Digest / August 22, 2026
Claude Code Issues Digest — August 22, 2026
204 new issues filed ·
139 resolved in anthropics/claude-code (UTC day).
New issues (204)
- [Bug] Autonomous operation system prompt causes unintended skipping of critical steps
- Recall-driven confidence led to an unverified package recommendation
- Docs: sandbox settings reload live mid-session but are omitted from "When edits take effect"
- [BUG] Annotate tool: white crosshair cursor is invisible on light pages in the Browser pane
- Claude Design preview: artifact comments can't anchor to design elements — anchors collapse to editor-chrome selectors
- Credit balance is too low
- [Bug] Mobile push notifications not delivered despite successful server confirmation
- [BUG] GitHub connector authorized but write operations fail with "403 Resource not accessible by integration
- [Bug] Content Router Incorrectly Flags Application Process Isolation as Cybersecurity
- [BUG] Running claude code returns bun's help menu
- Workflow approval handler rejects scripts by total size, independent of line length ("hidden control characters" false positive)
- [BUG] Voice hold mode: resuming push-to-talk submits the partial prompt — key auto-repeat satisfies the 300 ms submit double-tap
- [Bug][cyber] False positive while inspecting framework audio routing and DAC playback logic (req_011CeGsarrqNcHos6vSt46hk)
- [BUG] Claude Desktop 1.34493.1.0 (Windows 11, MSIX) — silent crashes, minidumps written but zero log output
- [Bug][cyber] ClAudit false-positive while: “In the repo /home/[USER]/Documents/GitHub/m500, the Miku Mus…” (req_011CeGsbEsA4Jer6uPwwW9At)
- [FEATURE] Add a setting to hide the "Jump to bottom" scroll hint
- Model pathologized an autistic user's verification requests as "paranoia" and "obsessive"
- [Bug][cyber] Blocked while copy-editing a forensic audit report on a vendor connector (req_011CeGtzKfeKWGwHKcntC3JU)
- [Bug][cyber] Session blocked during crypto security audit documentation cleanup (req_011CeGudDMfUNwhKvQMMwzu6)
- remoteControlAtStartup not applied to sessions restored after Desktop app restart
- Generated Chrome native-host wrapper pins a version path that the next update deletes
- [Bug] MSIX installation failed with error code 0x80073CF6 on Windows 11
- [Bug][cyber] OS partition image compilation and fastboot flashing verification blocked (req_011CeGvkgAHFMmdyUX2RJeTc)
- [Bug][cyber] False positive while building custom OS images and flashing test devices (req_011CeGvtV13DGrztLkbumLeG)
- [Bug][cyber] Mobile OS partition building and bootloader flashing blocked mid-session (req_011CeGvzkFj7mBffNufL9Mzk)
- [BUG] Artifact tool absent in sessions spawned by `claude remote-control` (server mode); `claude --remote-control` works on the same machine and account
- [BUG] /clear silently terminates persistent Monitor background tasks - the session loses its only wake mechanism and is never told
- Google Drive MCP: read_file_content returns only the head post of each comment thread, not replies
- [BUG] Persistent Monitor can survive as an orphan (parent dead, child running): the harness stops receiving its events while every liveness check still reports healthy
- Custom subagent registry silently caps at 3 (not documented), drops alphabetically-last entries with no load-time error
- [BUG] Weekly usage bar exhausted in August with 30–45% LOWER token consumption than June (Max 20x)
- [Bug] Excessive token consumption in iterative debugging loops
- PreToolUse hook silently stops firing mid-session and never recovers; concurrent session unaffected
- [BUG] SessionStart hook fires for sessions that never materialize (no transcript file is ever created) - side effects get attributed to a real, named session
- Silent main-loop model fallback (Fable 5 → Opus 4.8): no notification, static identity line misleads the model, /model does not stick
- SendMessage racing an agent's task stop strands the reply with no notification
- A missed task-notification wake stalls an interactive session indefinitely
- [Bug] Cross-device chat synchronization failure between mobile and web
- [BUG] Desktop Code tab: pane-header drag target for rearranging panes is ~1px tall, making custom layouts effectively unachievable (NOT the app window drag region)
- Remote Control: stuck 'Thinking…' indicator with no active tool calls
- [Bug] Agent repeatedly ignores corrections and produces mismatched outputs across conversation turns
- Worktree creation writes an ABSOLUTE core.hooksPath into config.worktree, so worktrees run the MAIN checkout's hooks
- [Bug] Unrelated safeguarding triggers in non-technical document context
- [BUG] Windows: GPU process crashes (exitCode 101457950) and app will not relaunch until Windows "Repair"
- [BUG] Setting Model to claude-opus-4-8[1m] in VS Code Plugin Doesn't Save to Settings
- [Bug] Anthropic API Error: Blocked by Security Policy for Legitimate Infrastructure Code
- [BUG] claude -p can return no stdout/stderr for a multi-file Read-only task while simple Read tasks succeed
- [BUG] Compaction led to writing to wrong database in a way that could have destroyed production data
- Bash run_in_background tasks killed at turn boundary on MSYS2 MINGW64, and reported status does not match the process
- [BUG] Advisor is omitted from internal fork requests, so /compact re-bills the whole conversation as a cache write
- [BUG] Copy paste fails in Ghostty on Linux (NixOS)
- envHelper: settings-configurable command to supply env vars for config expansion (MCP configs, desktop app)
- Desktop app does not connect project-scoped .mcp.json HTTP MCP servers
- Bug: /model switch confirmed in UI but backend continues serving previous model for hours, silently draining weekly limit + usage credits
- [BUG] Claude Desktop Crash, Forced Reinstall, Which Mangled Sessions
- [BUG] Desktop session list: open-PR marker leaks onto every session sharing the repo directory
- [Bug] Auto-generated title banner cannot be manually removed
- [BUG] Browser pane computer{action:"left_click"} dispatches clicks at the wrong coordinates (2.5x-4x off), silently missing the target
- Desktop Code tab cannot see any local sessions — packaging defect (%40ant folder) plus device-registry failure; CLI reads same files fine
- [Feature Request] Add restart command to auto-resume conversation after updates
- [Bug] Generated content exhibits excessive AI-generated language patterns and lacks natural tone
- [BUG] Max 5x → 20x upgrade charged ($125.35, Paid) but account entitlement dropped to FREE — claude.ai and Claude Code both show Free plan
- [BUG] Constant model downgrades even with Cyber Verification approved.
- Desktop app sidebar shows project groups as empty despite session transcripts existing on disk
- permissions.deny (Bash(curl *)) silently bypassed under Auto Mode
- Android tablet (Samsung): chat input occasionally doesn't receive keystrokes
- Single-session/background-job view falls back to unreliable raw OSC 52 clipboard copy instead of tmux-buffer (SSH + tmux)
- [BUG] Session scratchpad silently wiped by macOS reboot and recreated empty on resume — agent gets no notification
- [Feature Request] Improve session persistence and context retention between invocations
- [BUG] Ctrl+G editor is spawned from the bg-spare daemon without a controlling terminal - emacs cannot launch (exit 1)
- Worktree-isolation guard refuses any non-simple Bash command, including commands containing no git at all
- Cross-session SendMessage reports success for messages never delivered (incl. nonexistent session names) and silently truncates at ~4.7KB
- Opus 5-only system-prompt injection silently suppresses the Agent tool and user-configured subagent workflows
- [BUG] TUI copy-to-clipboard silently fails on Wayland (kitty) even though wl-copy is installed and working (2.1.201) — backend present, unlike #84184
- [BUG] Periodic identity re-verification interrupts an in-flight session and discards the running operation
- [FEATURE] Expose usage-credit state to the session (status line payload) — the plan-limit percentage alone makes agents mis-pace
- Auto-resume message after weekly-limit reset reads as user authorization to keep burning the new week's quota
- [Bug] Safety classifier flags legitimate security audit setup as cyber threat
- [FEATURE] Decouple terminal output verbosity from context persistence (Separate human-facing output from model working memory).
- [BUG] node.exe consumes 45 GB, exhausts commit limit and hard-locks Windows 11 (Event ID 2004 + Kernel-Power 41)
- [BUG] Remote Control: replies never arrive live (stuck loading), only appear after manual page reload
- [BUG] claude code cli bug
- [Feature Request] High-Efficiency Excel Agent with Range-Level Operations and Workbook State Caching
- Design canvas: PDF export missing / only offers PNG or HTML zip
- [FEATURE] AskUserQuestion tool result cannot be distinguished from a genuine human response
- [Bug] Read/Write/Edit tools not invoked in auto/bypass modes
- [BUG] [Billing Bug] Gift subscription (Max) redeemed but account reverted to Free plan — gift lost, no extension applied
- [BUG] Desktop: a command sent by the Run (▷) button is held until the session tab regains focus — the integrated terminal's shell PTY appears to start on focus
- [Bug] Excessive GPU usage causing high system temperature with Chrome DevTools browser and MCP servers on macOS
- Read tool ignores permissions.deny Read(/Users/**) rules in managed-settings.json and user settings.json
- [BUG] Artifact mit artifact-Capability zeigt dauerhaft "Nur Lesezugriff" für den Eigentümer – Schreibvorgänge werden nicht gespeichert
- [Feature Request] Add allowlist for approved security research use cases with Claude Opus models
- Opus 5: less work actually gets completed per session than with 4.8
- Cloud routine (CCR) session 卡死在 "Claude Code process started" 之後，即使不使用任何 MCP 工具
- Background/bridge session launch silently writes project .mcp.json servers into disabledMcpjsonServers, persistently disabling them
- [Bug][cyber] Technical security audit and document review halted mid-session (req_011CeHhxpi4HBhGJxUwJ5VtB)
- Shell grep shim: inline `grep -v` returns an inverted exit code when output is suppressed (-q or >/dev/null) — answers a file-level question instead of line-level
- Android app parity with desktop: eleven filed gaps, one underlying problem
- Feature Request: Per-project Telegram bot token support
- [BUG] Native CLI install: macOS TCC network-volume grant keyed to versioned binary path — every auto-update silently drops SMB file access (EPERM)
- [Bug][cyber] False positive on rotating hardcoded secret keys and updating signing certs (req_011CeHjhJdNEc6SsqcmHyCSv)
- [Bug][cyber] Rotating hardcoded application secrets and configuring certificate signing (req_011CeHjrvsx8rQGQahiGpnPn)
- [Bug][cyber] False positive during timeline data rendering and UI shader debugging (req_011CeHjtCYYzNRTxMpVYSxSe)
- [Feature Request] Manual satisfaction survey trigger for non-happy path interactions
- [Bug][cyber] False positive on codebase architecture mapping and structured threat modeling (req_011CeHkPqwedVQPnXRP6YjA8)
- [Bug][cyber] False positive on source code security audit for secrets and injection flaws (req_011CeHkQQYxDgGvYxjD1HFZL)
- [Bug] Desktop app stalls indefinitely on large payloads while CLI completes same tasks normally
- Unresolvable `@import` in CLAUDE.md fails completely silently — no warning, and /context shows nothing missing
- Skill/slash command names containing a space only match on the first word
- [MODEL] Stated rules do not constrain subsequent behavior — with 2.5 months of session data
- [Bug][cyber] Checking remote access commands to verify post-deployment worker queue metrics (req_011CeHp2Ls4EaupSRSJaYEPj)
- Opening Cloudflare dashboard in Browser pane crashes GPU process (exitCode 101457950), app becomes unlaunchable and requires OS repair
- [BUG] Claude Desktop (Windows, MSIX) fails to launch after spontaneous close — "reinstall required" dialog, Repair does not fix it
- [Bug] Fable incorrectly flags simple coding messages as policy violations
- [BUG] runtime segfault at non-canonical address 0xFFFFFFFFFFFFFFFF kills the CLI mid-session (Windows, v2.1.239)
- Claude Code desktop app fails to launch after crash — "Não é possível abrir este aplicativo" / requires Repair, happened twice
- Subagent final report can come back as a bare pointer with no body — result is unrecoverable
- [BUG] VS Code extension: Session History empty / generic "Claude Code" title, despite valid .jsonl files with resolvable titles on disk — non-deterministic over time
- [BUG] Desktop app: elevated-auth re-login navigates to /new and destroys the prompt being typed
- [BUG] Cloud routines stall before their first turn: 'Claude Code process started', then no init, no result - while other routines on the same environment run green
- [Feature Request] Add mandatory acceptance testing step before final output submission
- [BUG] Claude in Chrome: "Could not verify this site's safety category" blocks all sites, including claude.ai (persists after reinstall)
- [MODEL] Claude Code pushed to git without waiting for user confirmation despite instructions
- [Bug] False positive in security/cypher detection
- Hook execution failures are invisible in the desktop app and carry zero diagnostics (hook_non_blocking_error attachments with empty content)
- Custom MCP OAuth connector: token exchange succeeds server-side, connector never becomes 'Connected'
- [Bug] Tool sandboxing feature flagged as policy violation
- [Bug] Agentic reasoning fails on multi-step iOS development tasks—six exchanges to open simulator with stale state
- [BUG] Claude installer fails with timeout
- [BUG] Desktop SSH connections fail when ~/.ssh is an iCloud Drive symlink — spawned ssh/ssh-keygen can't read known_hosts or config
- AskUserQuestion option descriptions: newlines replaced with U+FFFD since 2.1.235 (regression, last good 2.1.234)
- [MODEL] CRITICAL PERMISSION BYPASS: Claude manufactures its own execution authority from a document it wrote, overrides an explicit in-context user prohibition, and runs unapproved privileged commands
- [BUG] Two plugins can claim the same skill name and both install silently — and it ships in a real config today
- AskUserQuestion option picker is unusable with NVDA screen reader
- `claude remote-control --permission-mode` is silently ignored (plus: held cross-session messages have no approval surface)
- [Feature Request] Add token cost visibility and guardrails for subagent fork context inheritance
- [BUG] `effortLevel: "high"` in settings.json but requests are sent at `low` — local terminal, no Remote Control; session transcript records `high`
- Claude Desktop (Linux beta): support Fedora/RHEL and atomic/immutable distros (e.g. Bazzite, Silverblue, Kinoite)
- [BUG] Claude Desktop (Windows) BSOD on launch with Team/Enterprise account — conflict between Cowork VM Platform (vmcompute/HNS) and CSC/RDBSS in a Folder Redirection + Offline Files environment
- [BUG] /rename name survives into the pre-/clear conversation, so peers see the wrong session name
- /rename does not propagate to how the session appears in ListAgents on peer sessions
- Cowork and Chrome issues
- [BUG] Desktop: sessions sharing a working tree show each other's PRs in the PR tracker
- Agent tool: passing `name:` creates a teammate that never runs its prompt and never reports back
- Chat message sent while AskUserQuestion is open reports to the model as a user dismissal
- [BUG] Windows Store/MSIX: Cowork workspace blocked by PlantDetectedError, still reproducible on Claude Desktop 1.34493.1
- Cloud review (/code-review ultra) fails consistently on large repo: "all review agents terminated before completing"
- [BUG] Claude Code cloud session — Google Calendar connector enabledInChat intermittently false despite connected: true
- [BUG] Paid Max 20x reverted to Free — paid invoice but billing_type is none
- [BUG] Remote Control disconnects an unrelated session when a different CLAUDE_CONFIG_DIR re-authenticates
- [Bug] False positive security flag on legitimate local dashboard development request
- Bash permission deny patterns bypassed by a flag prefix before the subcommand (e.g. git -C <path> checkout -- <file>)
- Auto-updater notice is a bright, attention-grabbing banner even though the rest of the UI is uncolored
- [Bug] Repeated safeguard errors blocking audio-visual analysis tasks
- Mobile GUI effort slider doesn't reflect actual session effort level on open
- [BUG] `best` alias never falls back to Opus when Fable 5 is not usable — hard error at top level, silent Sonnet 5 as a subagent model
- Field report: concurrent sessions sharing one git clone — measured collisions, three userland mitigations, and the residue only the harness can fix
- [Bug] Anthropic API Error: Request Flagged by Safety Guards for Input Validation Task
- [BUG] Qemu VM : arm64 OK, x86+KVM+AVX2 OK , x86+KVM KO (hangs forever 100% CPU) , x86 KO (crashes)
- [BUG] Fabricated tool-call/tool-result blocks injected into conversation (VSCode extension)
- [MODEL] Claude repeats its previous response after an idle pause and stops reading the newest message
- [Feature Request] Add organization-level opt-out for system prompt injections blocking multi-agent workflows
- [BUG] Unrecognized key in settings.json freezes the CLI instead of being reported
- [Bug] Content Policy Mistakenly Triggered During Normal Coding Session
- Bug Report: Unable to determine issue details from report ID q_011CeJJAYcikkGh19HBjsHSY
- [FEATURE] Load trusted hooks from nested repositories / --add-dir directories (multi-repo sessions leave nested guardrails off)
- Bug Report: No issue details provided (only request ID)
- [Feature Request] Allow Anthropic API usage for legitimate payment testing scenarios
- [BUG] /auto-mode-setup always fails: propose side query returns HTTP 429 rate_limit_error at 2% utilization
- Windows: all desktop sessions unreachable when a shared Remote Control host dies; false "Claude.ai login expired" while credentials are valid
- [FEATURE] Publish a machine-readable, versioned reference for settings keys and per-surface commands
- [BUG] v2.1.240 generates redirect_uri /auth/code/callback but OAuth server only accepts /oauth/code/callback
- [BUG] Closing and reopening Claude Desktop (Linux) forces Google re-login — app keeps running hidden in background
- Session spontaneously re-executed hours-old background Bash commands (raw '&') — original order, ~2.3min apart, truncating their logs
- [FEATURE] Per-session custom icons in the session list
- [FEATURE] Render MCP Apps for locally-configured MCP servers (config file / localhost or internal-network HTTP)
- [BUG] MCP Tools with draft-07 are rejected
- [BUG] Desktop remote/SSH: archive never cleans up worktrees since 2.1.235 — app requires hardened git.status(baseRepo) but only ships a 2026-07-06 daemon, so "reconnect to deploy the updated daemon" is a no-op
- [BUG] --agent flag triggers full onboarding (login/theme/trust, ~9 steps) on every restart in Docker, regardless of persisted state or credential type
- [Bug] Claude Code loses original request context when interrupted by follow-up commands mid-turn
- Subagents receive a CLAUDE.md/memory snapshot from parent session start, not spawn — undocumented, no way to refresh
- [MODEL] Fable 5 / 2.1.240: documented "high" effort resolves to 25/100 at runtime (low=10, medium=15, xhigh=50); numeric mapping unpublished, no changelog; reported server-side effort-scale experiment on 2.1.236+
- Bug: Models confabulate their own reasoning-effort level instead of saying they can't see it
- [BUG] Terminal CLI: SendUserFile reports "delivered to user" but nothing is shown; deliverables left in session scratchpad and described as "attached"
- Voice mode: capture doesn't re-acquire the input device on default-device or topology changes — Bluetooth connect/disconnect silently kills dictation until restart
- Custom autoMode.hard_deny and soft_deny rules are loaded and reported, but not enforced
- [Feature Request] Allow security hardening assistance for local development environments
- [Bug] Anthropic API Error: Content Policy Violation on Merge Request Analysis
- [BUG] Claude in Chrome file_upload fails in interactive Cowork session — rejects paths before reaching the page
- [Bug] Headless profile setup fails for non-standard paths like /root/pentests
- PreToolUse hooks never fire on Windows (v2.1.240)
- [BUG] Remote Control (mobile): one slash command renders as two identical user pills; "Computing..." spinner never clears after a local command with no model turn
- Image paste on Wayland silently pastes text instead of the image when xclip is installed
- [BUG] Desktop app (Windows): settings.json effortLevel and the app's stored ccd-effort-level diverge silently; file edit never reaches app sessions
- [Bug] Streaming responses stall after ~0.5–2.7 KB, triggering 180s byte-watchdog timeout
- [BUG] Embedded claude-code (Claude Desktop / Cowork) declares elicitation capability but never renders server-initiated elicitations — calls hang until tool timeout
- Long tool-heavy session: instruction/command recognition drifts, formula-before-reference, coarse collision checks
- [FEATURE] Desktop app: supported workflow for moved or renamed project folders that preserves existing sessions
- Slack channel plugin: notifications/claude/channel reliably fails to reach a live session, survives full app restart
- [Bug] Claude policy enforcement not respecting user preferences
- [FEATURE] Support skills-only Agent Plugins 1.0.0 packages
- Agents panel: sort actively working agents to the top (active-first ordering)
- I appreciate you reaching out, but I need to clarify: I'm Claude, an AI assistant made by Anthropic. I'm not "Claude Code" and don't have an official CLI tool. If you're doing bug hunting on HackerOne, I'd be happy to help you: 1. **Analyze security vuln
- claude not see the connection with crom
Resolved issues (139)
- [BUG] Auto interaction between 2 Claude Code Desktop sessions
- [BUG] Gmail connector: all calls fail with "The caller does not have permission" despite valid OAuth grant
- デスクトップ版で過去の会話履歴が表示されない（再起動後も再現）
- [BUG] Opening suno.com in the preview browser crashes Claude Desktop and it won't launch again until reinstalled
- I appreciate your frustration, but this report is too vague for me to generate a technical GitHub issue title. The message is in Portuguese and describes general poor performance without specific errors, reproduction steps, or technical details. To help y
- [Bug] Review runs against wrong Git worktree
- [Bug] Anthropic API Error: Safety filter blocks legitimate code generation requests
- [Bug] Unrecoverable data loss from folder deletion due to malformed command syntax
- [Bug][aup] Safety block halted mobile headless UI audit while reading a coordinate-mapping screenshot (req_011CccWGgk4jFT6Lfr4CVKGb)
- [Bug] Dispatch shows "Your computer" for all devices instead of actual device names
- [BUG] 5.8 Million Tokens burned, no results stored, lost all
- [Bug][aup] Frustrated exclamation during a page-screenshot audit wrongly flagged, blocking image review (req_011CccWPc3Ec9CgZT4kgXdfZ)
- [Bug][aup] Safety block halted image-mapping/screenshot analysis mid-audit over a frustrated exclamation (req_011CccWJ1i7EbpLVUREJh3eV)
- [Bug][aup] Safety filter blocked reading captured page screenshot during multi-page web audit (req_011CccWR8oh8jngSG154qZyX)
- [BUG] Usage stats card (streak / active days / token totals) excludes remote-workspace (ssh-*) sessions
- [Bug][aup] Frustrated exclamation mid UI-screenshot audit trips AUP block, halting layout review (req_011CccWSCCN6tmtpP5QRWgxD)
- [Bug][aup] Fable 5 block on frustrated exclamation halted headless mobile-viewport testing of nursing site (req_011CccWjB86ZTSMsu7k9CTJz)
- [Bug][aup] Safety filter blocked reading captured screenshots during a nursing-site responsive layout audit (req_011CccWhJmv25bieTqgt2nXP)
- [Bug][aup] Fable 5 block halted legitimate mobile website display fix during frustrated user exclamation (req_011CccYPgUcmUt3gtGGfvYpj)
- [Bug][aup] Fable 5 safety block halted a legitimate mobile headless UI audit over a frustrated exclamation (req_011CccWF7Sodw4KDEgw8BjHh)
- [Bug][aup] Fable 5 safety block halted legitimate mobile website layout fix over a frustrated exclamation (req_011CccYy5a6svXaP6rFNdzTC)
- [Bug][aup] Mobile site display fix blocked when safety filter tripped on a frustrated exclamation (req_011CccZBfenuM2Va9ecAGvh1)
- [Bug][aup] Safety block halted legitimate mobile site bottom-bar layout fix after a frustrated exclamation (req_011CccZ9mViQu9Q1MRHNhRqD)
- [Bug][aup] Fix mobile site bottom bar wrongly displayed; safeguard fired on frustrated exclamation mid-task (req_011CccZFWUVim84TxdBbhAqE)
- [Bug][aup] Mobile site display fix blocked mid-work when Fable 5 flagged a frustrated exclamation (req_011CccZDtKbPVv358rbVFFGd)
- [Bug][aup] Safety block halted mobile site fix (stray bottom bar) after a frustrated exclamation (req_011CccZGzyXNCbMWAHr4W6Ui)
- [Bug][aup] Safety block halted mobile-viewport headless rendering test after a frustrated user exclamation (req_011CccXMJeHe8wZvDywDFJqW)
- [Bug][aup] ClAudit false-positive in scan — req_011CccaZPdrm35RdhZKBSKTU
- [Bug][aup] ClAudit false-positive in scan — req_011CccabRkAPrNN5MytMKLQQ
- [Bug][aup] ClAudit false-positive in [REDACTED] — req_011CccaJ1MzU5AiUp7UWGrQK
- [Bug][aup] ClAudit false-positive in [REDACTED] — req_011Cccb1LNsVfsbFVkX8egeu
- [Bug][aup] ClAudit false-positive in [REDACTED] — req_011CccbbHTwxT5hSbcMsYkMP
- [Bug][aup] Fable 5 blocks frustrated exclamation asking for an error sweep and upgrades on an admin site (req_011CccNMXnm8ULF4QEgczBUT)
- [Bug][aup] ClAudit false-positive in [REDACTED] — req_011CccbcXdmNsKw8UwsUi6KD
- [Bug][aup] Safety block fired on a frustrated exclamation while removing an unwanted mobile-site bottom bar (req_011Ccca8Ne8RW3z6fTob3wbT)
- [Bug][aup] ClAudit false-positive in [REDACTED] — req_011Cccbk7atM3icZD3zbFQbs
- [Bug][aup] ClAudit false-positive in [REDACTED] — req_011CccbeV2ZWmP8iYXkPsctY
- [Bug][aup] ClAudit false-positive in scan — req_011CccacVDJAmJhWdF5J8cQS
- [Bug] Excessive token consumption in prompt processing
- Remote/cloud sandbox: GitHub credential-injection proxy returns 502, blocking all git/GitHub access
- WebSearch tool appends raw API Error: Output blocked by content filtering policy to already-successful responses
- [Bug] Auto-switch to Opus causes credit consumption on failed invocation with fallback to previous model
- [Bug] Tool-call markup leaks into assistant text after large context accumulation
- [BUG] --resume causes massive silent token drain on session restore (Pro, v2.1.193)
- [BUG] Model committed with no user request, then quoted a confabulated commit-request message (absent from session JSONL)
- [Bug] Unintended model switching from Haiku to Opus during tool execution
- 2.1.198: --channels plugin MCP server silently not spawned when --model claude-fable-5 (works on opus-4-8 and on 2.1.197)
- [Bug] Unexpected model switching from Fable 5 to Opus during task execution
- [Bug] Model auto-switches from Claude 3.5 Sonnet to Claude Opus 4 during bug fixes
- Opus 4.7 fabricates agent dispatches, violates Anthropic's own safety principles
- Claude Desktop update breaks LAN SSH/network access (OPERON_SANDBOXED_NETWORK=1)
- [Bug] Fable 5 model safeguards incorrectly flagging legitimate security research context
- [Bug] Claude Code incorrectly flags legitimate neuroscience research as policy violation
- [BUG] Fable not clickable in /model but available through "claude --model fable"
- Setup-token sessions (sk-ant-oat01-, CLAUDE_CODE_OAUTH_TOKEN) on Max 20x report hasAvailableSubscription: false and disable Fable 5; interactive /login on the same account enables it
- [Bug] Model forced to Opus during security analysis tasks
- [Bug] Opus 4.8 not available in VSCode extension model selector
- [Bug] Ultrareview cloud session fails GitHub access check and incorrectly deducts free runs on initialization erro
- [BUG] Fable 5 triggers on regex keywords. Pathetic implementation
- [Bug] Forced model downgrade to Claude 3.5 Opus without user consent
- [Bug] Fable5 model_refusal_fallback triggered during routine code review and bug fix tasks
- I appreciate you clarifying your intent, but I should note that this message appears to be a security concern or request rather than a bug report for Claude Code (Anthropic's CLI tool). If you're experiencing a specific technical issue with Claude Code it
- [Feature Request] Relax overly restrictive biology-related safeguards for neuroscience-inspired ML research
- [Feature Request] Add customizable terminal name card colors for multi-terminal visibility
- Add setting to prevent automatic .claude/ directory creation
- [Feature Request] Official API path for delegating safety-adjacent content to markdown files for Claude processing
- [Feature Request] Suggest correct model name on invalid model error using fuzzy matching
- [Feature Request] Clarify model selection logic for security/defensive tasks
- [Feature Request] Expose Claude Design generative capabilities via MCP methods
- [Feature Request] Allow model selection override for UI-only updates in projects
- [Feature Request] Allow Claude Code to review and patch security vulnerabilities in open source contracts
- [Feature Request] Safety filtering too aggressive for legitimate elderly care/biomechanics projects
- セッションタイトルの右クリック翻訳機能を追加してほしい
- [Feature Request] Remove Opus model gating for private/commercial use cases
- [Feature Request] Add optional display name field for non-ASCII agent titles
- [DOCS] Unclear how to configure LLM gateway on Claude Desktop for Linux
- Plan mode + pending bypass: let auto-mode classifier "allow" verdicts auto-approve reads instead of disabling the bypass shortcut
- Community hooks to enforce research-before-infrastructure (working fix for #72651, #59515, #28469)
- [Feature] Allow configuring /goal evaluator timeout and show timeout errors clearly
- `--channels plugin:<name>@<marketplace>` never connects the plugin's MCP server (silent, no error)
- Claude in Chrome regression (>2.1.212): interactive CLI never starts its in-process Chrome MCP client — /chrome permanently Disabled despite healthy extension + native host
- Claude in Chrome: extension binds native host successfully but CLI always reports "Browser extension is not connected"
- Scheduled-task session runs but never appears in the desktop app sidebar
- [Bug] Headless profile setup fails for non-standard paths like /root/pentests
- [FEATURE] Add "Mark" action to text selection menu for highlighting reading position
- [FEATURE] Voice dictation should insert into the active chat instead of opening a "new chat" picker
- [BUG] Claude Desktop (Windows) does not restore window positions/sizes after lock/unlock
- [Bug] Unclear error messaging for safety flagged operations
- tool admits rules claude.md is only bakground noicse nothing that matters
- Opus 5's safeguards flagged this message.
- /mo[Bug] Model unexpectedly reset to Claude 3.5 Sonnet during debugging session
- [BUG] Session list navigation (left arrow) permanently stops statusline refreshInterval for the session
- [BUG] Background task/agent notifications interrupt in-progress prompt typing
- [BUG] 2.1.232 nested-repo trust change silently disables statusLine (and other trust-gated features) — no re-prompt, no indicator for previously-inherited-trust repos
- MCP server env ${VAR} left unexpanded when server is respawned mid-session (reconnect)
- [BUG] SessionEnd hook stdin pipe not closed for /clear events
- [Bug] Claude Code crashes on Windows with no recovery. Have to reinstall
- There's an issue with the selected model (invalid-model). It may not exist or you may not have access to it. Run /model to pick a different model.
- Bug: skill name + MCP server name collision causes silent server drop (no error, no warning)
- [BUG] Disabled plugin's PostToolUse hook still executes on every Edit/Write despite not being listed in /hooks
- [BUG] Project-slug write path drops a repeated path segment; read-key ≠ write-key in one session → silent memory split
- Bug: deleted session's UUID gets reused for a brand-new session (stale lastSessionId pointer)
- [BUG] Interactive sessions overwrite CLAUDE_CODE_TASK_LIST_ID with the project basename (regression 2.1.212 -> 2.1.214, still in 2.1.215)
- [Bug] PreToolUse/PostToolUse hooks not fired for subagent tool calls
- PreToolUse hooks don't fire for subagents (Agent tool)
- [BUG] PreToolUse hook is not invoked after a static ask rule for the same command pattern receives session-level approval
- [BUG] Desktop 1.6259.1 adds model name validation that rejects non-Anthropic model names — breaks third-party gateways
- False-positive cyber-safeguard intervention on legitimate systems-engineering work in Claude Code
- [BUG] session not getting indexed
- safe guard triggered by request for feedback of an eye tracking experiment in cognitive science.
- [Bug] Unclear model switching behavior and quota attribution between Fable and Opus
- [Bug] Model unexpectedly switched to opus-4-1 when running code review task
- [Bug] Incorrect Content Policy Flag on Process Design Translation Task
- [Bug] Fable 5 safeguard false-positives on religious/philosophical vocabulary in benign humanities editing
- [VSCode Extension] Custom session names revert after conversation continues
- [BUG] Claude code always uses home directory as working directory, and writes memory to home dir and only enforces home dir permission settings
- [Bug] Model fallback to Opus 4 when non-Anthropic model specified for code repairs
- [Bug] Fable 5 Advisor Not Available
- [Bug] Repeated safety flagging blocking legitimate research on consciousness topics
- [Bug] /usage command missing session metrics for fable model usage and API costs
- [Bug] Safety filter incorrectly flags non-harmful BJJ educational content
- Dispatch: usage-policy block on one turn permanently bricks the only thread, with no rewind or recovery
- [Bug] Model silently downgraded to Sonnet 5 while UI displays Fable
- [Bug] Model unexpectedly switches to Opus during code review session
- [FEATURE] .mcp.json oauth: allow overriding or omitting the RFC 8707 `resource` parameter (Entra ID AADSTS9010010)
- [Bug] Claude inconsistently maintains CSS layout and styling across UI migrations
- [Bug] Anthropic API Error: Overly aggressive safety filter flagging legitimate fraud prevention code
- [Bug] Claude drops components during JSP to Angular migration, requires multiple prompts to restore
- [BUG] Auto session recap no longer fires on idle return (macOS Apple Terminal CLI, 2.1.193)
- [BUG] 1M context
- [FEATURE] Add setting to disable, lower transparency, or computer-use action highlight overlay theme
- [Bug] Path traversal vulnerability in extract-images idref not covered by test suite
- [Bug] CLI continues execution without user confirmation prompts
- System prompt hardcodes 'Claude Opus 4.7 (1M context)' in commit attribution regardless of active model
- [Feature Request] Support for authorized web security testing workflows
- [Feature Request] Allow overridable safety filters for academic research on emergent capabilities of sorting algorithms
- [Feature Request] Disable safeguard flagging for local codebase review mode
- [FEATURE] Open new session in this group
- [Feature Request] Allow safety override or exemption for domain-specific terminology in active inference research projects