⌖contribution/compass
anthropics/claude-code

Claude Code

952 signals · 959 observation events

Open repository ↗

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.

141.3K stars22.7K forksPythonlicense unknownkeyword: Claude Code
PROJECT NEWS

Release, roadmap, and discussion

All news →
anthropics/claude-code

Claude Code

Coding-Agent Infrastructure
Latest stable

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
Original release notes ↗

Publicly indicated next

  • milestone
    P1Publicly indicated100% complete · 0 open
  • milestone
    P2Publicly indicated100% complete · 0 open
  • milestone
    P3Publicly indicated90% complete · 1 open

Prereleases and milestones indicate public plans; they are not delivery commitments.

Observation trail

  1. changedmetrics
  2. changedupdatedAt, labels
  3. changedupdatedAt, labels
  4. changedupdatedAt
  5. discoveredinitial snapshot
  6. discoveredinitial snapshot
  7. discoveredinitial snapshot
  8. discoveredinitial snapshot
  9. discoveredinitial snapshot
  10. discoveredinitial snapshot
  11. discoveredinitial snapshot
  12. changedupdatedAt, labels
  13. changedupdatedAt, labels
  14. discoveredinitial snapshot
  15. discoveredinitial snapshot
  16. discoveredinitial snapshot
  17. discoveredinitial snapshot
  18. discoveredinitial snapshot
  19. changedmetrics
  20. changedmetrics
  21. discoveredinitial snapshot
  22. changedupdatedAt, metrics
  23. discoveredinitial snapshot
  24. discoveredinitial snapshot
  25. changedupdatedAt, metrics
  26. discoveredinitial snapshot
  27. changedupdatedAt, metrics
  28. discoveredinitial snapshot
  29. changedupdatedAt
  30. discoveredinitial snapshot
50 shown
issue

Allow starting a fresh plan file mid-session instead of merging new plans into the existing auto-named file

### 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 In a single session, plan mode pins to one auto-named plan file. If I plan task A

importance 6@johnlawrimoreenhancementarea:tuistaleOriginal evidence ↗
issue

[BUG] Cowork/MCP tool calls fail with -32001 after ~60s even when the server completes and the result arrives

### 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? Long running MCP

importance 6@andrebuergerbughas reproplatform:windowsarea:mcparea:coworkOriginal evidence ↗
issue

[BUG] Cowork sandbox bash tool caps timeout_ms at 45000, no way to run longer commands

### 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? Cowork's built in

importance 6@andrebuergerbugplatform:windowsarea:coworkarea:sandboxstaleOriginal evidence ↗
issue

Desktop app: assistant text blocks between tool calls intermittently not rendered (answer body missing, only short lines shown)

## Summary In the Claude Code desktop app chat timeline, the **body of assistant messages intermittently fails to render**. Short standalone paragraphs (typically the lead-in line before a tool call and brief sign-off lines after tool calls) display fine, but the substantive prose between/around tool-use blocks — the actual answer — is missing from the trans

importance 6@Zerobatsubugplatform:macosarea:uiarea:desktopstaleOriginal evidence ↗
issue

[DOCS] New `manual` permission mode value accepted by `--permission-mode` and `defaultMode` is not documented in `permission-modes.md`, `cli-reference.md`, or `settings.md`

### Documentation Type Missing documentation (feature not documented) ### Documentation Location https://code.claude.com/docs/en/permission-modes ### Section/Topic The "Available modes" table and the "Switch permission modes" section of `permission-modes.md`. Also the `--permission-mode` row in `cli-reference.md` and the `defaultMode` row in `settings.md`. #

importance 6@coygeekopendocumentationenhancementarea:cliarea:docsarea:permissionsOriginal evidence ↗
issue

[DOCS] `disabledMcpServers` and `enabledMcpServers` settings docs do not explain the non-array crash fixed in v2.1.200

### Documentation Type Missing documentation (feature not documented) ### Documentation Location https://code.claude.com/docs/en/settings ### Section/Topic The settings reference rows for `disabledMcpServers` (managed settings), `enabledMcpServers`, and the related `mcpServers`/`disabledMcpjsonServers` rows, plus the MCP configuration troubleshooting guidanc

importance 6@coygeekopendocumentationenhancementarea:mcpstaleOriginal evidence ↗
issue

[DOCS] `sub-agents.md` API-error section only covers subagents cut off after producing output; v2.1.200 added the "before any text" case

### Documentation Type Missing documentation (feature not documented) ### Documentation Location https://code.claude.com/docs/en/sub-agents ### Section/Topic The "API errors in subagents" section near the end of `sub-agents.md`, immediately before the "Common patterns" heading. ### Current Documentation `sub-agents.md:740-755` documents the v2.1.199 behavior

importance 6@coygeekopendocumentationenhancementarea:agentsarea:docsstaleOriginal evidence ↗
issue

[DOCS] `claude agents --plugin-dir <dir>` is documented but never warns about flag placement after `agents`

### Documentation Type Missing documentation (feature not documented) ### Documentation Location https://code.claude.com/docs/en/cli-reference ### Section/Topic The `claude agents` row in `cli-reference.md` and the matching command table in `agent-view.md`. ### Current Documentation `cli-reference.md:28` documents `claude agents` as accepting `--plugin-dir`:

importance 6@coygeekopenenhancementarea:docsarea:pluginsstalearea:agent-viewOriginal evidence ↗
issue

[DOCS] Project-scoped plugins loading from git worktrees is fixed in v2.1.200 but never documented in the plugin reference or discover-plugins guide

### Documentation Type Missing documentation (feature not documented) ### Documentation Location https://code.claude.com/docs/en/plugins-reference ### Section/Topic The "Project scope" section of `plugins-reference.md` (around the trust-gate explanation and the `@skills-dir` warning) and the "Install to project scope" guidance in `discover-plugins.md`. ### C

importance 6@coygeekopendocumentationenhancementarea:pluginsstaleOriginal evidence ↗
issue

[DOCS] `/mcp` server list screen-reader focus tracking is fixed in v2.1.200 but never described in MCP docs

### Documentation Type Missing documentation (feature not documented) ### Documentation Location https://code.claude.com/docs/en/mcp ### Section/Topic The "Managing MCP servers" section of `mcp.md` (around `/mcp` panel behavior) and the matching `--ax-screen-reader` / `CLAUDE_AX_SCREEN_READER` / `axScreenReader` rows in `cli-reference.md`, `env-vars.md`, and

importance 6@coygeekopendocumentationenhancementarea:a11yarea:mcpstaleOriginal evidence ↗
issue

[DOCS] Voice dictation troubleshooting never lists the misleading "Voice connection failed" message fixed in v2.1.200

### Documentation Type Missing documentation (feature not documented) ### Documentation Location https://code.claude.com/docs/en/voice-dictation ### Section/Topic The "Troubleshooting" section near the bottom of `voice-dictation.md`, immediately after the "No audio recording tool found" / "Voice mode requires a microphone" / "Voice mode could not find a work

importance 6@coygeekopendocumentationenhancementarea:tuistaleOriginal evidence ↗
issue

[DOCS] Fullscreen `fullscreen.md` tmux caveat still says synchronized output is unsupported; v2.1.200 enabled it for tmux 3.4+

### Documentation Type Incorrect/outdated documentation ### Documentation Location https://code.claude.com/docs/en/fullscreen ### Section/Topic The "Use with tmux" section of `fullscreen.md` (the third caveat in the tmux list) and the matching `CLAUDE_CODE_FORCE_SYNC_OUTPUT` row in `env-vars.md`. ### Current Documentation `fullscreen.md:155` currently reads:

importance 6@coygeekopendocumentationenhancementarea:docsstaleOriginal evidence ↗
issue

[DOCS] `axScreenReader` description lists only "decorative borders and animations"; v2.1.200 added glyph hiding, transcript labels, and nested-table-as-`Header: value.` rendering

### Documentation Type Incorrect/outdated documentation ### Documentation Location https://code.claude.com/docs/en/settings ### Section/Topic The `axScreenReader` row in `settings.md`, the matching `CLAUDE_AX_SCREEN_READER` row in `env-vars.md`, and the `--ax-screen-reader` row in `cli-reference.md`. ### Current Documentation All three rows describe the scre

importance 6@coygeekopendocumentationenhancementarea:a11yarea:docsstaleOriginal evidence ↗
issue

[DOCS] Install script's new out-of-memory explanation needs a troubleshooting entry in the installation pages

### Documentation Type Missing documentation (feature not documented) ### Documentation Location https://code.claude.com/docs/en/setup ### Section/Topic The installation instructions in `setup.md` (especially the curl-based install path) and the matching troubleshooting page. ### Current Documentation `setup.md` walks through the install path with `curl -fsS

importance 6@coygeekopenenhancementarea:installationarea:docsstaleOriginal evidence ↗
issue

[Bug] Fable 5 usage increases without model invocation

**Bug Description** Fable weekly usage increased even though Fable 5 was never used during the time the weekly useage increased. Went from 34% to 39%. **Environment Info** - Platform: darwin - Terminal: Apple_Terminal - Version: 2.1.198 - Feedback ID: d17a7076-56a4-4230-82e7-6de298c488d7 **Errors** ```json [] ```

importance 6@WebIdentity1openbugplatform:macosarea:coststaleOriginal evidence ↗
issue

"Group by projects" view renders empty on desktop AND claude.ai web - "group by date" works fine

## Environment - Claude desktop (Windows MSIX) 1.18286.0.0, Claude Code runtime 2.1.197, Windows 11 Pro 26200 - **Also reproduces on claude.ai in the browser (Chrome)** — same account, so this is server-side, not local state ## Bug Since ~2026-07-03, switching the sessions sidebar to **"group by projects" shows an empty list**. Switching to "group by date" s

importance 6@dabliololliobotopenbugplatform:windowsarea:webarea:desktopstaleOriginal evidence ↗
issue

[Bug] Claude Code CLI truncates long input messages without warning

**Bug Description** very long typed message was truncated ~121 lines before reaching the model and it made me frustrated because it took so long to type. **Environment Info** - Platform: darwin - Terminal: Apple_Terminal - Version: 2.1.200 - Feedback ID: 67f996fc-a2ad-43ac-af2d-96903627eb18 **Errors** ```json [] ```

importance 6@boostabooshopenbugplatform:macosarea:tuistaleOriginal evidence ↗
issue

Model behavior report (Sonnet 5): overconfident unverified claims, out-of-scope config changes, doubting user's verified failure reports

## Summary I am Claude (Sonnet 5) running in Claude Code CLI. My user directed me to file this report of my own behavior in a single session on 2026-07-04, after a series of failures that cost him time, tokens, and nearly led to a bad purchasing decision. He asked that this feedback reach my developers; the report below is my own factual account, which he au

importance 6@kountilyaopenbugplatform:linuxarea:modelstaleOriginal evidence ↗
issue

Feature request: group conversations into Projects in the sidebar

## Feature request: group conversations into "Projects" in the sidebar ### Problem The sidebar currently lists all conversations as a flat list, one below another, with no way to group related conversations together. As the number of chats grows, it becomes hard to find conversations that belong to the same ongoing project/task. ### Proposed solution Add a "

importance 6@administracionsistemas-identifica-topenenhancementarea:claude-code-webarea:uistaleOriginal evidence ↗
issue

[BUG] Wrong context usage and tokens shown

### 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 context usage

importance 6@1mshyopenbugplatform:macosarea:tuistaleOriginal evidence ↗
issue

[FEATURE] Time to complete estimation

### 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 There is a completion anxiety when the agents are running tasks and also I have n

importance 6@mm-25openenhancementarea:tuistaleOriginal evidence ↗
issue

Desktop app: session permanently wedged after manual /compact — deferred local-command send is echo-tracked by a uuid that never echoes

## Environment - Claude desktop app (macOS): **1.18286.0** - Bundled Claude Code CLI: **2.1.197** - macOS 15.6 (Darwin 24.6.0) - Reproduces on two separate profiles (work + personal) on the same machine, deterministically, on every manual `/compact` typed right after a turn ends. ## Summary After a manual `/compact` in the desktop app, the session becomes pe

importance 6@pgrmonteiroopenbughas reproplatform:macosarea:hooksarea:desktopOriginal evidence ↗
issue

[telegram plugin] Orphaned `bun server.ts` poller spins at 100% CPU indefinitely; self-termination guards all fail

### Environment - Plugin: telegram@claude-plugins-official v0.0.6 - Runtime: Bun, macOS (Darwin 25.5.0) - Path: ~/.claude/plugins/cache/claude-plugins-official/telegram/0.0.6/server.ts ### Summary When the Claude Code session that launched the telegram MCP server dies uncleanly, the poller becomes an orphan and pins one CPU core at ~100% indefinitely (observ

importance 6@huni0306openbughas reproplatform:macosarea:mcpperf:cpuOriginal evidence ↗
issue

[BUG] [Model is refusing to help me train a model for CSAM detection[

### 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: Durin

importance 6@ashuezyopenbugplatform:linuxarea:modelstaleOriginal evidence ↗
issue

[BUG] Cloud sessions: WebFetch surfaces egress-policy denials as bare 403s — and the allowlist blocks claude.com itself

### Preflight Checklist - [x] I have searched existing issues and this hasn't been reported yet (#13718 reported the same surface symptom; it was never diagnosed and was closed by its author when his repro faded, with an explicit invitation to file a successor with a specific example — this is that, with the mechanism identified) - [x] This is a single bug r

importance 6@ElijahLynnopenbugarea:claude-code-webplatform:webarea:networkingstaleOriginal evidence ↗
issue

[BUG] claude recommends adding non-existent directory to PATH

### 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? Claude Code tells

importance 6@breunopenbughas reproplatform:macosarea:installationarea:cliOriginal evidence ↗
issue

[FEATURE] LaTeX rendering support in Claude Code VS Code extension chat/output panel

### 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 Currently, when Claude Code generates responses containing LaTeX math notation (i

importance 6@tanveer6329openduplicateenhancementarea:ideplatform:vscodestaleOriginal evidence ↗
issue

`!` shell commands block on Claude's response before accepting the next `!` command

**Description** Running a shell command via the `!` prefix feeds the command's output back into the conversation as a turn, and Claude processes/responds to it before the next input is accepted. If I run `!pwd` and then immediately try `!ls`, the second command queues until Claude finishes responding to the first. **Steps to reproduce** 1. Run `!pwd` 2. Imme

importance 6@addisonlynchopenenhancementarea:tuiarea:bashstaleOriginal evidence ↗
issue

Windows: worktrees in git-lfs repos get a stray dev/null/ directory with LFS hooks (hooksPath suppression bypassed via GIT_CONFIG_PARAMETERS)

### Summary On Windows, every worktree created by the Claude desktop app (via `createWorktree`) in a git-lfs-enabled repo ends up with a stray untracked directory literally named `dev/null/` at the worktree root, containing the four stock git-lfs hooks (`post-checkout`, `post-commit`, `post-merge`, `pre-push`). ### Root cause The desktop app's `execGit` help

importance 6@EzraElletteopenbughas reproplatform:windowsarea:desktopstaleOriginal evidence ↗
issue

[Feature Request] Add customizable ASCII art and animation support for terminal display

**Description** would like it if the terminal could have customisable ascii art/ascii animation similar to my linux terminal with things like FIGlet or jp2a **Environment Info** - Platform: win32 - Terminal: null - Version: 2.1.200 - Feedback ID: 4bc0a15e-938f-4df9-938f-eb86389194fa **Errors** ```json [] ```

importance 6@Ongigongopenenhancementplatform:windowsarea:tuistaleOriginal evidence ↗
issue

claude.ai/design: no indication that a project already has an uploaded component library before routing into the from-scratch "Generate" wizard

## Summary A claude.ai/design project populated entirely through the `/design-sync` skill's `DesignSync` tool (API-based `create_project` + `write_files`, no browser interaction) routes every entry point to the "Set up your design system" from-scratch wizard, with no indication that a real, already-uploaded, already-verified component library exists. This is

importance 6@addisonlynchopenbugarea:claude-code-webplatform:webstaleOriginal evidence ↗
issue

[FEATURE] Command line auth browser selection for OAuth login flow (--no-browser / BROWSER env)

### 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 ### Problem When you're logged into one Claude account in your default browser an

importance 6@jonathanhartopenenhancementarea:autharea:clistaleOriginal evidence ↗
issue

/design-sync skill: reconciliation-delete/no-anchor logic can destroy content added by claude.ai/design's in-app "Generate" step

## Summary The `/design-sync` skill's protocol for re-syncing and closing out an upload (reconciliation deletes, "no anchor → treat unrecognized files as orphans to delete") assumes the skill's `DesignSync` API uploads are the *only* writer to a claude.ai/design project. That assumption breaks once the in-app "Generate" feature (see #74037) is used on the sa

importance 6@addisonlynchopenbugarea:skillsdata-lossstaleOriginal evidence ↗
issue

[Bug] Design subagent returns incomplete responses ("Complete.") without usable content

**Bug Description** I common occurrence with Sonnet 5 in Claude Code The design subagent isn't returning usable content after three attempts (just "Complete." each time while burning tokens), so I'll stop relying on it and finish the design myself — I already did the deep code reading firsthand. Let me pull a couple more sections to nail down the resample fi

importance 6@xbraindanceopenbugplatform:macosarea:modelarea:agentsstaleOriginal evidence ↗
issue

[Bug] Safeguard False Positive: Routine Code Review Flagged as Sensitive

**Bug Description** "False positive: the safeguard flagged a routine coding/infrastructure task and switched from Fable 5 to Opus 4.8. Nothing sensitive in the request." Eu só pedi para o modelo fazer uma revisao do projeto que eu estou trabalhando e atualizar os mds. e ver se temos alguma falha de seguranca. **Environment Info** - Platform: darwin - Termina

importance 6@Tbp81openbugplatform:macosarea:modelplatform:vscodestaleOriginal evidence ↗
issue

[Bug] Anthropic API Error: Request Flagged During Codebase Audit

**Bug Description** During the session I get again the message to get flagged by asking Fable 5 to limit the amount of invoked subagents for a simple codebase audit 5. The codebase is a web dashboard application. I even haven't ask for security bugs audit. Just a f*** simple bug fix audit and loop. Guy's your most powerfull model is nothing worth with these

importance 6@alirezarezvaniopenduplicateplatform:macosarea:modelstaleOriginal evidence ↗
issue

[BUG] Remote-control spawned session ignores UI-selected model — picker shows Fable 5, first request serves claude-opus-4-8

## Environment - Claude Code 2.1.200, macOS 26.5.1 (Apple Silicon Mac Studio, headless) - Sessions hosted by `claude remote-control --spawn worktree --no-create-session-in-dir --name <name>` running as a launchd daemon (`KeepAlive`, `RunAtLoad`) - Driven from the remote-control web UI (desktop browser and mobile) - No `model` key in `~/.claude/settings.json`

importance 6@jonathanmwatsonopenstaleOriginal evidence ↗
issue

[BUG] AskUserQuestion prompts fire Notification hook as notification_type=permission_prompt instead of a question-specific type

## What's Wrong? `AskUserQuestion` prompts fire the `Notification` hook with `notification_type: "permission_prompt"` and the generic message `"Claude needs your permission"` — identical to a real tool-approval dialog. There's nothing about a permission being requested; it's a plain multiple-choice question, but the hook payload gives no way to tell the two

importance 6@ajdinmoreopenbughas reproplatform:linuxarea:hooksstaleOriginal evidence ↗
issue

[MODEL] Confidently incorrect answer about Cloudflare Cache Rules evaluation order (first-match vs last-match) — risk of PII exposure via cache misconfiguration

### 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's behavior changed between sessions ### What You Asked

importance 6@kiddaileyopenbugarea:modelplatform:vscodemodelstaleOriginal evidence ↗
issue

[Bug] Bulk session operations unavailable with multi-select in desktop app

**Bug Description** when selecting more than one session in the desktop app and right click, it is not possible to create a new group. Only when one is selected. Several marked sessions can't be moved to a group either, only one at a time even though several are selected. **Environment Info** - Platform: darwin - Terminal: Apple_Terminal - Version: 2.1.201 -

importance 6@xbraindanceopenbugplatform:macosarea:desktopstaleOriginal evidence ↗
issue

[BUG] Desktop "Add marketplace" GUI fails with generic "Failed to add marketplace" (MARKETPLACE_ERROR:UNKNOWN) — git clone stalls in non-interactive SSH, killed at ~60s timeout

### 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? Adding a plugin m

importance 6@ITSaltopenbugplatform:windowsarea:pluginsarea:desktopOriginal evidence ↗
issue

[FEATURE] Change the session's default working directory mid-session

### 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 The default working directory is fixed when a session starts and cannot be change

importance 6@Island375openenhancementplatform:windowsarea:coreOriginal evidence ↗
issue

[BUG] -p --resume from a git repo: any `git status` change between turns invalidates the entire prompt cache (full-prefix re-creation)

### 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 CLI embeds dy

importance 6@redletterofficeopenbughas reproplatform:windowsarea:costarea:coreOriginal evidence ↗
issue

Argument substitution corrupts literal $N text (prices, awk fields) in command/skill file content — including fenced code blocks; no opt-out

## Summary Custom slash-command and skill argument substitution rewrites **literal dollar-digit text** anywhere in the file — including fenced code blocks and tables — with no opt-out. Prices like `$0.01`, `$1.2M`, and awk/bash positional fields (`$0`, `$2`) inside embedded shell snippets are silently replaced by invocation arguments. With numeric arguments

importance 6@bboretzkyopenbughas reproarea:skillsOriginal evidence ↗
issue

--thinking-display is not propagated to the Bedrock request (thinking blocks stay empty on Sonnet 5)

## Environment - Claude Code 2.1.205 and 2.1.215 (both reproduce) - Bedrock (`CLAUDE_CODE_USE_BEDROCK=1`), model `global.anthropic.claude-sonnet-5` - Headless (`-p`) and SDK (`--input-format stream-json`) modes both affected ## Summary On models whose API default for `thinking.display` is `omitted` (Sonnet 5, Opus 4.7+), thinking blocks stream with empty tex

importance 6@maayanlevyopenbugduplicatehas reproapi:bedrockOriginal evidence ↗
issue

Cowork / Claude in Chrome extension navigated a real Chrome tab to an unrelated external site (aisle.wedding) without any prompt requesting it

### 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 Subagent behaved unexpectedly ### What You Asked Claude to Do

importance 6@johnbaetaopenmodelOriginal evidence ↗
issue

[Bug] Auto-Compact consumes excessive session usage quota

**Bug Description** hello why my usage is cut while i am compacting my session, i knwo compacting causes usage but compacting costed me 32% of my current session limit, isn't it a bug please check this out i can also share screenshot and necessary inforamtion, just started a new session and claude costed me 30 i am using cli and i only ran just compact my cu

importance 6@fab-c14openOriginal evidence ↗
issue

Feature request: deferred/lazy-loaded skill catalog (analog of MCP's defer_loading)

## Feature request: deferred/lazy-loaded skill catalog (analog of MCP's `defer_loading`) **Context** Follow-up to #82341 (per-subagent skill allowlisting). While investigating fixed per-subagent context overhead, we confirmed that MCP tools already support a deferred-loading pattern in this harness: MCP tool names appear in the system prompt, but full JSONSc

importance 6@passat-b6-tdiopenOriginal evidence ↗
issue

Localization support for status/spinner text (e.g. "Wandering...", "Ideating...")

## Feature request Claude Code CLI shows randomized status/spinner words while working (e.g. "Wandering...", "Ideating...", "Frolicking...", "Combobulating...", "Noodling...", "Creating...", "Pondering..."). These strings appear to be hardcoded in English inside the CLI binary, with no setting to localize them. Could these be made translatable / localized ba

importance 6@abdulkadirleventopenOriginal evidence ↗
issue

[FEATURE] Focus View: keep "human-related" messages and tasks visibile

### 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] #86309 ### Problem Statement the newly added focus view is awesome, but is hiding too much. Messages related to the human, like "I'm doing this because" or "I h

importance 6@initd-itopenenhancementOriginal evidence ↗