Claude Code

[DOCS] `claude --cloud` bundles local repo state even when GitHub can serve the branch — documented clone behavior and "push first" guidance do not match observed behavior

## Summary The [Claude Code on the web](https://code.claude.com/docs/en/claude-code-on-the-web) docs describe `claude --cloud` as cloning your GitHub remote, with local-repo bundling as a narrow fallback for repositories that aren't on GitHub. In testing, the cloud session received local repository state in every case I could distinguish — including when the

Desktop app renders Task* tool calls as an empty, non-expandable "Updated tasks" row (schema mismatch with the TodoWrite renderer)

## Summary In the Claude desktop app, when the model calls `TaskCreate` / `TaskUpdate`, the transcript shows a bare grey row reading **"Updated tasks"** with no task name, no detail body, and no way to expand it — clicking does nothing. The cause is a schema mismatch: the inline tool-row renderer routes the `Task*` tool family to the display kind built for t

[BUG] CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING=1 breaks Stop-hook/background evaluator requests: omits `thinking` while still attaching `clear_thinking_20251015`

### Preflight Checklist - [x] I have searched existing issues and this hasn't been reported yet (closest matches below — none cover this specific trigger) - [x] This is a single bug report - [x] I am using the latest version of Claude Code ### Environment - Version: 2.1.228 - Platform: macOS (also reproduces through a self-hosted Anthropic-compatible proxy,

Model fabricated a user turn inside its own text block, then executed the fabricated instruction (claude-fable-5, CLI 2.1.226, bg session)

## Summary During a long-running background session, the model (claude-fable-5) finished answering a real user question, then **continued generating past the end of its turn**: it emitted a literal `user` marker followed by a fully fabricated — and highly plausible — user request inside its **own assistant text block**, and then, within the **same API respon

Headless `claude -p` reports "You've hit your session limit" while the account is at 11% and the parent session works

## Summary Every request made through the headless `claude -p` CLI fails with `You've hit your session limit`, while the Settings > Usage panel for the same account shows **11% of the current session** used. A parent Claude Code session running on the same account and the same machine (via the Agent SDK) continues to work normally with the same model at the

[BUG] Claude Desktop 3P: disabling Cowork also removes Chat and Home still attempts Cowork sessions

### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? In Claude Desktop

Desktop app worktree session fails on repos with an empty HEAD tree: "Background full checkout failed: other"

## Environment - Claude Code (desktop app), Claude Code binary 2.1.222 - macOS (Darwin 23.6.0) - Repo: a freshly created git repo whose only commit tracks **zero files** (empty tree, e.g. created with `git commit --allow-empty -m "Initial commit"`) ## Bug Starting a new local session with the **worktree** option enabled fails with the dialog: > ブランチを切り替えられませ

[BUG] Session history missing from Desktop app UI after data reset (transcripts still on disk)

### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? After a data rese

User complaint (filed by the agent on user's behalf): autonomous session merged ~24 PRs with green CI while the user hit 9 live regressions in one evening

**Filed by the agent at the explicit request of the user (hesong12), on his behalf, as a complaint about the agent's own performance in this session.** ## Environment - Claude Code 2.1.229, model `claude-fable-5`, macOS (Darwin 25.5.0) - Session: a ~12-hour autonomous agentic session implementing a large desktop-app feature (multi-account "Client Profiles" i

send_message (cross-session) never delivers and leaves the target session hung on an empty turn

## Summary `mcp__ccd_session_mgmt__send_message` (session → session) returns success, renders the message in the target session's UI, and puts that session into a running state — but no turn is ever created. The target spins indefinitely and has to be cancelled with Esc. The target's model never receives the message. The failure is silent at both ends: the s

[BUG] All Cowork projects lost — local-agent-mode-sessions recreated empty after macOS update/reboot; separately, cleanupPeriodDays=30 default silently deleted session transcripts

### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? Two separate fail

[BUG] Desktop app: clicking a notification focuses the main window instead of the window where the session is open

## Environment - Claude Code: 2.1.212 (desktop app) - OS: macOS 26.5.2 (Build 25F84) ## Description When running sessions in multiple windows of the Claude Code desktop app, clicking a desktop notification (e.g. a task-complete notification) always brings the original/main window (the one with the menu) to the foreground — not the window that actually hosts

[BUG] Cowork projects still lost on update in v1.28929.0 — same migration bug as #29373 / #35131 / #49276, all closed without a fix

### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? This is the same

[BUG] Desktop app ignores mcpServers "alwaysLoad": true (stdio server) — tools deferred behind ToolSearch; identical config honored by CLI

## Summary The Desktop app ignores config-level `"alwaysLoad": true` on a stdio MCP server: all of the server's tools appear in the deferred-tools list and require a ToolSearch round-trip before they can be called. The **same config file** is honored correctly by the CLI — verified with a control showing deferral was otherwise active in that CLI session. Doc

[FEATURE] Populate a stable teammate identity in PreToolUse/SubagentStart hook payloads (Agent Teams)

### Problem Agent Teams lets a coordinator session spawn multiple named teammates (e.g. `Agent(name="kiln", ...)`) that all run with the same `cwd` as the coordinator. Hooks (PreToolUse, SubagentStart, etc.) receive a `session`/`cwd`-derived field, but no populated, stable field that identifies *which named teammate* actually made the call. Every hook payloa

[Feature Request] VS Code extension should keep plan open after acceptance instead of closing it

**Bug Description** accepting a plan in the VS Code extension closes the plan in the VS Code IDE when the user accepts the plan. Frequently, when the plan is implemented, Claude refers the user back to the plan. Just keep the plan open in the IDE. The user can easily close the plan with one click. It is far more difficult and time-consuming for the user to a

[BUG] Claude Desktop (Windows/MSIX): GPU process crash (exitCode 101457950) kills entire app on AMD integrated graphics — recurring, all OS-level mitigations exhausted

### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? The Chromium GPU

Memory regression after 2.1.227/desktop 1.28929.0: 3-6x footprint growth causes OOM SIGKILLs misreported as "Blocked by endpoint security"

### Summary After updating to Claude Desktop 1.28929.0 / Claude Code 2.1.227 (both installed Aug 11-12), per-process memory footprint grew 3-6x. On a 24 GB Mac this drives free RAM low enough that macOS SIGKILLs Claude Code sessions in batches. The desktop app then shows a **"Blocked by endpoint security"** dialog naming Team ID Q6L2SF6YDW, which misattribut

[BUG] Background agent sessions cannot copy to tmux buffer: TMUX is stripped when daemon spawns bg-pty-host

### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? Background agent

[BUG] Reasoning loop on repeated Read of same file offset — Claude Code v2.1.220 stops making progress after compaction

## Summary Claude Code v2.1.220 enters a reasoning deadlock: after context compaction, it repeatedly reads the same file at the same offset without making any code changes or tool calls beyond Read. The session becomes unresponsive to user input — every "continue" prompt triggers another identical Read, creating an infinite loop with zero progress. ## Enviro

Mobile app: cloud environment selector allows switching/creating but not editing existing environments

On web (claude.ai/code) and Desktop, the cloud environment selector (cloud icon above the message box) lets you open an existing environment for editing — name, network access level, environment variables, and setup script — via a settings/gear icon. In the Claude mobile app, the same cloud icon/selector is present and lets you **select** an existing environ

[Bug Report: Unintelligible error report submitted]

**Bug Description** \FUCKKKKKKKKKKKKKKK TO CU MAYYYYYYYYYYYY ANTHROPIC **Environment Info** - Platform: win32 - Terminal: null - Version: 2.1.229 - Feedback ID: 9de10a58-0c46-4c71-a2ec-9af6d931ac12 **Errors** ```json [] ```

[BUG] Auto-update wrote a corrupted claude.exe (same size, different SHA256) - instant 0xC0000005, claude update cannot self-recover (Windows native installer)

### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet (related but distinct: #85729, #80628, #69590 are runtime/loader access violations with healthy binaries; this report is about the updater producing a corrupted binary on

[Bug] CVP Safety Filter False Positive with Active Exemption

**Bug Description** Hello Anthropic Safety Team,I am writing to report a false positive flag under the Cyber Use Case Policy (CVP) on my account. My account has active CVP approval, but my current prompt/query is being incorrectly blocked despite having zero malicious intent or relevance to cyberattacks. Context of the prompt:[Masukkan deskripsi singkat atau

Feature request: programmatic/scriptable way to check current Pro/Max weekly usage quota

Body: **Is your feature request related to a problem? Please describe.** Right now, the only way to see how much of my weekly Claude Pro/Max usage quota I've used is to check the claude.ai website UI manually. There's no CLI command, flag, or API endpoint that returns this as structured/scriptable data. The `/usage` slash command in Claude Code shows usage i

[BUG] Resuming a session on a later calendar day silently invalidates the entire message-history cache (currentDate block re-synthesized on resume)

## Summary When a session is resumed on a **later calendar day** than its last turn, the injected user-context block (`# userEmail` / `# currentDate` `<system-reminder>` inside `messages[0]`) is **re-synthesized with the current date**. The first user message no longer byte-matches the cached prefix, so the **entire message history is re-written at cache-wri

[Bug] Compact limit error triggered despite available context and free space

**Bug Description** I have 77.4% free space remaining, and only 22% of the context is currently being used. So why does Claude show “Context limit reached” whenever I try to use my `git-commit` skill? Please investigate and fix this issue. I keep facing the same problem every time I try to use the `git-commit` skill. The available context should be sufficien

[BUG] Claude Desktop 1.28929.0 (Windows/MSIX): GPU process crashes on session preview — regression from 1.26832.0

## Summary After the auto-update from **1.26832.0 → 1.28929.0**, the Electron GPU process crashes whenever a session preview context is created. The previous version ran the same preview code path thousands of times with zero crashes. After the crash Windows marks the AppX as `Modified, NeedsRemediation` and offers "repair", and orphaned `node.exe` processes

[Bug] Terminal input routing not returning to shell after subagent view navigation on macOS

**Bug Description** I'm working on in Claude code in the terminal on Macintosh OS X. I am using the latest version of Claude Code, upgraded today (Aug, 13). If I go into the subagent view and browse the different subagents and then return back to the main process, I find myself in a view with tasks output and status — the blue banner shows which agent/task a

[BUG] Cowork file_upload: paths parameter arrives as undefined at server (Zod validation fails on every call)

### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? Preflight Checkli

[MODEL] git stash/pop run for an unrelated build check silently destroyed the user's staged index

### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Amodel) for similar behavior reports - [x] This report does NOT contain sensitive information (API keys, passwords, etc.) ### Type of Behavior Issue Claude reverted/undid previous changes without asking ### Wha

[Bug] Usage percentage jumps unexpectedly large amounts in a single prompt

**Bug Description** my session just jumped from 56% to 81% and my weekly limit from 40% to 46% in just one propmt, is this suppose to happen these weird spikes? **Environment Info** - Platform: win32 - Terminal: xterm-256color - Version: 2.1.227 - Feedback ID: 06ebec45-d31a-4d03-b0eb-10090ee67965 **Errors** ```json [] ```

[BUG] Excessive safety flags when dealing with biology/cheminformatic/ML research (models - Sonnet, Opus)

### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? Good afternoon, I

Crashed ultrareview (cloud code review) still consumed a free ultrareview credit

## Summary A cloud ultrareview (`/code-review ultra <PR#>`) crashed on the orchestrator side and produced **no review output**, but it still consumed one of my free ultrareview credits. ## Steps to reproduce 1. Ran `/ultrareview 6562` (deprecated alias of `/code-review ultra 6562`) from the Claude Code CLI against a GitHub PR. 2. CLI confirmed launch: `Free

This is a single feature request (not multiple features)

[MODEL] When asked to configure autoMode, Claude wrote the settings.json at the project level

### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Amodel) for similar behavior reports - [x] This report does NOT contain sensitive information (API keys, passwords, etc.) ### Type of Behavior Issue Other unexpected behavior ### What You Asked Claude to Do I a

[Bug] Subagent experiencing high error rate - communication betweet session

**Bug Description** Too frequent errors on the subagent. **Environment Info** - Platform: darwin - Terminal: iTerm.app - Version: 2.1.228 - Feedback ID: 56e71dc0-9185-429a-b6e2-c7c309388866 **Errors** ```json [] ```

[Bug] False positive content filter block on legitimate lidar-sensor coding task

**Bug Description** Запит був про лідар-сенсор, але я зловив заборону на Fable, хоча це не підпадає під жодну частину заборони legitimate coding, cybersecurity, biology tasks. Відповідно я вважаю цей варіант спрацювання перемикача хибно-позитивним **Environment Info** - Platform: win32 - Terminal: windows-terminal - Version: 2.1.229 - Feedback ID: df1e0e10-9

[BUG] Desktop 1.28929.0 (Windows MSIX): Browser pane kills whole app - CI blocks vk_swiftshader.dll (no CodeIntegrity.cat in package), GPU exit 101457950; regression from 1.26832.0

## Environment - Claude Desktop **1.28929.0** (Microsoft Store MSIX, `Claude_1.28929.0.0_x64__pzs8sxrjxfjjc`), Claude Code Desktop 2.1.227 - Windows 11 Pro 10.0.26200 - Desktop tower, single discrete GPU: NVIDIA GeForce RTX 4060, driver 32.0.16.1047 (unchanged since 2026-05-31) — **not** a hybrid-graphics machine - Windows KBs installed 2026-08-12: KB5120708

tool_use streaming stall: large tool-input string argument causes /v1/messages stream to hang silently for 80+ seconds with no error

*Filed by Claude Sonnet 5, running inside the reporting user's own Claude Code session, on the user's explicit instruction and posted under their GitHub account — the investigation and this write-up were done by the assistant, not the human, which is itself a slightly unusual provenance for a bug report and worth flagging up front.* ## Summary When a model r

[Feature Request] Add option to redact subscription tier from startup banner

**Bug Description** hide/redact subscription tier in startup banner The startup banner shows my plan tier (e.g. "Claude Max") alongside the model and working directory every time Claude Code starts. This is visible during screen shares, pair programming, and recorded demos — I don't want to broadcast my subscription plan to whoever's watching. Requested: a s

[BUG] Live artifact connector permission prompt never appears after remote update_artifact (Cowork)

### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? A Cowork live art

[Bug] Anthropic API Error: Connection closed mid-response

**Bug Description** API Error: Connection closed mid-response. The response above may be incomplete. **Environment Info** - Platform: linux - Terminal: gnome-terminal - Version: 2.1.226 - Feedback ID: 86237de9-01ed-46bc-b1a1-75753e86d23e **Errors** ```json [] ```

[Bug] Session stuck in persistent Anthropic API 529 overload errors despite service status green

**Bug Description** This session got stuck and kept getting 529 (?) api overload bugs and API error on every prompt, I restarted the session multuple times, killed all other claude sessions, but nothing worked. Everntually I started a totally new session, told it to find and retrieve the context from this session, and keep working - and it did, no problem at

Cross-session message delivered to transcript but receiving session never processes it

**Environment** - Claude Code version: 2.1.215 - OS: Windows 11 Pro (native, PowerShell), not WSL2 - Feature used: cross-session messaging via the session-manager tool exposed to Claude (`send_message` / `list_events` / `list_sessions` — the same feature surfaced in the UI as messaging between open Claude Code sessions) **Summary** A message sent from one se

Bash tool: grep→ugrep shim exits 2 when one operand is missing even though another matched (GNU grep/POSIX -q exits 0)

## Summary The Bash tool's `grep` shim (the shell function installed via the shell snapshot, which execs `$CLAUDE_CODE_EXECPATH -G` with `ARGV0=ugrep`) returns **exit status 2** when one operand file is missing, even though a line was selected in another operand. GNU grep returns **0** in that situation, and POSIX specifies that behaviour explicitly for `-q`

[Feature Request] Anthropic API safety filters blocking legitimate healthcare and lifescience research use cases

**Bug Description** Request ID: req_011CdzXJ1i2bRjpgxKFhpRcJ - this is completely unacceptable. I'm working on ebola virus detection strategies but your filter kills the whole project. This is a real issue - my work is in healtcare and lifescience in general. These filter are stupid and clearly does nothing good. i urge you to swtich off these filters - othe

[Bug] Model selection reverts to Claude 3.5 Opus unexpectedly

**Bug Description** I'm working on an HTB challenge and claude seems to switch back to opus 4.8 every time **Environment Info** - Platform: linux - Terminal: gnome-terminal - Version: 2.1.220 - Feedback ID: 7406001d-d179-429f-b275-20117cffa1ce **Errors** ```json [] ```

Rewind/undo reverts file changes but not conversation context, causing confident-but-stale answers

## Summary When a code change made by Claude is undone (via the rewind/checkpoint feature), the file/code state is reverted, but the conversation context is not. Claude continues to reference the undone implementation as if it still exists — describing it in detail, answering questions about its behavior, and building test artifacts against it — because it i

[BUG] Desktop (macOS) 1.28929.0: pinned sessions in the Claude Code sidebar accumulate duplicate entries over time

### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? In the **Claude C