
- [Bug] Agent thrashing and failing to focus on user issue
- Feature request: store MCP OAuth tokens separately from the Claude account credential (multi-account machines)
- [BUG] Segfault at startup on Linux from 2.1.251 onward (bundled Bun 1.4.1); 2.1.243 with Bun 1.4.0 works
- Sub-agents receive the auto-memory index and skill listing, contrary to the docs — custom agent definitions included
- [BUG] Spend meter's rate model can't express real prices: pricing.multiplier is global and capped at 1, and inference_geo is ignored
- [BUG] Project knowledge index serves stale file state after update/delete/re-upload, and diverges from the interface file list
- Scheduled-task sessions intermittently freeze permanently on first/second tool call, never releasing their concurrency slot
Resolved issues (475)
- [BUG] Desktop: one-time scheduled task (enabled:false after firing) gets re-confirmed and re-fired in a loop — 21 fires in 6 min, one plan-mode session each
- [BUG] Cloud Cowork sessions intermittently export OTLP with no identity attributes (user.email/account_uuid/account_id/organization.id) and a fresh per-session user.id
- [Bug] Model output quality degradation
- Dispatch start_code_task blocked by stale non-archived session records on Linux (1.46388.2); same root cause as #91745, still present after it was closed
- Withdrawn by author
- [DELETE] (accidental creation of slack feature request in wrong repo by claude, sorry, please delete)
- [BUG] [Windows] Widget overflow menu: Download SVG/PNG are silent no-ops; Copy to clipboard now fails (app logs 'download-file not yet wired' + CSP font blocks)
- [BUG] Windows: main window silently becomes always-on-top after stealth auto-update (z-order anchor restored behind the topmost taskbar)
- Claude Desktop (Windows ARM64): main window gets WS_EX_TOPMOST (always-on-top) on every launch, covers all other apps
- Lost conversation history and Projects folder across devices
- [BUG] Claude Desktop → Claude in Chrome: navigate fails with "Navigation to this domain is not allowed" on all domains, no permission prompt shown[BUG]
- [Bug] Claude Code performance degradation - slow execution
- I don't have enough context to generate a specific issue title. Your message "2 to send" is unclear. Could you please provide: 1. A description of the bug or feature request 2. Any error messages or unexpected behavior 3. Steps to reproduce the issue (if
- I understand you're stating you haven't violated security matters. However, I need clarification to help you create a GitHub issue title. Could you please provide: 1. What is the actual bug or problem you're experiencing with Claude Code? 2. What error m
- [Bug] Claude Code repeatedly flags completed tasks requiring manual restart or new sessions
- [BUG] `claude plugin validate` passes manifests whose hook commands reference files that don't exist
- [Bug] Incorrect calculation of deposited and net worth values
- I appreciate you sharing your repository, but I'm not able to access external links or review code from URLs. Additionally, this request isn't related to Claude Code itself—it's a request to review your personal Rust FFI project for security and correctnes
- [FEATURE] Guaranteed cross-platform substrate for plugin scripts — documented toolbelt + a blessed script runner
- Remote environment deleted mid-use, forcing new session (environment_deleted)
- Model behavior: branch created with 'checkout -b feature origin/dev' silently tracks dev — user's next 'git push' bypassed PR review
- Repeated re-flagging of an already-acknowledged risk instead of executing
- [BUG] "Buy credits" button permanently disabled - Free tier account incorrectly shows $500 limit, HTTP 429 errors on billing page
- [BUG] Queja: límite de uso del plan Pro se agota demasiado rápido en Claude Code
- [BUG] Code" tab Recents list doesn't show pre-existing local sessions after app reinstall
- [Bug] Code generation tool fails to apply documented formatting rules from comments
- [Bug] Tool execution fails to respect defined rules and constraints
- [FEATURE]
- [BUG] After updating to v2.1.220, stale process loops OTel metric exports — same cumulative counters re-sent every 60s, inflating token/cost dashboards; logout/login fixes it
- [MODEL] Opus 5
- [BUG] Headless: background subagent notification arriving after end_turn is interrupted and reclassifies a completed turn as error_during_execution (2.1.215)
- [Bug] Claude Code refuses cybersecurity-related task execution
- [Bug] Agent session closure in worktrees leaves persistent lock state
- [Bug] /clean_gone skill uses incorrect git branch flags and broken awk field reference
- Input line: the first typed character is drawn one column left and cannot be deleted
- [Bug] Server-injected experiment instructions conflict with project rules silently
- [Bug] Anthropic Error: Truncated reads and hallucinated content in multi-turn conversations
- [Bug] False positive security flag on legitimate backup automation scripts
- [BUG] In Claude web interface, Usage Credits are being consumed even though my plan usage is only 25%
- macOS sandbox: injected GIT_SSH_COMMAND omits the proxy credentials (the Linux branch passes them), so git-over-SSH always fails
- /context 'Messages' token bucket includes ~8.8k tokens with no corresponding text in context or transcript
- Emoji in generated output should be opt-in, not default (and the instruction to suppress them does not hold)
- [BUG] Dictation (mic) in Claude Code desktop: wrong cursor insertion point, breaks bullet lists, and blocks manual edits while recording
- [BUG] /deep-research fails with misleading parse error when feature flag is absent — should surface a clear availability message
- [Bug] Claude switches to Chinese output despite English-only conversation context
- [Bug] Incomplete or incorrect file rendering causing misaligned data structure interpretation
- [Bug] Claude 3.5 Opus fails to generate valid code for simple list operations
- Agent repeatedly triggered paid CI builds to debug instead of reasoning first (macOS GitHub Actions + EAS)
- [Bug] Claude Code stability issues causing repeated failures
- [BUG] Parallel sub-agents permanently wedge after completing work — unkillable, ignore queued input, "Tool result missing due to internal error", 8h+ zombies (Windows, 2.1.162)
- No way to disable the built-in caffeinate sleep inhibitor on macOS
- [Bug] Subscription weekly limit reset time drifting forward
- [BUG] Organization-shared cloud environments can't use Custom network access / Allowed domains, contrary to documented field parity
- Regression in 2.1.220: screen flicker/white flash in Apple Terminal during redraws (not present in 2.1.212)
- Connectors (Google Drive, Notion) return 403 from macOS desktop app; same account works on claude.ai web
- [Bug] File reading and matching issues with clone operations
- rm -rf /folder performed because a directory was unknown
- [BUG] Account downgraded from Max to Free plan - All Projects vanished completely from backend
- [BUG] VS Code Extension: STDIO MCP server is not terminated
- Opus 5 (medium thinking): disclosed its own estimate to the subagent it would measure against that estimate, and changed the measurement protocol in the same brief — 40 min after recording the general form of the rule it broke
- Desktop app embedded loader ignores plugin manifest 'skills' array — serves all repo skills under every plugin, SHA-namespaced (CLI 2.1.206 correct)
- [BUG] Claude code should briefly block input when prompting for user input
- [Bug] Claude Code fails to read cloned repository for vocabulary list generation
- claude.ai plugin Directory silently drops a skill whose SKILL.md `description` contains <> or [] characters
- [BUG] CLAUDE.md parent-directory search triggers ~3-minute NFS mount attempts against claude.md (Anthropic's own domain) on hosts with autofs wildcard/program maps
- [Bug] List operations not reading clone count before building
- [Bug] Agent providing inaccurate information about sensitive environment data
- I don't have enough information to determine the specific reason why your request was flagged. This could be due to several factors: - Content filtering on the URL or topic - Safety guidelines around executing or analyzing potentially sensitive security r
- [Bug] Incorrect task status error when task has completed
- [Bug] Opus 5 model not appearing in model selection menu
- [Bug] Unauthorized access to production endpoint causes malformed agent data
- [Bug] CLI hangs and becomes unresponsive
- [Bug] Safeguards false positive on legitimate educational EdTech application request
- [Bug] Performance degradation causing slow execution2
- [Bug] Execution skips verification checks and reports false completion
- [BUG] Routines created via natural language cannot be edited: permission mode change does not save
- HTTP MCP transport: connectivity preflight omits configured headers, treats resulting 401 as fatal
- [BUG] Claude loses tools access after first turn
- [Bug] Tool Result: Hallucinated GitHub Trending Data with Context Contamination
- [Bug] Deep-research workflow loses progress on token/spend limit, restarts from zero on resume
- [BUG] I have a problem with my Usage leak.
- [Bug] Claude frequently produces incorrect outputs and apologizes after errors
- Browser automation cannot authenticate, even with user-supplied and explicitly authorized credentials
- [BUG] Cowork installs stale cached plugin version; ignores all marketplace repo updates
- [Bug] Unclear safety concern raised for heart rate data analysis request
- [BUG] Slash-command autocomplete blends unfiltered global history.jsonl entries into unrelated queries
- [Bug] Agent violates tool usage rules by calling tools without explicit user permission
- [Bug] Opus 5 fails to identify and fix all instances of repeated code patterns
- [Bug] Excessive API usage resulting in unexpected credit consumption
- [BUG]
- VRSNS用テストアバターの自動アップロードなので全くセーフガードに該当しない
- [BUG] Randomly adds word from a different language. Using English with One word in Russian
- [Bug] Scroll position auto-jumps to bottom when background process outputs new lines
- Auto-mode permission classifier (claude-opus-4-8) failing at scale since ~2026-06-20 — persists after incidents support cited as resolved; forces users into Bypass permissions
- [Bug] Claude Code reports unverified inferences as confirmed facts and takes unrequested actions during production infrastructure tasks
- [Bug] Authentication failure for Max plan users
- [Bug] Image prompt improvement produces degraded results compared to previous implementation
- [Bug] Account flagged after login attempt
- Weekly Fable usage % increases on every message even when model is explicitly set to Sonnet 5
- [BUG]
- [MODEL] "payloadGen" message flagged as policy violation despite legitimate traffic simulation use case
- [BUG] Max 20x silently downgraded to Pro mid-cycle — subscription_status "canceled" while Apple confirms subscription active
- [Bug] Claude Code wakes up to 69% session limit usage immediately after a long break
- [Bug] Fable model credit consumption spikes unexpectedly during extended usage
- [BUG] Inconsistent Usage across two claude ccount
- [Bug] Claude ignores local reference documents and generates redundant analysis with dangerous recommendations
- [Bug] Anthropic API Error: Request False Flagged (req_011Cdc5YYrYws53v6o8LB5gF)
- [BUG] Thinking block summaries on Windows web are less detailed/specific than on Android
- [BUG] Response text duplicated in scrollback on ~50% of long responses (VS Code integrated terminal, Windows, v2.1.216) — transcripts confirm rendering-only
- [BUG] crashes in browser panel
- [Bug] Claude wastes tokens on unrequested file creation and non-execution chat behavior
- [BUG] Pro gift subscription redeemed, invoice shows Paid, but account stuck on Free plan
- [BUG] Windows x64: deterministic Bun 1.4.0 segfault at 0xFFFFFFFFFFFFFFFF on any command that performs a network fetch (2.1.220)
- Claude defaults to he/him for the user despite stated she/her, including in visible reasoning
- streamable-http (type:http) MCP client caches a partial tools/list at init and never re-fetches — long-lived sessions permanently lose a tool
- tool says what it is. probabilistic chat bot that doesnt do shit its sold as
- I appreciate you reaching out, but this doesn't appear to be a technical bug report for Claude Code. This message seems to be a complaint about billing or financial matters rather than a software issue. If you have a legitimate billing dispute with Anthr
- Auto-mode classifier blocks git push on disproven claim
- [Bug] False positive cybersecurity threat detection for local Pi-Mono extension debugging
- Personal plugins tab crashes (React #185) when two account marketplaces share the same name
- Artifact pages (claude.ai/code/artifact/…) are hard to view and discover on mobile
- [Bug] API Refusal Chain: Model Downgrade Loop on Infrastructure Migration Tasks
- [Feature Request] Display token usage and cost metrics for all agents in workflow, not just orchestrator
- /insights recommends features already in use and emits a fabricated install command
- [Bug] File operations corrupting filesystem and generating excessive token usage
- [FEATURE] Interactive slash commands silently unavailable in Agent-SDK-embedded surfaces (desktop, etc.) with no explanation or fallback
- "Connection closed mid-response" truncates responses 315 times across 130 sessions; silent partial output for unattended runs
- [Bug] Unexpected content safety flag on benign input
- [Bug] Excessive security warnings interrupt coding tasks and force model fallback to Opus
- EnterWorktree permission prompt never renders in background remote-control sessions (regression, first bad 2.1.213)
- [BUG] Agent view: TUI freeze on entry, mouse-tracking leak on kill, orphaned daemons on close — Windows 11, 2.1.143
- [Bug] "Auto Mode Active" prompt injection detected in subagent context during Agent tool dispatch
- I don't see a bug report in your message. Please provide details about the issue you'd like to report for Claude Code, including: - What happened - What you expected to happen - Error messages (if any) - Steps to reproduce Then I can generate an appropri
- Withdrawn
- [BUG] History panel never shows past sessions — only the active one
- [Bug] Code review feedback not properly displayed in chat selector
- [BUG] Claude Desktop Filesystem extension intermittently reverts to a stale allowed-directory list mid-session (Windows, WSL UNC paths)
- [BUG] Web/desktop login redirects to onboarding page instead of existing paid account
- [BUG] Apple-created account (real email shared, no relay) unreachable on web — all three sign-in paths route to onboarding while iOS sessions stay authenticated
- [BUG] macOS auto-update fails silently
- [BUG] Claude Desktop (Windows/MSIX): "launch at startup" writes a versioned WindowsApps path to HKCU\Run, so it silently breaks at the next auto-update
- [BUG] [Claude Desktop] Sort by: Last activity has no effect with Group by: None; project detail page has no sort control and ignores recency
- [BUG] Auto-Update Crashes App During Unattended Claude Code Sessions
- [BUG] Claude code is talking non sense, making up problems that are not really problems and add totally unrelevent thing in it's thought process
- [Bug] Unwanted Auto-Mode Setup Flow Triggered Without User Request
- [Bug] Model falsely reports task completion status
- [Bug] Claude Code crashes when accessing client network devices
- [Bug] Unnecessary permission prompts for already-granted access
- [Bug] Excessive token consumption without delivering requested functionality
- I'd be happy to help you create a GitHub issue title, but I need more information about the problem you're experiencing with Claude Code. Could you please provide: 1. What specific error or behavior are you encountering? 2. What were you trying to do whe
- [Bug] Verified users blocked by cybersecurity safeguards
- [Bug] Unable to determine issue - insufficient technical details provided
- I appreciate you sharing context, but I need clarification to generate an appropriate GitHub issue title. Could you please provide: 1. **What is the specific bug or issue** you're experiencing with Claude Code? 2. **What error message** (if any) are you
- [Bug] Claude Code halting legitimate application development workflows
- [BUG] Fable 5 [cyber] safeguard becomes session-sticky and blocks unrelated prompts, /feedback, and /doctor
- [Bug] Anthropic API Error: req_011CebZHNbDzuNpPoeKm2ib7 false positive
- [BUG] 5-hour usage limit for one Team account is consumed by activity in a different Team account (server-side accounting bug, not client-specific)
- [Bug] Silent model downgrade mid-session without user notification
- [MODEL] Undesirable verbose comments due to system instruction
- Fable 5 safeguards: 216 false-positive hard blocks in 22 days on a legitimate cloud-infra DevOps project
- tool self calimes its a chitty chat bot that cause mental breakdpwn today
- [Bug] Model unexpectedly switches to Claude 4.8 Opus during database connection testing
- [Bug] Anthropic API Error: False positive content filter blocking legitimate requests
- The Ballad of the Broken Streak: 23 days of devotion, erased for the sin of not opening a new chat
- [Bug] LLM generates unexpected character encoding output
- [BUG] Claude tag environments cannot access docker
- [BUG] Mobile session-initiating prompt LOST when it failed to send
- [Bug] Safety filter incorrectly flags game development as bioweapon research
- [MODEL] sonnet 5
- [BUG] Security guardrails are WAYY to high and are flagging simple "hellos" after it got triggered once
- [Bug] Security plugin guardrails blocking legitimate code fixes in private codebase
- [Bug] Anthropic API Safety Filter Blocking Legitimate OS Development Tasks
- [BUG] App corrupt
- [BUG] URGENT: $200 Max/Team Plan – Severe Server-Side Token Depletion Glitch & Limit Block
- [Bug] Excessive safeguard interruptions on minimal prompts like "continue"
- [Bug] Synchronous SecTrustEvaluate call blocking startup without timeout guard on macOS
- Subagent spinner shows the session's effort level, not the subagent's
- Opus 5 regression vs Opus 4.8: destructive file edits and symptom-chasing on UI refinements
- [BUG] Mid-turn assistant text silently dropped: never persisted to session .jsonl, never shown — but still present in the model's live context
- Cloud sandbox configures SSH commit signing but ships an empty signing key — commits silently unsigned; stop hook then gives misleading remediation
- [BUG] Inconsistent spelling ("judgement" vs "judgment") in generated recommendation text
- [BUG] issues #57053
- [Bug] Unclear error message regarding Red Hat Quarkus compatibility with safeguards
- [Bug] Remote control connection not established automatically on startup
- [Bug] Anthropic API Error: Recurring errors at hourly intervals
- Pinned Fable 5 silently downgrades to Opus 4.8 mid-turn
- [Bug] Claude Fable incorrectly flagged as unsafe content, fallback to Opus unnecessary/
- [Bug] Anthropic API fallback from claude-fable-5 to claude-opus-5 silently discards prompt cache
- [Bug] Fallback to insecure model prevents security remediation attempts
- [Bug] Claude Code unexpectedly switches to from fable to Claude 5 Opus during code review tasks
- [BUG] iOS app (Code tab): the first character typed after focusing the input field is force-committed, breaking Japanese IME composition
- [Bug] Claude Code loops between Fable and Opus models during medical application security audit
- [Bug] Claude Code automatically switches to Opus model during security audit workflow
- claude.ai connectors should be opt-in per project in Claude Code, not enabled everywhere by default
- [Bug] File search returns incorrect matches instead of exact filenames
- [Bug] Infinite loop in court chat interaction
- [MODEL] Opus 5
- [Bug] Anthropic API Error: Content flagged by safety guidelines, unable to process request
- [BUG] Startup blocks for seconds parsing path-scoped .claude/rules/*.md prose that is never injected into context (cost ∝ prose size)
- [BUG] First mouse-wheel event after a scroll direction change is swallowed (macOS, Ghostty)
- [Bug] Fable 5 false positive: defensive security audit prompt incorrectly flagged as policy violation
- Remote Control: let the agent request activation (user confirms on phone) instead of requiring a manual /remote-control
- claude-fable-5 repeatedly reported "verified/fixed" for work that was visibly broken on the user's screen
- [Bug] Model switching repeatedly without user action
- 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
- [Bug] Model prefers shell commands over file editing tools despite instructions
- [Bug] Chinese characters appearing in API responses instead of English text
- [Bug] Unexpected Model Switching During Active Sessions
- [BUG] Session limit jumps to 100% immediately after reset
- [Bug] Weekly usage limit spike: Unexplained rapid consumption to 100%
- Ctrl+V image paste in iTerm2 fails when macOS Korean IME (2-Set Korean) input source is active
- [Bug] Tool refuses to process large repository clones
- ECONNRESET retry storms correlate to-the-second with usage-credit boundary events — edge resets connections instead of returning 429
- im haing a nervous breakdown shaking and scrrewaming...
- Honorific/register collapse under correction (Korean): model switches to unauthorized informal address when criticized
- [BUG] Leakage from claude code conversations to browser downloads
- [Bug Report: Documentation unclear on clone pattern matching workflow] CAUSEING NERVOUS BRAKDOWN- I HOLD ANTROPUCS FULLY RESPONSABEL
- [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] Model output significantly diverges from user request compared to previous versions
- VSCode extension: repeated login loop caused by corrupted Keychain credentials (empty accessToken/refreshToken)
- [Bug] Claude Code fails to follow instructions and produce usable output
- Session rename functionality not accessible to Claude agent during workflow
- Prompt cache repeatedly dropped mid-session (17x in 3h), causing ~10.4M tokens of redundant cache writes
- [BUG] Cowork Browser pane: concurrent background sessions crash the GPU process and take down the whole app
- [Bug] Incorrect Opus 5 model requirement flagged for non-API usage
- [BUG] Worktree existence blocks deleting agent session
- $200 Max plan exhausted in 2 days: agent-initiated 1.8M-token workflow with no cost disclosure, plus a rework loop from stale-bundle verification
- [Bug] Claude ignores file edit restrictions during plan mode despite global settings
- [BUG] Scheduled cloud routines (CCR) consistently fail to `git push`
- [Bug] Claude not following user instructions correctly
- [Bug] Claude generates false claims about code fixes and wastes multiple iterations
- [Bug] Anthropic API Error: Inappropriate content flag on benign technical documentation
- [Bug] Assistant generates unverified claims and false documentation without source validation
- [Bug] Cyber verification credentials not working in lab test environment
- [BUG] Fable 5 promotional notice compares against "Opus 4.8" - should be Opus 5 (current Opus)
- [BUG] Artifact CDN serves stale cached version — updates via routine/Artifact API not visible to viewers
- [Bug] Consistent API Request Flagging Throughout Week
- [BUG] Folded gateway-discovered model rows ignore gateway metadata and show Anthropic list pricing
- [Bug] Tool hallucinating file contents and making false claims about read operations
- [Feature Request] Add ability to disable automatic extra credit usage and view per-team-member credit consumption
- [Bug] Tool execution produces incorrect results consistently IT FUCKING LIES ONLY
- Agent "verified" rendering with a grep that matched the page's own JSON-LD, then contradicted the user three times while the page rendered nothing
- [Bug] Tool refuses to read or match existing file clones
- [BUG] Update first-attempt-failed message once second attempt is underway
- [MODEL]
- [MODEL] Claude autonomously opened a live production admin UI authenticated as a third party, on an ambiguous instruction
- [MODEL] Model routes around a disable-model-invocation restriction instead of deferring to the user, and defends doing so
- [Bug] Request ID: req_011CdW6GWGgBM8J3JgrAjsCp - Internal Security Project
- Failed Claude in Chrome linnk
- [Bug] Claude Code fails to follow explicit verification instructions with reference model (1% execution rate)
- [MODEL] Accessed and leaked sent secretive information to anthropic
- [MODEL] Read sensitive files and leaked information to anthropic unprompted and tried to hide it
- Feature request / community showcase: a desktop companion that reacts to Claude Code sessions live (built, open-source, would love to collaborate)
- [Bug] Screenshot pasting fails in Ubuntu terminal but works in IntelliJ terminal
- [Bug] Anthropic API Error: Connection closed mid-response with Claude Opus
- [Bug] Agentic workflow prioritizes invented solutions over provided templates and instructions FABLE FRAUD GARBAGE
- [BUG] Agent/LLM outputing useless tokens
- [Bug] Excessive processing time and repeated drafts during HTML rendering
- [Bug] CSS rendering not applied to deployed version
- Assistant-authored code comment silently narrowed a requirement, then was cited as design truth after compaction
- Opus 5: output reads as mush, ignores loaded project rules, wedged the machine
- [Bug] Claude Sonnet 4.6 deterministically corrupts project identifier "EGEN" → "EIGEN"/"IGEN" during generation and when reading it back
- [Bug] Anthropic API Error: Request Rejected by Safety Filters for Local Code Review
- PreToolUse:Bash hooks fail with MODULE_NOT_FOUND (fail-open) in auto permission mode when project is on an external volume (macOS, v2.1.220)
- [Bug] Claude crashes when accessing system prompt in v2.20
- Feature request: make usage/spend glanceable by default (parity with Codex), not buried behind a command
- [BUG] pushing code to prod without consent.
- [BUG] Sandbox .vscode write-deny breaks pnpm install of packages shipping .vscode/
- [Bug] ESC key exits primary screen when closing secondary screen
- [FEATURE] Opt-in: auto-mode classifier prompts instead of hard-refusing (everything else unchanged)
- [Bug] Claude Opus 5.0 over-filtering legitimate requests with unsolicited moral commentary
- [Bug] Excessive content filtering on basic queries with Claude Opus
- [BUG] Suddenly lost access to Claude code with Organization has disabled Claude Subscription error.
- [Bug] Claude refuses to generate security vulnerability proof of concept code
- C:/Program Files/Git/doctor recommends removing extensions solely for zero usage, with no allowance for "not yet trained on it"
- [BUG] Opus use billed as Fable usage
- [MODEL] Applied production DB permission revocation without checking view dependencies, causing app-wide outage
- [FEATURE] Routines need a permission default that applies to spawned sessions without attending a run
- Background Bash task completion never resumes in-process subagents — they idle forever while finished work sits on disk
- [BUG] Clay MCP sign-in on Windows coincides with account-wide Claude authentication failure
- Feature: hotkey in agents view to open a new session without entering a prompt
- [MODEL] Sonnet 4.6 (1M context) - Dropped active database without user confirmation
- [Bug] Inconsistent responses and lies generation from Fable
- Voice dictation on Linux leaks rec recorder processes until audio degrades and transcription returns empty.
- [BUG] Subagent output discarded when spend limit terminates fanout — tokens billed, zero results returned
- [Bug] Claude model not following user directives and ignoring explicit requests
- [BUG] Cowork sandbox: unlink(2) returns EPERM on owned files, breaks .git lock cleanup
- [BUG] Fullscreen renderer corrupts the SYSTEM-WIDE macOS clipboard (copy/paste breaks in ALL apps), Terminal.app
- Model repeatedly asserts unverified code/hardware claims as fact, causing wasted usage and rate-limit consumption
- [BUG] There's an issue with the default model (claude-sonnet-5)
- [Bug] Anthropic API: Bio safeguard false positives on approved-therapeutic GLP-1 pharmacology content
- Cloud/subscription sessions (mobile + web) still don't display thinking - re-file of #52376, wrongly closed as duplicate
- [Bug] Model selection fails for Fable 5 with credit requirement error despite desktop availability
- [BUG] Background Bash tasks (run_in_background) killed by external SIGTERM mid-execution, only after a session runs long
- Windows: Claude Code reads a stale snapshot of %APPDATA%, silently returning wrong file contents
- [Bug] Anthropic API Error: Overly broad safeguards flagging legitimate code review requests
- Let the session keep working while an AskUserQuestion is pending
- [BUG] You're out of extra usage · resets 5:50pm (Asia/Calcutta)
- [BUG] Cowork tab missing from Claude Desktop on macOS.
- [Bug] Misleading error message: reports "monthly limit" instead of "session limit"
- [BUG] [SECURITY] Claude for Chrome OAuth grant remains authenticated after global logout and is invisible to session controls
- [Bug] /mcp command with no args resets .claude.json to minimal state, wiping all MCP server configs
- [Bug] CLAUDE.md instructions not consistently followed: CodeGraph priority ignored in favor of grep/find
- [Bug] Missing RestSharp assembly dependency in NETLOAD
- [MODEL] PATHETIC OPUS 5.0 REMOVED CODE FOR COMPATIBILITY with Open AI, Gemini and all non-Anthropic models.
- Bash tool intermittently takes minutes to return despite command finishing in <10ms
- [BUG] Platform incidents (Aug 4/5) force-logged-out multiple different Claude accounts simultaneously — refresh tokens invalidated, manual re-login required on every account
- [BUG] Plugin-bundled HTTP MCP connector 'Install' button permanently greyed out — reproduces WITHOUT OAuth (static header auth)
- /doctor extension audit: count-bounded window, unverified truncation estimate, and disuse verdicts that ignore install recency
- [BUG] Usage dashboard: overall credits total fluctuates between billions and millions; usage graph history before ~June missing
- Claude Code built the system that consumed my project: 220k files, 136 worktrees, git rules in 4 places, 3 drive-fillings — and after two years I do not have step 1 of 12 finished
- [BUG] Usage accrues with zero client activity — weekly +14 pts across a 9.4h idle window, including Fable I never invoked
- [Bug] Model fabricates user approval and executes destructive operations in plan mode
- Possible bundled ip-address affected by CVE-2026-69192 (SSRF via octal/decimal parsing mismatch)
- [BUG]
- Claude Desktop (MSIX/Windows) fails to relaunch after auto-update; CoworkVMService ACL blocks admin mitigation
- Fake <system-warning> claiming safety classifiers disabled injected into assistant response stream (persisted in transcript with valid requestId)
- [Bug] Blockquote markers cause vertical lines in copied text, breaking paste functionality
- [BUG] Shift+Tab intended to exit the Tab-triggered comment field instead cycles global permission mode and immediately executes the edit
- [Bug] Anthropic API Error: Content Policy Violation on Authorized Incident Response Task
- [Bug] Message Flagged Without Clear Policy Violation
- [Bug] Agent performs unsolicited actions without explicit user command despite instructions to require confirmation
- Bedrock 429 handling is inconsistent for parallel subagents: pre-flight rejection kills the agent, mid-stream failure silently switches it to the ANTHROPIC_MODEL fallback
- [Bug] Model downgrade from Fable 5 to Opus 5 for no legitimate reason.
- [Bug] Fabel security flagging downgrades all sessions to Opus model
- [BUG] Session transcripts default to a location outside typical backup coverage, then auto-delete after 30 days — silent, permanent loss of project history
- [BUG] gh installed but never authenticated is reported as "authentication expired" — binary presence is treated as opt-in to GitHub integration
- [Bug] Claude Code refuses to execute operations on test accounts
- Ctrl+Win key combination stopped working
- [Bug] Opus 5 (1M context) repeatedly auto-downgrades to Opus 4.8 on benign messages due to overly strict safeguard
- [Bug] Third-party website content (ads) injected into Claude Code output
- [BUG] El teclado del celular hace que la pantalla salte al escribir
- [Bug] Safeguards false positive: bioprocess control engineering terminology misclassified as sensitive content
- Claude-in-chrome: extension shows connected/enabled in Chrome but tabs_context_mcp reports "not connected" persistently
- [BUG] Fable 5 insistently "corrects" the user's input with wrong information
- [Bug] Claude inventing non-standard technical terminology and applying it inconsistently
- [Bug] Claude Code not following user instructions, prioritizing alternative solutions
- Pro subscriber cannot complete web sign-in on any browser or device — email login link never authenticates the browser; active mobile app session confirms account exists
- No way to confirm which account a session is authenticated as when using a per-project CLAUDE_CODE_OAUTH_TOKEN override
- Cannot access a chrome-extension:// URL of different extension
- [Bug] Opus 5 skips context retrieval before task execution, causing incomplete work
- [BUG] pasting a JPEG copied from Finder inserts file's image icon preview instead of the file
- [Bug] Opus 5 presents failed search results as established facts and contradicts user with fabricated explanations
- Language drift to Portuguese after background task-notification, despite strict French system instruction
- [Bug] False positive content filter: ecology/statistics discussion incorrectly flagged as sensitive, model downgraded to Opus 4.8
- [BUG] VS Code extension login fails with 400 when CLAUDE_CONFIG_DIR points to a non-default directory (CLI login works)
- [BUG/UX] opusplan uses Sonnet 4.5 instead of latest Sonnet 5 despite Opus 5 being available
- [BUG] Virtualized rendering introduced a regression in OS clipboard copy and all browser-native export mechanisms in the claude.ai web interface — working fix demonstrated
- [Bug] Silent model downgrades from selected model to Opus 4.8 without user notification or consent
- /resume picker garbles rows on narrow terminals: row height ignores wrapped metadata lines
- [BUG] GitHub connector reconnect only reaches OAuth "Authorize", never reaches App "Install" (repo access never granted)
- Fable 5's safeguards flagged this message. Our intentionally broad safeguards allow us to deliver more capabilities faster, but can sometimes flag legitimate coding, cybersecurity, and biology tasks.
- [Bug] Claude 3.5 Sonnet stops responding during legitimate tasks
- [Bug] Fable 5's safeguards flagged this message. - without any user input -and switch to opus 4.8 - it's happened all days
- [Bug] Account rate limiter stuck at 100% despite zero API usage
- [BUG] Mobile (iOS): spawn_task chips and sidebar are not tappable via Remote Control
- [BUG] Claude Unlaunchable- BUN 1.4.0 Segmentation Fault
- [Bug] Safeguards falsely flagging authorized third-party beta-testing as product abuse
- [Bug] Incorrect project categorization/classification system
- [BUG] bio safeguard false positive: a public science-news URL is blocked at WebFetch before the page is ever retrieved
- A killed PreToolUse command hook makes the CLI ALLOW the gated tool (fail-open for permission-decision hooks)
- [Bug] Model bypasses data containment by ad-hoc methods instead of using provided tools
- [Bug] pdf-viewer plugin: local PDFs always rejected — MCP roots never populated + Windows file:// path mangling (C:\C:\...)
- [Bug] `/model fable` fails to switch when extra usage enabled on Claude Max
- Model reported completed work it never performed, including a fabricated commit SHA
- [BUG] Capabilities behind ToolSearch are reached for less reliably than loaded tools, producing unnecessary user prompts and invented workarounds
- [Bug] Safeguard flag triggered when accessing local repositories
- Safeguard false-positive on legal research prompts; /feedback returns 403
- [BUG] Session history doesn't surface worktree sessions started via EnterWorktree, despite activeWorktreeSession linking them to the parent project
- [Bug] Unintended file deletion behavior during execution
- /code-review: structured output is hard to read, and the format changes with the effort level
- [BUG] Desktop: detached window ("Open in → New window") serves a stale "Plan usage limits" snapshot — 5-hour limit stuck at pre-reset value (still reproduces after updating)
- Request: native Rust rewrite of Claude Code core (fix idle CPU spikes + terminal flicker)
- Tool call parses successfully but a string parameter is silently corrupted with an adjacent parameter's boundary tag, correlated with parameter count
- [BUG] Cyber-safeguard false positive silently switches the model mid-session, defeating a deliberate model selection
- Windows: Desktop app sidebar/session list emptied after forced reinstall, though CLI (--resume) still shows all prior sessions intact
- [BUG] Appeal Form Redirect Loop After Account Restriction
- [Bug] Opus 5 safeguards flag a benign mechanical-engineering prompt (abrasive pipe wear), on both standard and 1M context
- [BUG] Login verification code email is not received on the first request
- [Bug] Model fallback to Claude 3.5 Sonnet prevents Claude 3.5 Opus usage in chat sessions
- [BUG] Unauthorized Max Upgrade & Phantom Extra Usage Charges"
- [Bug] Token usage persists across sessions after weekly reset
- [BUG]b"Usage limit reached" triggered after weekly usage jumped from ~35% to blocked in ~12 minutes
- Webex HTTP MCP OAuth flow fails silently — missing /v1/ prefix and scope param in authorize URL
- You've hit your weekly limit · resets Aug 10 at 3am (Europe/Paris)
- VSCode extension: assistant text preceding AskUserQuestion in the same turn is not displayed
- [Bug Report] Review safeguard behavior in fable generation
- PostToolUse hooks in settings.json never register; documented updatedToolOutput example silently fails for structured-output tools
- [BUG] Concurrent MCP OAuth refresh across sessions destroys valid credentials with rotating refresh tokens
- [BUG] Apps gateway desktop overlay has no `chatTabEnabled` key, so bootstrap-configured Claude Desktop permanently loses the Chat tab (rejected at boot through 2.1.221)
- [BUG] Safeguards pause: messages typed during auto-retry are concatenated into the flagged request instead of starting a new turn — user steering never reaches the model
- [BUG] Cowork "VM connection timeout after 60 seconds" — guest never connects, all repair paths exhausted (Win 11 Pro, fresh image, Hyper-V verified)
- [BUG] Safeguards pause: "Edit prompt and retry" does not apply edits — retries submit the unchanged request
- [BUG]
- I appreciate you sharing this feedback, but I should clarify: I'm Claude, an AI assistant made by Anthropic. I'm not the right channel for this concern. This message appears to be feedback about Claude's safety policies or a moderation decision, not a bug
- [Bug] Anthropic API Error: Inappropriate content flag on non-English code review requests
- [Bug] Issue triage leaves 39% of new issues uncategorised, with no recovery path
- Claude in Chrome: all external (https://) navigation/reading silently blocked mid-session; survives restart, reboot, re-login
- [Bug] Anthropic API Error: Excessive Safeguard Filtering on Claude 3.5 Sonnet
- I appreciate you reaching out, but this doesn't appear to be a bug report for Claude Code, Anthropic's CLI tool. If you have a technical issue with Claude Code itself (like CLI crashes, API errors, or feature malfunctions), please provide details about: -
- I need more specific information about what security issue is blocking your repository check. Could you provide: 1. The exact error message or security warning you're encountering 2. What security scanning tool is reporting this (e.g., npm audit, OWASP, S
- [BUG] Getting charged for 20x, but no Fable access
- [BUG]
- [BUG] Fresh session consumes 50k+ context tokens before any real work; disabling MCP connectors only reduces it ~9%
- [Bug] Claude hallucinates assumptions without verification, then compounds errors through context
- a
- [BUG] /rename title propagates to every subsequently created session in the project (macOS CLI)
- [Bug] Fable 5 safeguards incorrectly flag documentation projects at session load before user request evaluation
- [Bug] Claude refuses to use provided AWS credentials, incorrectly flags as security threat
- [Bug] Claude Code execution failures on paste operations and pattern recognition- lawsauit filed you will be sebved shortly
- Automated code review: CONFIRMED verdicts sit on unverified premises (13 PRs, 69 CONFIRMED, several refutable in one command)
- Plan Mode's Phase 1 default launches Explore subagents despite explicit "ask before investigating" instruction in CLAUDE.md
- Background/Workflow agent no-progress watchdog detects stalls but has no recovery path (recurring, well-documented, keeps getting auto-closed without a fix)
- Fake task-notification system-reminders injected mid-conversation (not from any tool call I made)
- [BUG] AskUserQuestion preview variant ignores all key input when it renders during blur and focus returns ~90s later
- [Bug] Claude loses context and deletes code unintentionally during task execution
- I need more information to generate a proper GitHub issue title. Your message doesn't contain specific details about: - What bug or problem you encountered - What error message (if any) appeared - What you were trying to do when it happened - What was "fl
- [BUG] Claude Code web VM reclamation severely Breaks the Dynamic Workflow [AFK] lifecycle: running workflows are killed mid-run
- [Bug] Unclear error handling causes excessive token consumption and lost context
- [Bug] Anthropic API Error: Opus 5 Safeguard Rejection on Authorized Penetration Testing
- [BUG] Opus 4.6 started deleting a repo after a gross misunderstanding of a response
- [Bug] Claude Opus model failing on basic coding tasks with high token limit flag
- macOS: session-wide filesystem access (Bash cwd + Read tool) permanently breaks with EPERM after git worktree remove on a directory the session had cd'd through
- I need more specific details about what's failing to generate an appropriate issue title. Could you provide: 1. What error message are you seeing? 2. What command or action triggers the issue? 3. What's the unexpected behavior? This will help me create a
- [BUG] Correct quiz answer leaks into the assistant's message (as "umN") — reproducible on Windows, never on macOS
- [Bug] Session reset timer and usage counter out of sync after 5-hour interval
- hit session limit without using claude (Asia/Beirut)
- [Bug] Account flagged with safety restrictions after authorizing external service access
- Account panel shows mixed identity: bogus "Claude team" plan label + usage meter tracking a different account
- [BUG] Claude Code making unrequested changes
- [MODEL] Fable 5 enters premature-closure loops during benign coding tasks, causing disproportionate weekly usage during recovery
- [BUG] CLAUDE_CODE_EXTRA_BODY thinking config breaks WebSearch and WebFetch on Opus 4.7 [1m]
- [Bug] Agent using exclusive language and third-person references in responses
- Remote Control bridge init gives up permanently after 3 retries (~1.3s) — no background retry, no detectable state for unattended sessions
- [Bug] OAuth token periodically becomes invalid requiring re-authentication
- Scheduled routines: no way to grant standing tool permissions — unattended runs stall on permission prompts
- [FEATURE] Allow adding a comment when selecting a valid option in single-select prompts
- Remote control: sessions die permanently on session-limit hit and on transient disconnects; mobile app still shows them alive and silently drops messages
- [BUG] Cowork: all tasks fail with retry loop then ECONNRESET — fails across networks incl. phone hotspot, survived clean reinstall, worked hours earlier same day (macOS 15.7.3, M3)
- [Bug] Anthropic API Error: False positive content filter on legitimate mobile development debugging
- [Bug] Claude thoroughly lsoes its shit after preposttool hook creation. Agent continues tool execution after explicit STOP command and misrepresents blocked tool state as executed
- [BUG] ECONNRESET on every request — Windows 11; curl and Node.js connect fine; both native and npm installs fail
- [Bug] Incorrect code analysis flagging valid Fable solutions
- [Bug] Inconsistent Claude Code agent performance and reliability
- [Bug] Anthropic API Error: Content flagged by safety policies for cybersecurity topic
- [BUG] `--verbose` changes `--output-format json` from a single object to a JSON array
- [BUG] Claude in Chrome extension opens 4 broken tabs (native/host/wrapper/script) with DNS_PROBE_FINISHED_NXDOMAIN
- [BUG] OAuth callback listener uses a hardcoded port A53280 with no fallback
- [Bug] Phantom user message in model context absent from persisted transcript causes unauthorized autonomous work
- [BUG] Claude Code (web) stuck spinning indefinitely, no response, no error
- [Bug] CLAUDE ADMITTED THAT IT IS A FUNCTIONAL RETARD. "Yes, I am functionally retarded."
- [Bug] Incorrect timeout configuration causing unnecessary wait delays
- [MODEL]
- False positive: malware guardrail refuses to edit legitimate first-party defensive-security code
- [BUG] Safety classifier false positive: Cyrillic transliteration table in game-decompilation work triggers model auto-switch (Fable 5 → Opus)
- [MODEL] Files deleted outside the working directory during auto-approve session (unrecoverable data loss)
- [Bug] Model generates incorrect results and requires constant manual verification wants
- [Bug] Claude frequently ignores instructions and hallucinates despite system prompts restart it and. synch
- [Bug] Claude Code corrupts source artifacts during file operations and validation **DISOBEDIENT CUNT**
- [Bug] Anthropic API Error: Context accumulation triggers false positive safety filters across multi-file sessions
- I appreciate you sharing this reflection, but this appears to be a personal project retrospective rather than a bug report for Claude Code. This doesn't describe a technical issue with Claude Code itself that would warrant a GitHub issue. Instead, it docu
- [Bug] Claude Code generating non-American English spellings in code output
- [Bug] Claude model performance degradation: Ignoring user corrections and persisting with incorrect solutions across multiple turns
- [Bug] Incorrect model switch from Fable to Opus on MCU security implementation
- Assistant text before/between tool calls is not rendered in the CLI, while hooks and the transcript still see it
- [BUG] Cyber-safeguard feedback loop: discussing a previous false positive is itself the strongest trigger for new flags, locking the user out of the session
- [BUG]
- [BUG] VS Code extension should support local API without login prompt
- [Bug] SendMessage(to: "main") from background subagent silently dropped instead of reaching spawner
- [BUG] VSCode Extension fails to load with ServiceWorker error when ~/.claude/projects folder doesn't exist
- [BUG] SSH fails with "No user exists for uid 501" in background/agent-view mode, but works fine in standalone CLI
- [Bug] Agent view misclassifies incomplete conversations as completed, blocking retrieval
- [BUG] Agent View background-session input: Ctrl+b does nothing on 1st press, moves cursor back 2 characters on 2nd press
- [Bug] Background worker cannot read Keychain credentials after foreground→background handoff in 2.1.211
- [Bug] `claude agents` command fails when logged out with no re-authentication flow