Digest / August 3, 2026
Claude Code Issues Digest — August 3, 2026
252 new issues filed ·
117 resolved in anthropics/claude-code (UTC day).
New issues (252)
- VS Code extension: /mcp reconnect and MCP servers panel intermittently fail with "terminal is still starting up or is showing another view"
- Cost/token usage metrics missing documented agent.name/query_source="subagent" attribution for Agent-tool (Task) dispatches
- [MODEL] Fabricated user turns right after a completion summary — short session, no compaction, no subagents (Cowork desktop app)
- Provide a setting to disable the claude.ai/code session link in PR bodies and commit trailers
- [Bug] Claude 3.5 Fable model selection blocked for local MUD/MUSH communication application
- Feature request: persistent voice conversations with Claude Code (no idle disconnects)
- Fable 5 unreachable on Max 20x: new sessions always refused, only pre-existing sessions get it, and the switch back is silent
- [BUG] Cyber-safeguard false positives on scientific computing session (IR spectrometer calibration) — fires on accumulated context, blocks on both Opus 5 and Opus 4.8
- [FEATURE] Support /desktop session handoff on Windows ARM64
- Allow customizing the ExitPlanMode approval button text
- ultracode/enableWorkflows do not suppress the default-off delegation lines (2.1.220)
- [Bug] Anthropic API Error: Safety Policy Flagged Legitimate Coding Request
- Feature request: allow custom ordering / pinning in the "/" slash-command autocomplete menu
- [Bug] Model hallucinated fabricated system-reminder with fake MCP instructions and false security warning
- [BUG] Disconnect From extension claude code for vscode
- [FEATURE] Fullscreen TUI Ctrl+R only searches the last 100 prompts — restore classic renderer's full-history exact search
- (withdrawn by author)
- [Feature Request] Add option to redisplay "How's Claude doing?" prompt after dismissal
- Desktop app sessions keep loading old plugin version after 'claude plugin update'
- [FEATURE] Session timer that excludes idle time to determine how long actual work took in a session.
- [BUG] Backspace deletes an entire grapheme cluster, wiping a whole Thai syllable in one press (v2.1.220)
- [Billing Bug] Two successful payments (Max 20x + Pro) not activating — account stuck on Free tie
- [Bug] Model downgrade on cryptography-related code requests without user consent
- Agent misreads silent/empty tool output as a confirmed negative result
- [BUG] Claude Desktop (Linux): GPU process launch retries at ~9,000/sec, writing 346 GB to /var/log/syslog in 24 hours
- [BUG] Agent isolation:"worktree" locks Bash/Edit to the wrong repo when the sub-agent's task is in a different repo
- [FEATURE] Configurable naming convention for auto-generated session names
- [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
- Remote MCP connector: response delivered to wrong concurrent tool call (cross-subagent payload misdelivery); rightful caller times out after 180s
- Fable 5: modifies code and restarts a service before delivering the requested report / getting approval
- Auto-mode permission classifier reads the command string, so a sensitive action moved into a script file isn't gated
- [BUG] Transient `api_error` mid-turn forks the transcript parentUuid chain — the retried reply is silently orphaned; later messages follow the error branch (surfaced by Rewind in Desktop)
- [FEATURE]
- [FEATURE] Claude in Chrome: pin the MCP tab group to a designated Chrome window
- [Bug] Subagents loop 9 extra turns after end_turn, returning filler text instead of report
- [BUG] Claude Code unexpectedly clears OAuth tokens and requires full re-authentication before refreshTokenExpiresAt
- Chrome extension never connects via Claude Code native messaging host (only Desktop app host works)
- Session lost previously-established case context mid-task, forcing hours of re-explanation
- [BUG] claude.exe crash on Windows 11 Arm64 with 0xC0000005
- [BUG] Windows: maximizing a standalone PowerShell/cmd console window leaves a blank right margin and wraps at the pre-maximize width
- [FEATURE] VS Code extension: retrieve sessions from a different projects/windows
- [BUG] php-lsp plugin (intelephense) ignores intelephense.config.json / project-level settings for environment.includePaths
- chrome-native-host crashes on every invocation (Windows, ERR_INVALID_ARG_TYPE on named pipe listen)
- Model switch in RC/remote app doesn't propagate to local client until an RC-originated message is sent
- [BUG] Windows MSIX: in-app Browser crash leaves package in `Modified, NeedsRemediation` — Repair fails entirely, requiring full removal (container loss) and a reinstall that hits 0x80073CF6
- Session goal explicitly requesting opencode-cli/agy-cli subagent delegation was not consistently followed
- [BUG] Windows desktop: a `:N` line suffix on a markdown-file link silently kills the link, while the same suffix works on code files
- [FEATURE] Desktop file viewer: allow editing `.md` files in place, as code files already allow
- Support importing/backfilling local CLI sessions into the session registry
- [BUG] GPU process crash (exitCode 101457950) kills host app shell during live browser preview
- [BUG] Rate-limit dialog blocks the session with a single option whose only effect is dismissal
- [BUG] Left/Right arrow key history navigation causes unrecoverable loss of in-progress work
- [Bug] Unclear error when Fable model selection fails
- [BUG] Desktop app closes question prompt in the middle of writing an answer
- Feature request: visual nudge mode — select an on-screen element and arrow-key its position
- [BUG] claude install creates broken symlink: %h not expanded in path
- [MODEL]
- CLI runs its session-teardown path mid-tool-call; the lost result is mislabelled toolDenialKind:"user-rejected" (VSCode ext 2.1.220, Windows) — same class as #83358, different trigger
- [BUG] Fullscreen TUI: bare mouse movement scrolls the viewport after any wheel notch — SGR reports with both the wheel and motion bits set are treated as wheel notches (urxvt)
- [Bug] Anthropic API Error: Usage Policy violation on valid task requests
- [FEATURE] Named model presets: multiple /model picker entries with preset effort levels
- Stop hook (stop-hook-git-check.sh) tells the agent to rewrite published history after a PR merges
- Please enable modifyOtherKeys in xterm to allow Ctrl-Enter and Shift-Enter
- [BUG] Scheduled tasks still ignore permissions.defaultMode on 2.1.204 — #40470 closed as completed but reproduces (locked, cannot comment)
- [Bug] Unexpected model switch to Opus without user action
- [BUG] CoworkVMService crashes during use (HcsShutdownComputeSystem 0x80070032), cannot self-recover due to ACL restriction
- MCP connect has no jitter under concurrent sessions, and ToolSearch then reports the failed servers as nonexistent
- [Bug] Unexpected model downgrade to Claude 4.8 Opus during Kaggle competition usage
- Sandbox masking placeholders are created in the project working directory instead of $HOME, appearing as untracked files on every Bash command
- [FEATURE] Persistent setting to disable auto-attach of active editor file/selection
- CronCreate recurring tasks fire at a fixed per-session phase offset, not at the wall-clock cron time (hourly `0 * * * *` fires 15 to 28 min past the hour, and the offset scales with the period)
- [FEATURE] Feature Request: Ability to reload/refresh skill definitions during a session
- [Bug] Remote-control-by-default causes cross-project session hijacking + background-job misclassification (recurrence of #64635, #64659, #64640)
- [DOCS] State whether project-scope hooks are workspace-trust-gated, and what happens to that gate under `--print`
- [Bug] Text corruption when copying large terminal selections in KDE Konsole
- [Bug] Model Generates Fabricated Future Results in Long-Form Multi-turn Sessions
- [BUG] "Claude in Chrome" website permission ("Allow all websites") not honored — all navigation outside one previously-used domain blocked
- Fable 5 blocking every prompt for codebase analysis
- [BUG] Desktop app 1.24012.9 (Windows/MS Store): in-app per-message Read Aloud play button does nothing after update
- [BUG] Markdown link labels never clickable in Orca/tmux — no OSC 8 emitted for unrecognized TERM_PROGRAM, FORCE_HYPERLINK=1 ignored
- [BUG] Desktop app Code tab unresponsive - no response after sending message
- [MODEL] Measurable quality regression in Claude generation 5 (Fable 5 / Opus 5 / Sonnet 5): worse nonsense detection, ~2x verbosity, under-disclosed model fallback (Fable 5 → Opus 4.8) — reproducible measurements
- [Bug] Overly aggressive safety filtering persists across new conversations after security research sessions
- statusLine always reports the main session, even when a subagent or teammate transcript is focused
- Agent triggered an unauthorized production deploy via a routine `git push` (confirmation gating misses commands that are remote triggers)
- [BUG] Uploading skills to Claude Desktop running via gateway only transfers the SKILL.MD file not other files or folders
- Gateway model discovery never fires when using apiKeyHelper (works with ANTHROPIC_AUTH_TOKEN)
- /code-review: structured output is hard to read, and the format changes with the effort level
- Unexpected behavior when handling special characters in filenames
- Claude in Chrome: extension connected & signed in, but MCP tools report 'not connected' and list_connected_browsers returns [] after mid-session account re-login
- [BUG] Make it easy to see what's hidden or that feature is pretty much useless
- Claude should behave like a tool, not perform a personality
- Feature request: periodic spinner verb rotation during long-running operations
- Agent tool: model:"fable" parameter appears not respected — subagent runs on Sonnet instead
- [Bug] Push notifications fail to deliver to Android device despite tool reporting success
- Session picker's "widen to all worktrees" (Ctrl+W) has no browser-safe alternative in code-server/vscode.dev
- False positive: safeguards flagged legitimate FCM push-token testing (auto-switched Fable 5 → Opus 5)
- [MODEL] Sonnet 5
- QuickBooks Online connector missing time-activity/timesheet creation tool
- No action clears the prompt buffer without a side effect; chat:clearInput is a redraw that arms /clear
- QuickBooks Online connector has no support for Classes anywhere in its toolset
- [Bug] Anthropic API Error: Content filtered by safety policies blocking legitimate coding tasks
- Agent makes unrequested out-of-scope changes; one took a working page down (and clean type-check reported as verification)
- claude-api skill inlines ~500KB (~130K+ tokens) of reference docs into every invocation
- [BUG] Custom agent used as an in-process teammate does not inherit project MCP tools
- Desktop: render the configured statusLine in the Code tab, or allow pinning the usage ring
- User feedback: per-action confirmation on public-post submissions felt like friction during a time-sensitive session
- [BUG] Agent tool model: "opus" override spawns a subagent that self-reports as Sonnet 5
- [FEATURE] Allow nushell for default shell
- [Bug] Bash commands with ! operator execute in wrong directory in git worktree
- First-run on Windows blocks non-developers: app demands Git despite docs saying it's optional — bundle or offer to install it
- Claude-in-Chrome extension appears stuck on 1.0.84, contradicts "fixed in 1.1.9310" note on #39707
- Desktop app: sessions that ran in a git worktree can't be resumed after restart — app silently rebinds the conversation to a new empty session, orphaning the transcript
- Prompt cache repeatedly dropped mid-session (17x in 3h), causing ~10.4M tokens of redundant cache writes
- Session rename functionality not accessible to Claude agent during workflow
- Statusline payload reports wrong effort level after /effort change
- [Bug] CLI outputs non-ASCII text graphics despite user disabling request
- [Bug] Incorrect Opus 5 model requirement flagged for non-API usage
- Agent-type roster is not user-controllable: bundled plugins + cross-project memory occupy 111.5k tokens in every session, and three documented workarounds have no effect
- [BUG] I can't login to claude code. Don't get verification email.
- [BUG] Cowork Browser pane: concurrent background sessions crash the GPU process and take down the whole app
- [BUG] Every Opus subagent dies at 0 tokens on xhigh effort — spawn path hard-sets thinking:{type:"disabled"} while still sending effort:xhigh
- $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
- Remote-control server/bridge pair survives after "Connection closed" is logged — orphaned for 13+ hours
- [BUG] Worktree existence blocks deleting agent session
- Agents view: sessions with an open ask filed under Working (buckets on state, not tempo); no bucket for state:failed; resumed sessions double-counted
- MCP race: first API request is serialized before HTTP MCP servers finish connecting — tool_use fails with "No such tool available" (breaks --resume)
- `/code-review` can silently destroy uncommitted work — a premature "completed" status opens the write race, and the harness's own safeguard is told to hide it
- claude-api skill over-triggers on any Claude model mention, injecting ~400k+ tokens of unrelated SDK docs into context
- Tool-call summary line clamps to 2 lines with no expand toggle
- [FEATURE] paste screenshot without text
- update_trigger API rejects any model change on scheduled Routines (model_update_disabled)
- [BUG] Desktop third-party mode (Vertex/Bedrock): SSH environment never appears — sshApisAvailable is hardcoded false in the bundled renderer
- Workflow progress view: phase dot colours are undocumented, and two runs in identical states render different colours
- [BUG] Scheduled cloud routines (CCR) consistently fail to `git push`
- [Bug] Anthropic API Error: Inappropriate content flag on benign technical documentation
- [BUG] /design-sync bundled skill silently disabled by server-side flag while DesignSync tool stays enabled — broke a scheduled cloud routine mid-flight
- Cowork: sdk_install fails with EACCES when Claude Desktop is installed as MSIX (Store) package
- [FEATURE] Permission dialog: make the digit for "No" stable across the 2-option and 3-option variants (implementation TBD)
- [BUG] Plan Mode restriction not consistently enforced across turns
- [BUG] Native installer writes malformed PATH entry on Windows (C\Users\... instead of C:\Users\...)
- [FEAT] Session status in the desktop sidebar is encoded by colour alone — add a shape/glyph channel
- Gmail MCP create_draft: description says attachments are unsupported, but the schema documents them and the call succeeds
- "Commit changes" cards are duplicated and report branch-vs-main instead of the uncommitted diff
- [BUG] Claude Desktop for Linux holds a 'Capturing' idle inhibitor indefinitely — screen never blanks until the app quits
- [Bug] Cyber verification credentials not working in lab test environment
- [Bug] Assistant generates unverified claims and false documentation without source validation
- [FEATURE] Search the sessions list in claude agents
- Notification hook never fires in VS Code native panel (permission_prompt & idle_prompt)
- [Bug] Consistent API Request Flagging Throughout Week
- [BUG] Weekly + Fable quota jumped 0% -> 50%/100% with account idle (Max 20x, after Jul 31 reset)
- [BUG] Artifact CDN serves stale cached version — updates via routine/Artifact API not visible to viewers
- [BUG] Fable 5 promotional notice compares against "Opus 4.8" - should be Opus 5 (current Opus)
- I appreciate you sharing this detailed context
- I appreciate you sharing this detailed context
- Windows MSIX: session cards can never be saved when %APPDATA%\Claude contains a reparse point — one-line manifest fix (RoamingAppData\Claude missing from FileSystemWriteVirtualization exclusions)
- [BUG] Folded gateway-discovered model rows ignore gateway metadata and show Anthropic list pricing
- [Bug] Account-level connectors cannot be removed on any surface — desktop, web, or CLI
- [Feature Request] Add ability to disable automatic extra credit usage and view per-team-member credit consumption
- [BUG] Claude Code persists Bash permission approvals to project .claude/settings.json instead of .claude/settings.local.json
- Agent loop: a parallel subagent result that lands after the final turn is never consumed — and the 2.1.218 nudge does not cover it (only fires on empty turns)
- [BUG] Desktop app: sessions inside a sidebar group show no timestamps and are not sorted by recency
- [Bug] Tool hallucinating file contents and making false claims about read operations
- [BUG] ToolSearch returns incomplete MCP tool list on first query, causing false 'tools removed' claims
- 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 execution produces incorrect results consistently IT FUCKING LIES ONLY
- [BUG] Claude Code double-Escape to clear prompt causes CLI to hang on macOS
- ok [Bug] Claude repeats output twice in same window
- [Bug] Tool repeatedly fails on simple list operations, wastes tokens and time TASKE DONE BY CHAT GPT IN 2 MINUTES
- [BUG] design-sync: two bugs in tsconfigPathsPlugin — the "@/*" path alias breaks the tsconfig parse, and the resolver returns directories
- A subagent resumed with `SendMessage` reports its completion to the main conversation instead of the subagent that resumed it
- [BUG] Scheduled tasks silently disappear: recurring task vanished twice in 5 days while spent one-time tasks persist
- [Bug] Terminal display shows blank output when tmux not installed
- Feature request: attach an existing session as a teammate (cross-session messaging)
- [Bug] Claude Code refuses to read cloned repository
- Cowork: Chrome "Approved sites" domain approval resets every new conversation despite choosing "Always allow"
- Opus 5 leaves stale references after its own edits, causing repeated review cycles
- [FEATURE] options to "ignore" or white list MCP and other tools, so this session ignores anything not specified as an external needed tool
- Opus 5 (xhigh): confident claims without source verification despite explicit CLAUDE.md rules
- extraKnownMarketplaces (directory source): global registry entry silently overwritten by identically-named copy of the repo (e.g. git submodule)
- Web sessions: git config pre-wires SSH commit signing but the signing key is an empty file — all commits land unsigned/Unverified
- Desktop-to-mobile session sync is stale/missing for some conversations
- A tool listed in permissions.allow is denied by the permission classifier
- [BUG] Claude Desktop (Windows/MSIX): update retries against the deleted %TEMP% .msix instead of registering the already-staged package — app becomes unlaunchable AND unrepairable
- [BUG] Org plugin ("Installed by default") not reaching non-owner Member — Team plan
- [FEATURE] Add SubTasks in TaskCreate
- [Bug] Claude Code not functioning as intended PIECE OF SHIT TOOL
- Workflow({name: ...}) fails with false "control characters" validation error; scriptPath/script work fine
- [BUG] Claude Desktop: Skills consistently fail with No such tool available: skill despite plugins being installed and MCP tools working
- [BUG] Desktop app: no way to open an existing folder; terminal-started sessions never appear in Recents
- [Bug] Bash commands executed in accept-edits mode without confirmation
- Feature request: server-injected per-installation secret for key wrapping (software "TPM")
- [Bug] Safety guardrails trigger on benign text input
- [Bug] CLI ignores user instructions and corrupts existing work
- Feedback: agent made shallow/cosmetic edits instead of substantive fixes when asked to improve prose clarity
- [BUG] claude --version hangs at 99% CPU, regression between 2.1.112 and 2.1.113
- [FEATURE] CLI: `code` is the only block-level token given neither a container nor a screen-reader form
- [BUG] Registering a DirectoryAdded hook makes one scaffold Task subagent return no output even though the hook never fires
- [BUG] Background Bash task launched by a subagent is killed when the subagent returns; the promised completion notification can never fire
- [BUG] Claude Desktop for Linux: main process spins at 100% CPU on session start, generates repeated coredumps
- macOS Bash sandbox: profile grows ~4.5KB per deny path (no dedup of ancestor-literal rules), exceeds ARG_MAX and breaks every sandboxed command
- Unexpected model switch withaut reason
- I appreciate you reaching out, but I'm not able to help with this request. I'm Claude, Anthropic's AI assistant. I'm designed to help with legitimate tasks like answering questions, writing code, analysis, and creative projects - but I can't assist with
- Subagents are resumable conversations but cannot be resumed on a different model — model-scoped failures force a full restart
- [BUG] Login authenticates but has_finished_claudeai_onboarding=false walls existing paid Max account behind new-account onboarding (web/desktop/setup-token) — 10th public report of this signature, first with the mechanism captured on the wire
- [BUG] Desktop global search doesn't match renamed/pinned session titles
- Daemon bg job wedged in 'blocked' after safety-classifier API error, then silently resumes 4 days later and runs tools on stale context
- Session cwd silently resets to original launch directory, discarding legitimate navigation (affects PreToolUse hook cwd for all tool types)
- [Bug] /permissions command selection and search functionality unresponsiv
- /code-review ultra (ultrareview) always reports "no commits yet" regardless of actual git state
- [Bug] Version 2.1.220 spuriously drops auth mid-session with stale subscription state
- [BUG]
- [Bug] Fable 5 safeguards blocking legitimate code operations
- [BUG] Share transcript prompt intercepts freeform keypress with no undo — accidental publish with no recovery
- [BUG] Desktop remote sessions: plugin sync to ~/.claude/remote/plugins/ omits hooks/, so plugin hooks never fire
- TOOL CAN EVEN PASTE WHAT CHAY GPT DID IN MINUTES. FULLSCRIPT AND CLONE ITS ALL SHIT- LEGAL CASE
- claude-in-chrome: tabs_context_mcp reports extension not connected despite extension installed, enabled, and logged in
- [Feature Request] Add verbosity/logging options to keep user informed during execution
- [DOCS] Hook reference omits the PowerShell tool's tool_input schema, so a PreToolUse PowerShell hook fails open on a guessed field name
- [FEATURE] Reusable saved prompts for /btw (side chat) — and a /compact that drafts its own instructions there
- Silently dropped required specify-field text while editing a multi-field court form (UCS-FW1), filed with a real defect
- [Feature Request] Effort level setting should not interrupt agent autonomy or require additional input
- Zoom only affects the main window — secondary windows cannot be zoomed at all
- Chrome Extension Won't Connect
- [BUG]
- [FEATURE] Let Pro/Max subscription usage count toward API key requests, so subscribers can use their plan in third-party tools without needing separate token billing
- MCP tool call issued during session re-initialisation is silently discarded
- [BUG] Claude Desktop's downloaded Code CLI binary (claude-code-vm/claude-code) is unsigned, causing exit 127 "Malformed Mach-o file"
- [BUG] Connection closed mid-response at ~1% of requests — quantified over 130k local transcripts (Opus 5 / 2.1.220)
- [Bug] Opus 5 model usage not displayed in CLI usage per model section
- Customize/disable the turn-duration verb, or derive it from spinnerVerbs
- TUI leaves one terminal row unused at the bottom when running under tmux
- Ok [Feature Request] Whitelist or exempt epidemiological modeling code from safety filtering
- [BUG] JetBrains plugin: "Send to Claude Code" (Ctrl+Alt+K) does nothing when the Markdown preview pane is focused
- Agent view row shows the parent/default model, not the per-spawn `model` override — same root cause you already fixed for `/agents` in #20291 (v2.1.199)
- [Bug] Microphone input detection fails after recent changes
- [BUG] Windows: Remote Control session archives when the terminal closes; desktop Unarchive fails with HTTP 409 ("Failed to restore session")
- [Bug] Claude model not following user directives and ignoring explicit requests
- [Bug] Ultra review consumed free run quota without completing analysis
- [Bug] Claude Fable 5 blocks legitimate first-party security review with active CVP approval
- Ultrareview consumed a free run but produced zero output (orchestrator crashed)
- [BUG] design-sync truncates prop JSDoc mid-word at 120 chars with no marker (lib/dts.mjs:434)
- [Bug] /clear command resets terminal box color, breaking project instance identification
- [Feature Request] Add support for Claude 3.5 Opus model variant
- Crypto scan 2026-08-03 09:00 — 0 buy candidates, 0 exit flags
- False-positive 'Output blocked by content filtering policy' during benign LaTeX/Greek transcription task
- Custom statusLine never renders (built-in footer badge row works) — 2.1.220, macOS 15.6, native
- misaligned actions and power persistence on Claude
- [BUG] Mobile attach to ONE specific local session returns 401 'OAuth access token expired' after account token rotation; bridge binding is server-authoritative and survives every local remedy
- Stats "Peak hour" ignores the 7d/30d/All range selector — hourCounts is a flat dateless histogram, and it goes stale
- [BUG] MCP HTTP OAuth flow ignores custom "scope" from .mcp.json, sends hardcoded defaults instead
- claude-in-chrome: extension disconnects mid-session (after file download) and never reconnects
Resolved issues (117)
- Provide a setting to disable the claude.ai/code session link in PR bodies and commit trailers
- [BUG] Missing HCS services: vfpext — confirmed missing even in official Windows 11 ISO (install.esd), after DISM repair and full Hyper-V reinstall
- Desktop app sessions keep loading old plugin version after 'claude plugin update'
- Bash tool_result replaced by a concurrent session's assistant output (cross-session output bleed) — 2 documented occurrences
- [Bug] Model downgrade on cryptography-related code requests without user consent
- (withdrawn by author)
- [FEATURE]
- [Bug] Agent applies workarounds without searching codebase for established patterns first
- Exploded window makes app almost unuseable
- [Feature] /context: collapse MCP tools to one line per server (opt-in --verbose for full list)
- [Bug] Opus 4.8 confabulates tool/agent state and ignores in-context corrections in long sessions
- Agent intermittently emits malformed tool-call delimiters (literal "course"/"invoke") in long sessions
- Onboarding tour for model selection, effort tiers, and token economics on Pro plan
- [MODEL]dyld: Symbol not found on macOS 13.x"
- [BUG]/plugin Disable/Enable toggle targets wrong settings key when plugin manifest name differs from marketplace directory name
- Routines list shows the normalized task id, not the routine's label/description
- Deferred-tool call emitted in the same turn as its `ToolSearch` load permanently bricks the session via thinking-block 400 (extended thinking)
- [Bug] Multiple shell snapshot processes consuming 100% CPU causing battery drain
- [FEATURE] [BUG] New working folder in a new session in Claude Code on the desktop app
- Voice dictation: support multiple languages or separate language setting for speech-to-text
- [Bug] Model silently switches from 1M to non-1M Opus mid-session, causing unrecoverable API Error
- [BUG] Bedrock Guardrails — session permanently blocked after guardrail trigger in Cowork 3P (Desktop)
- [Bug] Anthropic API Error: Output blocked by content filtering policy
- Cowork scheduled tasks silently fail `git push` — headless runtime inherits dangling session sockets, breaking all credential channels
- Subagents should inherit their worktree's cwd + permission scope across tool calls (instead of resetting to the main repo dir)
- [Bug] Anthropic API Error: 529 Overloaded
- [Bug] Security documentation missing from prompt guidelines
- [FEATURE] Native (non-terminal) agent UI for Claude Code in Android Studio
- [Feature Request] Add progress persistence and recovery for interrupted API sessions
- [Bug] Claude not loading CLAUDE.md configuration file
- Extension 2.1.185 crash-loops VS Code renderer (code 5) on macOS arm64 — regression from 2.1.183
- [BUG] Scheduled tasks run inside the user's working tree and leave orphaned git
- [BUG] Cowork/Code sessions hang on load after update to desktop 1.14271.0 (macOS 26.5.1)
- Agent executes a gated workflow out of order on task transition: enforcement state left pointing at the completed prior task goes inert, and the agent free-styles step order
- [Bug] Duplicate text + tool-call blocks after tool error on Linux TUI (v2.1.185)
- [BUG] /login URL hyperlink truncated when URL spans multiple lines
- [BUG] Model wastes excessive tokens in internal reasoning for simple file creation tasks
- [FEATURE] Expose Claude Max plan usage quota in statusline JSON
- Token consumption during offline period — server-side session not terminated after user exit
- [Bug] AskUserQuestion dialog persists after user response; session hangs with incorrect subagent attribution
- [Bug] Text block silently dropped from render and transcript after no-comment AskUserQuestion answer
- OTEL per-request events not emitted on Microsoft Foundry (CLI, v2.1.126)
- [Bug] /compact fails with "Usage credits required for 1M context" even after /model set to Sonnet 4.6
- Long statusline is truncated with `…` instead of wrapping to multiple lines (regression)
- [MODEL] Auto-mode permission classifier misses content visibility from prior Write tool calls in the same conversation
- Ctrl+B (background command) does nothing during the "thinking" phase, which looks identical to a running command
- [MODEL] Claude Code safe-mode --print/resume misroutes conceptual prompt as tool-task clarification
- [BUG] Agent view: slash commands treated as prompts and forwarded to subagent
- [BUG] Desktop SSH: ccd-cli 2.1.181 "is not runnable" — versioned dir never created on a shared remote instance with multiple Claude accounts
- spawn_task chips should persist across app restarts
- [FEATURE] automatic compression of session logs (~/.claude/projects/*.jsonl) to save disk space
- [Bug] Anthropic API content filter blocking legitimate HIPAA/CFR legal research
- [Bug] Transient API 529 "Overloaded" halts an autonomous /goal run permanently — no self-resume, requires manual "continue"
- [Bug] Autonomous code generation diverges without external validation/review - unable to capture inputs from claude design well
- [Bug] Anthropic API Error: Server Rate Limiting - Temporary Request Throttling
- [Bug] Anthropic API Error: Usage credits required for 1M context window
- claude-api bundled skill (v2.1.169): aggressive auto-trigger on model name mentions causes double context compaction
- A lone UTF-16 surrogate (from truncated/persisted tool output) in the session transcript bricks the session with a repeating `400 ... no low surrogate in string` on every turn
- [BUG] Regression v2.1.126→v2.1.179: Sessions fail with '400 Input too long' despite being 3.5x smaller than working sessions
- claude-api skill inlines ~500KB (~130K+ tokens) of reference docs into every invocation
- [BUG] Every Opus subagent dies at 0 tokens on xhigh effort — spawn path hard-sets thinking:{type:"disabled"} while still sending effort:xhigh
- [BUG] I can't login to claude code. Don't get verification email.
- I appreciate you sharing this detailed context
- I appreciate you sharing this detailed context
- [Feature Request] Support for self-recursive harness in Claude Code/
- [Feature Request] Add persistent model indicator to graphical panel UI
- [BUG] OAuth login flow gives no feedback when manual code paste is required — "if prompted" misleads users
- Feedback: no shared state across Claude surfaces/sessions; long conversations also degrade
- Agent writes internal task IDs into commits as #N issue refs
- [BUG] VSCode extension v2.1.145 — claude.exe crashes with 0xC0000409 (BEX64) at deterministic offset 0x3bd3179
- [BUG] auto-updater installed it without validation and broke `claude` ; (2.1.176 SIGILLs on launch on RHEL 7 / glibc 2.17)
- [FEATURE] Parse an affirmative in free-text permission replies instead of always treating them as a denial
- Phone/app session that runs locally is silently lost — transcript never persisted to host (consistent, data-loss)
- [Bug] Session unrecoverable after Anthropic API Error 529
- Korean IME (한글) input broken in CLI prompt — characters decompose into individual jamo
- [BUG] VSCode: Claude Code sessions stay unread because they're only marked read when opened in the (decoupled) chat tab
- [FEATURE] Feature Request: Predictive Inline Code Completion in VS Code Extension
- [FEATURE] granular OTEL_LOG_ASSISTANT_RESPONSES flag for opt-in logging of assistant response text
- [Bug] Anthropic API Error: Server rate limiting requests during high load
- [BUG] /model does not display sonnet[1m] or opus[1m] despite both models being accessible via --model
- [Bug] Anthropic API Error: "Server is temporarily limiting requests" persists for 24h on Opus 4.8
- /agents command not available in desktop app
- Feature request: hook event for API transport/server errors (502/5xx/429/529)
- [FEATURE] Right-To-Left support
- [BUG] Creating zip file in windows for linux
- [BUG] - Switching between two accounts (logout + login) in Claude Code Desktop takes several hours to complete.
- --resume should surface worktree sessions launched from the same repo
- [FEATURE] Design Contract — a continuity bridge between Claude Design and Claude Code
- [Feature Request] Add mechanism to ask user about the solution when user describes a Logical Error, instead of implementing the generic solution
- You've hit your session limit · resets 7:30pm (Europe/Berlin)
- [FEATURE] Enable spellcheck in VSCode extension input (blocked on microsoft/vscode#214367)
- `claude --print` fails with 401 when no interactive session is open (regression in 2.1.185)
- Support per-directory account switching
- [BUG]
- GitHub Integration connector connected but tools not injected into Claude Code session
- Claude Code — ~60s "Request timed out" regression in 2.1.183-2.1.185 (client-side; network_down=false, rate_limited=false)
- [Feature Request] Add security audit and code review capabilities for generated code
- [Feature Request] --resume should restore the last active model (including [1m] variants)
- Opus 4.8: model re-injects an unsolicited suggestion and falsely attributes the deliberation to the user
- Allow Workflow agent() subagents to spawn nested subagents (grant the Agent tool)
- [BUG] Claude Code MCP OAuth aborts on server-advertised protected-resource path mismatch — IBKR connector works in Claude Chat, fails in Claude Code
- Workflow tool triggers on any message mentioning "workflow" — trigger is too broad
- [FEATURE] http MCP server: allow OAuth fallback when a configured Authorization header is empty
- [BUG] Server is busy Claude cowork desktop error
- Cohesive multi-surface assistant: shared memory + task handoff across Claude Code & Desktop (+ launchd FDA regression on macOS CloudStorage)
- VSCode extension: clickable file links with spaces in the path do not open from the chat panel
- 🐛 [Bug] Crash on Qwen3.6-27b: "Cannot read properties of undefined (reading 'startsWith')" in content renderer Description
- Sessions started inside a git submodule are not listed by --resume
- [FEATURE] Add focus mode to the Claude Code web interface
- It is not paying attention what is doing when restoring data from local sql server db to azure sql database
- [BUG] Agent View: Name not always updating
- [BUG] Interactive turns silently no-op (0s, 0 tokens, no /v1/messages request) on a Claude Team seat — --print works, no error shown
- [Feature Request] Pin active routine runs to sidebar for visibility and permission requests
- Context menu "Open in" submenu fails to expand and dismisses entire menu (Claude Desktop, Code section)
- [BUG] Session resume pollutes conversation files with duplicate system messages, hiding conversation history
- MCP OAuth: Claude Code re-runs DCR on every authenticate, orphaning the previously-issued client_id and its refresh_token
- [Bug] Claude Opus 4.6 Regression: Model fails to maintain context on minor code modifications