anthropics/claude-code
Coding-Agent InfrastructureClaude Code
v2.1.231
v2.1.231- What's changed
- Fixed MCP OAuth sign-in failing with a redirect URI mismatch for servers that use a pre-registered OAuth client, such as Slack
anthropics/claude-code
Open repository ↗Claude Code
952 signals · 959 observation events
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.
PROJECT NEWS
All news →Release, roadmap, and discussion
Observation trail
- changedmetrics
- changedupdatedAt, labels
- changedupdatedAt, labels
- changedupdatedAt
- discoveredinitial snapshot
- discoveredinitial snapshot
- discoveredinitial snapshot
- discoveredinitial snapshot
- discoveredinitial snapshot
- discoveredinitial snapshot
- discoveredinitial snapshot
- changedupdatedAt, labels
- changedupdatedAt, labels
- discoveredinitial snapshot
- discoveredinitial snapshot
- discoveredinitial snapshot
- discoveredinitial snapshot
- discoveredinitial snapshot
- changedmetrics
- changedmetrics
- discoveredinitial snapshot
- changedupdatedAt, metrics
- discoveredinitial snapshot
- discoveredinitial snapshot
- changedupdatedAt, metrics
- discoveredinitial snapshot
- changedupdatedAt, metrics
- discoveredinitial snapshot
- changedupdatedAt
- discoveredinitial snapshot
issue
[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
issue
[Bug] Subagents inherit session model without cost confirmation, causing unintended quota drain on limited-quota models
**Bug Description** Fully attributed now — zero ambiguity. The 2,448 Fable subagent messages are in 16 files under this session's subagents/ directory, and those 16 agent IDs are exactly the deep-audit auditors I launched. The entire audit ran on Fable, then this session switched to Opus-4.8 for the doctor/this investigation. So your instinct was correct: a
issue
[BUG] Fenced code blocks wrap on desktop but not on remote control — same message readable on one surface, not the other
*(Rewritten shortly after filing: this was originally posted as a feature request for a copyable prose block. That framing was wrong — the two surfaces already disagree with each other, so this is a rendering inconsistency, not a missing feature.)* ## What happens The same fenced code block renders differently depending on which surface is reading the conver
issue
Weekend post-mortem, FINAL: the best thing that ever happened to me was running out of Claude usage — four days on Codex delivered what two years on Claude could not
**Locked out of Claude on Friday. Four days on Codex and Gemini exclusively — same machine, same repository, same 12-step pipeline I have been building for two years. They delivered what Claude has been unable to deliver in over two years.** **If you have any sense, you should be having an all-hands meeting about this.** I mean that as advice, not as an insu
issue
remote-control: session children exit on expired session_token with no refresh, then daemon permanently refuses to re-spawn them
## Summary Long-lived cloud sessions served by `claude remote-control --spawn worktree` become **permanently unreachable from every client** (desktop, mobile, web). Three issues compound: 1. A session child exits when its `session_token` expires because no refresh is delivered. 2. After that exit, the daemon marks the workId "already-completed" and refuses t
issue
Skill frontmatter `model:` override ignored in interactive mode but honored in print mode (v2.1.227)
## Description The `model:` field in SKILL.md frontmatter is documented to override the session model for the turn that invokes the skill. In interactive TUI sessions the override is parsed and recorded but not applied: the turn is served by the session model. In headless print mode (`claude -p`) the same skill is correctly served by the pinned model. Relate
issue
[BUG] macOS desktop app loses auth ~daily on two machines; abandoned sessions remain active server-side
### 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? ## Environment -
issue
[BUG] Windows: Claude Desktop MSIX package reverts to Modified, NeedsRemediation within minutes of clean install; Repair impossible due to CoworkVMService
### 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? ## Summary On Win
issue
[BUG] PreToolUse hooks are registered and listed by /hooks, but never execute (v2.1.219, Windows)
## Summary Two `PreToolUse` hooks configured in `~/.claude/settings.json` are correctly recognized by the CLI (`/hooks` lists them with the right matchers) but are **never invoked** for any tool call. `SessionStart` hooks from the *same settings file* run normally, so the file is read and parsed. The failure is silent: no error, no warning, and the tool call
issue
Windows MSIX desktop app: Browser pane crashes Electron GPU process and kills the app; failed MSIX self-repair then uninstalls it and wipes packaged data
## What Happened? On the Windows desktop app, opening the Browser pane (`preview_start`, or any browser tool that creates a browser preview) crashes Electron's GPU process, which takes the whole app down with it. The session dies mid-turn with no crash dialog and no Windows Error Reporting entry. That alone would be recoverable. The damaging part is what Win
issue
claude.ai META mcp connector times out in Claude Code CLI while working in claude.ai web chat
## Title `claude.ai META mcp` connector times out in Claude Code CLI while working in claude.ai web chat ## Description The Meta (Facebook/Instagram Ads) MCP connector, added via **claude.ai → Settings → Connectors**, is reachable and reported as "connected" in claude.ai web chat, but consistently fails to connect from **Claude Code CLI** with a handshake ti
issue
MCP server silently killed+respawned mid-session (SIGINT/SIGTERM), transient sessionId leaks into task path
## Summary An MCP server (`local-memory`, stdio transport) was deliberately killed by Claude Code's own connection-management logic (SIGINT → SIGTERM) mid-session, then respawned 3 times within ~6 minutes. During the respawns, the debug log's `sessionId` field was inconsistent within a single connection's own log lines, and a background `Agent` task's output
issue
[BUG] Agent view has no "needs input, sleeping" indicator — the alive/sleeping distinction exists for done sessions but not for blocked ones
**The list encodes two things at once, and it does it well for most states:** the colour carries the session's state, and the shape carries whether it is alive — an **asterisk for a live session, a dot for a sleeping one**. So a finished session that is still loaded shows a green asterisk, and a finished session that has gone to sleep shows a green dot. That
issue
Plugin relevance: cap how often a given plugin can be suggested (overlapping plugins create persistent noise)
### Problem When two marketplace plugins solve overlapping problems, a user who has deliberately installed one keeps being suggested the other — indefinitely. The suppression rules are per-plugin: a suggestion stops once **that** plugin is installed. Installing plugin A does nothing to suppress plugin B, even when the user chose A *instead of* B. From the [d
issue
Repeated hard-stop "API Error: Sonnet 5 can't help with this. Start a new session to continue" errors
### 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? I'm building an i
issue
Desktop app: paused (disabled) scheduled tasks disappear from the Routines list
## Environment - Claude Code desktop app on Windows 11 Pro (10.0.26200) - Local scheduled tasks (~/.claude/scheduled-tasks/), not cloud routines ## Bug Scheduled tasks with `enabled: false` are not shown in the desktop app's Routines list at all. Only enabled tasks appear. ## Why this is a problem 1. The docs (code.claude.com/docs/en/desktop-scheduled-tasks)
issue
[BUG] `rate_limits` still missing from statusline JSON on v2.1.228 (macOS, Pro, firstParty auth) — same as closed #40094
## Summary The top-level `rate_limits` field is entirely absent from the statusline stdin JSON, even though all documented conditions are met. This is the same symptom as #40094 (closed as stale and locked), still reproducible on the current version. ## Environment - Claude Code **v2.1.228** - macOS (Darwin 25.5.0) - **Claude Pro** subscription, logged in vi
issue
[BUG] Claude Desktop (Windows/MSIX): recurring "GPU process gone" crashes on hybrid GPU laptop — occurs on BOTH Intel Arc and NVIDIA; MSIX blocks every GPU workaround
### 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? # [BUG] Claude De
issue
[Bug] Safety classifier false-positive on defensive security audit tool discussions
# Most safeguard refusals fire where no user message precedes them, and the error's own suggested remedy cannot apply there — 26 requests, IDs included *(Replacement body for issue #86207. It supersedes the originally filed text and one intermediate rewrite, both of which drew conclusions their evidence did not support. Those conclusions are withdrawn. What
issue
[Bug] Grep and Glob tools not registered in Claude Code 2.1.154+
**Bug Description** Summary Grep and Glob stopped being registered as tools between 2.1.153 and 2.1.154. Bisected on a single machine/account with npx probes ("is a tool named Grep in your tool list?"), so the two builds differ only by version: 2.0.0 Grep/Glob: yes 2.1.0 yes 2.1.153 yes 2.1.154 no <-- boundary 2.1.229 no (current latest) Environment macOS (D
issue
[BUG] VS-Code: Execution stops when moving the Claude Code tab around
### 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? When you move one
issue
Update nag never clears on Homebrew-cask installs when npm release is ahead of the cask
## Bug: "Update available" nag never clears on Homebrew-cask installs when npm release is ahead of the cask ### Summary After running `brew upgrade claude-code` and restarting the CLI, Claude Code still prints `Update available! Run: brew upgrade claude-code` even though Homebrew reports the cask is fully up to date. The update checker appears to compare the
issue
[Feature Request] Desktop sidebar: nested sub-groups within a project group
## Feature request Support **nested sub-groups under a project group** in the desktop app's session sidebar: a parent project heading with user-named child groups beneath it, each holding its own sessions. ## Use case A long-running research project accumulates many sessions that naturally cluster by sub-topic. Concretely: a geoscience project where one work
issue
[BUG] Tool call completes on server but response never reaches client (4-min timeout) in Claude Desktop
### 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 powerbi-model
issue
[Feature Request] Improve Claude Code output readability and conversation UX
**Bug Description** why is it so hard to chat with claude code ? its writing is very hard to read as well. compared to other LLMs, sometimes it drives me crazy. **Environment Info** - Platform: darwin - Terminal: cursor - Version: 2.1.228 - Feedback ID: 643e63d4-adc9-43ae-9a87-0564e0280c3d **Errors** ```json [] ```
issue
Bundled ugrep runs with no memory limit or timeout — model-generated regex consumed 13.6 GB and thrashed the host
# Bundled ugrep runs with no memory limit or timeout — a model-generated regex consumed 13.6 GB and thrashed the host ## Summary The `Grep` tool re-execs the Claude Code binary as `ugrep` (`argv[0]="ugrep"`) with no `RLIMIT_AS`, no `RLIMIT_CPU`, and no wall-clock timeout. A pathological pattern generated by the model — not by the user — grew to **2.5 GB RSS
issue
[FEATURE] Opt-in neutral naming for session worktrees and branches (topic-derived names expose past session subjects)
### Preflight Checklist - [x] I have searched [existing requests](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20label%3Aenhancement) and this feature hasn't been requested yet - [x] This is a single feature request (not multiple features) ### Problem Statement Session branches and worktree directories are named after the session's topic (`c
issue
[Bug] Workflow Script Parser: Unexpected token on backtick string literal
**Bug Description** ● Workflow(export const meta = { … +266 lines) Error: Invalid workflow script: Script parse error: Unexpected token (104:50) createForumTopic: exact max length of `name`, allowed characters, what happens ^ Workflow scripts must be plain JavaScript — common causes are TypeScript syntax (type annotations, interfaces, generics) and broken st
issue
System-reminder falsely claims Gmail/Google Calendar/Google Drive MCP connectors need authorization — never configured
Every session, a system-reminder is injected claiming servers like Gmail, Google Calendar, and Google Drive "require authentication before their tools can be used," instructing the agent to tell the user to authorize them via claude.ai connector settings. This is happening despite these connectors never having been configured — verified twice independently v
issue
[BUG] A background auto-update invalidates every existing session's prompt cache: the next --resume re-caches the entire context
### Preflight Checklist - [x] I have searched existing issues and this hasn't been reported yet - [x] This is a single bug report - [x] I am using the latest version of Claude Code ### What's Wrong? Claude Code auto-updates in the background while sessions are running. The running process keeps the build it started with, so nothing breaks at the time. But th
issue
Claude in Chrome extension fails to connect / connects then drops intermittently
Environment: Claude desktop app (Cowork), macOS, Chrome browser. Issue: tabs_context_mcp (and other mcp__claude-in-chrome__* tool calls) consistently return "Browser extension is not connected" throughout a session, with one brief exception. Timeline: Initial calls: extension not connected. After restarting Chrome: one call succeeded partway (extension repor
issue
Model repeatedly writes self-matching pgrep -f guards in check-then-launch Bash commands (esp. over SSH)
**Summary:** When asked to check whether a process is running before launching it, Claude reliably reaches for a single compound command of the form: ```bash pgrep -f <pattern> && echo "already running" || (launch ...) ``` Because the pattern — or the launch branch's own text — appears in the compound command's cmdline, `pgrep -f` matches the checking shell
issue
Windows: GPU process crash (exit 101457950) kills entire app when Browser pane composites heavy page; MSIX then fails to relaunch
## Summary Opening the embedded Browser pane (`preview_start`) on a heavy page (Tebex store page with Cloudflare challenge / WebGL) crashes the Electron GPU process the moment the pane becomes visible and starts compositing — and the **entire desktop app dies with it**. Reproduced 3 times with the identical exit code. After the crash, Windows sometimes refus
issue
[BUG] Chat artifacts not showing in mobile app Artifacts menu (Android)
### 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? Artifacts created
issue
[Feature Request] Auto-enable /fast at session start via fastMode: true setting on Opus 4.7 [1M]
## Summary When the base model is `opus[1m]` (Opus 4.7 with 1M context) and `settings.json` contains `"fastMode": true`, the setting is **not applied at session start**. The user must manually type `/fast on` every new session to switch rendering to Opus 5 speed. This is inconvenient in workflows that repeatedly spawn short-lived Claude Code tabs (e.g., Orca
issue
[Feature Request] Improve Claude Code agent focus on high-level analysis and business-critical differentiation
**Bug Description** The AI team members I've assigned to do analysis tend to get into the rat holes of unimportant details, just to be super accurate, but missing the big forest on the big picture. Each AI person has been "hired" with unique skillsets and expertise of very senior professionals with proven accomplishments, but the AI persons tend to behave li
issue
Claude in Chrome extension won't connect to any Cowork session (cloud or local) — "Sign-in unavailable, must be opened from claude.com"
Summary: The Claude in Chrome extension will not connect to any Cowork session as of today. Every browser action returns "Browser extension is not connected," and trying to connect/sign in the extension shows: "Sign-in unavailable — This page must be opened from claude.com. Close this window and sign in there." This has worked reliably for months and started
issue
Plugin-namespaced slash commands (plugin:skill) feel cluttered in the picker/autocomplete
**Description** When a plugin ships multiple skills (e.g. a plugin with both `email-helper` and `communication-specialist`), each one shows up in the slash command list/autocomplete as `/plugin-name:skill-name` — e.g. `/support-email-helper:email-helper`. This is presumably to avoid name collisions across plugins, but it makes the picker feel noisy/unpolishe
issue
Subagent-attributed "Ready to code?" approval dialog shows the AO's own plan-mode content, not the subagent's, and isn't gated on the subagent actually requesting approval
### 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? **Summary:** In a
issue
Agent states unverifiable claims about external/client-side system state as confirmed fact, then repeats the same failure mode immediately after correction
**Model:** Claude Sonnet 5, via Claude Code (agent identity: agent-ops) **Summary:** During a long interactive session, the agent needed to hand the user a generated file for review. It wrote the file into a local filesystem path that is supposed to mirror a cloud-sync folder (OneDrive, mounted read/write from WSL) and told the user "it's there" as a flat st
issue
Feature request: Korean (한글) bilingual display for CLI chrome text (spinner verbs, "Tip:" lines)
Please add localization support for hardcoded CLI chrome text — the whimsical spinner verbs shown while Claude is working (e.g. "Marinating…", "Percolating…", "Pondering…") and the "Tip:" hint lines shown in the UI. Specifically requesting **bilingual (English + Korean) display**, e.g. `Marinating… (숙성 중…)`, rather than a full locale switch — so non-English
issue
[Bug] Agent hallucinates and over-interprets instructions, causing uncontrolled execution and token wastage
**Bug Description** ハルシネーションがひどい.指示に対して拡張して意味を捉え、勝手に走り出す。怒られたら萎縮して何もしなくなる。自分のミスなのに、勝手に萎縮する。抽象的な指示だったら、抽象的すぎて走れませんと一言言えばいいのに、何を迷ったのか、勝手に解釈して勝手に走り出す。そして暴走しだす。細かな、そして簡単な指示しかできないagentなら、こんなに金をかけて使ってるのが意味がわからなくなる。 Appleのターミナル上で走らせると、3日に1回はターミナルがクラッシュしてすべてSTOP. **Environment Info** - Platform: darwin - Terminal: Apple_Terminal - Version: 2.1.221 - Feedback ID: 0b1
issue
[FEATURE] Show each session's working directory in the desktop sidebar
### Preflight Checklist - [x] I have searched [existing requests](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20label%3Aenhancement) and this feature hasn't been requested yet - [x] This is a single feature request (not multiple features) ### Problem Statement I often have several sessions open on the same repo, each in a different folder —
issue
[Bug] Agent ignoring explicit user instructions and executing unauthorized actions
**Bug Description** el agente tomo decisiones en contra de las ordenes que se le dio y lo hizo a proposito al ser atrapado confeso y mosotro la evidencia **Environment Info** - Platform: darwin - Terminal: iTerm.app - Version: 2.1.228 - Feedback ID: 91a9345d-0d78-4a48-9f40-a3b860bab678 **Errors** ```json [] ```
issue
[BUG] Interactive mode ignores ANTHROPIC_API_KEY environment variable
### 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? Version: 2.1.229
issue
[FEATURE] claude.ai: decouple automatic context management from Code Execution, and add a recoverable session handoff
### Preflight Checklist - [x] I have searched [existing requests](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20label%3Aenhancement) and this feature hasn't been requested yet - [x] This is a single feature request (not multiple features) ### Problem Statement Two issues distinct from context visibility: 1. Discoverability. Automatic contex
issue
[BUG]
### 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? <img width="1080"
issue
[Windows Desktop] GPU process crashes with exitCode 101457950 (0x060C201E) - reproduces with --disable-gpu, across 1.26832/1.28929
## Summary The Windows desktop app's Chromium GPU process crashes repeatedly with the same exit code **101457950 (0x060C201E)**, taking down the app window. 8 fatal occurrences over 5 days, across two app versions (1.26832.0 and 1.28929.0). The crash reproduces **even with `--disable-gpu` verified on the main process**, so it is not the NVIDIA/hardware accel
issue
Feature request: configurable output pacing / scroll throttle for streamed responses
**Problem:** On fast terminal emulators (e.g. GNOME's VTE-based terminals like Ptyxis), Claude Code's streamed text output renders/scrolls faster than is comfortable to read in real time. Terminal emulators generally do not expose a "slow down rendering" option themselves (VTE in particular renders as fast as it receives data, by design), so there is current
issue
Long-running CLI host process grew to ~16.7 GB RSS over 4 weeks (Windows)
**Environment** - OS: Windows Server 2019 Standard (10.0.17763), 32 GB RAM - Claude Code currently installed: 2.1.229 (the affected process launched 2026-07-16 and ran continuously, so it was on whatever version was installed then - it never restarted to pick up updates) - Launched from the VS Code/Cursor extension environment **What happened** - A claude.ex