data.json

{
  "metadata": {
    "total_fetched": 5856,
    "total_issues": 1774,
    "total_clusters": 8,
    "generated_at": "2026-03-07T06:26:32.144888+00:00"
  },
  "clusters": [
    {
      "id": 2,
      "issue_count": 607,
      "stats": {
        "total_reactions": 1194,
        "total_comments": 2073,
        "avg_age_days": 14.9
      },
      "analysis": {
        "cluster_title": "Keyword Group: Session Context Json Behavior Agent",
        "summary": "This semantic cluster was defined by the high density of the following keywords: session, context, json, behavior, agent.",
        "hardest_problems": [
          "[BUG] /resume is broken \u2014 session history inaccessible since v2.1.31 (3 root causes identified)",
          "[BUG] Cowork UI: Responses not rendering until user types; \"working through complex response\" hangs indefinitely",
          "[BUG] [1M] Sonnet 1M context window disappeared after Claude Code update to v2.1.45 \u2014 regression on Max plan ($200/mo)"
        ],
        "hallucination_check": "Local heuristic analysis based on TF-IDF term frequencies and engagement metrics."
      },
      "top_issues": [
        {
          "number": 26123,
          "title": "[BUG] /resume is broken \u2014 session history inaccessible since v2.1.31 (3 root causes identified)",
          "body": "## Summary\n\nSince ~v2.1.31 (Feb 4, 2026), **users cannot access their session history**. `/resume` shows only ~5-10 recent sessions despite hundreds existing on disk. `claude -c` fails to find previous sessions. This affects **all platforms** (macOS, Windows, Linux) and **all versions from 2.1.31 through 2.1.42+**.\n\nSession data is NOT lost \u2014 `.jsonl` files are intact on disk \u2014 but there is no working UI to access them.\n\n## Three Distinct Root Causes\n\nCommunity reverse-engineering of `cli.js` (v2.1.39) has identified **three separate bugs** causing this:\n\n### Bug 1: `sessions-index.json` stopped being written (~Feb 4, 00:33 UTC)\n\nNew sessions are no longer added to the index. Across multiple reporters:\n- 3,080+ `.jsonl` files on disk, only 740 indexed (24%), **0% indexed after Feb 4**\n- `~/.claude/usage-data/` also stopped updating on the same date\n- Summary generation degraded earlier (late January) \u2014 all new entries have `\"summary\": \"\"`\n- Affects every project directory, not project-specific\n\n**Timeline correlation:** This coincides with Anthropic's token usage reduction rollout (ref #16157). Session summary generation costs 100K+ input tokens per session \u2014 disabling it would reduce platform-wide token consumption.\n\n### Bug 2: `/resume` picker hardcoded to 10-session initial batch\n\nSource analysis of minified `cli.js` (v2.1.39) by @tracymelody:\n- `sessions-index.json` is **never read** \u2014 the string doesn't appear in the source\n- `/resume` uses `readdirSync` to scan `.jsonl` files, sorts by mtime\n- **Initial batch is hardcoded to 10** (`K=10` parameter)\n- Load-more trigger depends on terminal height: `page_size = Math.floor((terminal_rows - ~10) / 3)`\n- With a 24-row terminal, `page_size=4`, the load-more trigger doesn't fire (Ink SelectInput wraps at list boundary instead of scrolling)\n- `C7z` filter discards \"lite\" sessions (no `firstPrompt` AND no `customTitle`), further reducing the visible list\n\n**One-line fix** (from @tracymelody on #24435):\n```bash\nsed -i 's/async function _c1(A,q,K=10)/async function _c1(A,q,K=500)/' cli.js\n```\n\n### Bug 3: Case-sensitive worktree path comparison (Windows)\n\nFrom @agatho on #24729:\n- Multi-worktree code path does `dir.name === s` (case-sensitive)\n- `git worktree list` returns `C:/...` but project directory stores `c--...`\n- Comparison fails silently, sessions not found\n- Only triggers with >1 git worktree\n\n## Steps to Reproduce\n\n```bash\n# Check how many sessions exist on disk vs. what /resume shows\nfind ~/.claude/projects -name \"*.jsonl\" | wc -l\n# Compare to /resume picker count (typically shows 5-10)\nclaude --resume\n```\n\n```bash\n# Check index staleness\nfor idx in ~/.claude/projects/*/sessions-index.json; do\n  echo \"$idx: $(stat -f '%Sm' \"$idx\" 2>/dev/null || stat -c '%y' \"$idx\" 2>/dev/null)\"\ndone\n# All will show dates on or before Feb 4, 2026\n```\n\n## Expected Behavior\n\n`/resume` and `claude -c` should list ALL previous sessions, not just the 5-10 most recent.\n\n## Workarounds\n\n1. **`claude --resume <keyword>`** \u2014 bypasses the picker entirely, searches all `.jsonl` files on disk\n2. **`claude --resume <session-id>`** \u2014 direct resume by UUID\n3. **`claude install 2.1.29`** \u2014 last working version\n4. **Community repair scripts:** [bosmadev/claude](https://github.com/bosmadev/claude), [tirufege/gist](https://gist.github.com/tirufege/0720c288092c1a3a4750f7c198aa524b)\n\n## Environment\n\n- **Versions affected:** 2.1.31 through 2.1.42 (all current)\n- **Last working version:** 2.1.29\n- **Platforms:** macOS, Windows 11, Linux (Ubuntu), WSL2\n- **Terminals:** Ghostty, iTerm2, Terminal.app, Windows Terminal, VS Code integrated\n\n## Consolidates These Issues\n\nThis issue consolidates 12+ reports of the same bug cluster:\n\n| Issue | Title | \ud83d\udc4d |\n|-------|-------|----|\n| #22878 | /resume not listing sessions in v2.1.30 | 13 |\n| #23614 | sessions-index.json stops being updated after v2.1.31 | 7 |\n| #25032 | sessions-index.json not updated, /resume shows stale sessions | 3 |\n| #24435 | Resume picker only shows ~8 most recent sessions | 2 |\n| #22462 | sessions-index.json not updated, claude -c fails | 1 |\n| #23421 | sessions-index.json stops updating (Linux/WSL2) | 1 |\n| #22205 | Sessions not added to sessions-index.json | 1 |\n| #25729 | /resume only shows ~5-10 most recent sessions | 0 |\n| #25552 | /resume missing sessions: index out of sync | 0 |\n| #25685 | Session resume picker shows no sessions | 0 |\n| #24729 | sessions-index.json not updated (Windows worktree bug) | 1 |\n| #25130 | [FEATURE] Allow /resume to display full session history | 0 |\n\n**Combined engagement: 29+ thumbs-up, 50+ comments, 12+ independent reports across 3 platforms.**\n\nPlease close the above as duplicates pointing here if this is adopted as the canonical tracking issue.",
          "state": "closed",
          "age_days": 18,
          "comments": 16,
          "reactions_total": 60,
          "upvotes": 60,
          "downvotes": 0,
          "created_at": "2026-02-16T14:54:09Z",
          "url": "https://github.com/anthropics/claude-code/issues/26123",
          "labels": [
            "bug"
          ],
          "cluster_id": 2
        },
        {
          "number": 26805,
          "title": "[BUG] Cowork UI: Responses not rendering until user types; \"working through complex response\" hangs indefinitely",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\nIn Claude Cowork (desktop app), two related UI issues make the app nearly unusable during longer sessions:\n\n**Issue 1: Responses don't render until the user types something**\nClaude's response completes processing but the UI does not display it. The response only appears after the user types a new message. This means the user cannot read Claude's output and cannot know what to do next \u2014 they only discover responses retroactively after sending a follow-up message.\n\n**Issue 2: \"Working through a complex response\" spinner hangs indefinitely**\nDuring sessions with multiple tool calls or background processing (file reads, Bash commands, agent tasks), the UI shows \"Working through a complex response...\" and hangs for several minutes with no progress. The response may have completed on the backend but the UI never updates.\n\n**Combined effect:**\nThe user cannot have a functional conversation. They cannot see responses, cannot answer questions Claude asks, and lose typed input when the UI is stuck. This occurred repeatedly in a single session today (2026-02-19) causing significant workflow disruption and lost context.\n\n**Environment:**\n- App: Claude Cowork (desktop, macOS)\n- - Session type: Long session with multiple tool calls (Bash, file reads, Notion MCP, Task agents)\n- - - Vault: Obsidian vault mounted as working directory\n\n### What Should Happen?\n\nResponses should render immediately when Claude finishes generating them, without requiring user input to trigger display. The \"Working through a complex response\" state should either resolve in a reasonable time or surface an error \u2014 it should never hang indefinitely with no feedback or recovery option.\n\n### Error Messages/Logs\n\n```shell\n\n```\n\n### Steps to Reproduce\n\n1. Open Claude Cowork (macOS desktop app)\n2. 2. Mount an Obsidian vault as the working directory\n3. 3. Start a long session involving multiple tool calls (Bash commands, file reads/writes, Notion MCP calls, Task agent spawning)\n4. 4. Claude completes a response after tool use\n5. 5. **Bug 1:** The response does not appear in the UI. The chat window shows no new message. Type anything in the input box \u2014 the response appears only after sending the new message.\n6. 6. For Bug 2: During a session with heavy processing (multiple sequential tool calls), observe the UI display \"Working through a complex response...\" and wait. The spinner persists for 3-5+ minutes with no resolution and no error message. The UI provides no way to cancel or recover.\n**Notes:**\n- Both bugs appear more frequently as session length increases (longer context window)\n- - Bug 1 appears to be a rendering/state update issue \u2014 the response exists but the UI doesn't reflect it\n- - - Bug 2 may be related to the UI losing sync with the backend during long processing chains\n- - - - Reloading the app clears the session context, so there is no viable recovery path\n\n### Claude Model\n\nNone\n\n### Is this a regression?\n\nYes, this worked in a previous version\n\n### Last Working Version\n\n_No response_\n\n### Claude Code Version\n\n2.1.45 (Claude Code) \u2014 running via Claude Cowork desktop app\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nmacOS\n\n### Terminal/Shell\n\nTerminal.app (macOS)\n\n### Additional Information\n\n_No response_",
          "state": "closed",
          "age_days": 15,
          "comments": 45,
          "reactions_total": 24,
          "upvotes": 24,
          "downvotes": 0,
          "created_at": "2026-02-19T06:40:27Z",
          "url": "https://github.com/anthropics/claude-code/issues/26805",
          "labels": [
            "bug",
            "oncall"
          ],
          "cluster_id": 2
        },
        {
          "number": 26428,
          "title": "[BUG] [1M] Sonnet 1M context window disappeared after Claude Code update to v2.1.45 \u2014 regression on Max plan ($200/mo)",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\n\nAfter Claude Code auto-updated to **v2.1.45** (with the Sonnet 4.6 release on Feb 17, 2026), the `sonnet[1m]` model option that was **previously working** now returns:\n\n```\nSonnet 4.6 with 1M context is not available for your account.\nLearn more: https://code.claude.com/docs/en/model-config#extended-context-with-1m\n```\n\nBoth attempts fail:\n\n```\n\u276f /model sonnet[1m]\n  \u23bf  Sonnet 4.6 with 1M context is not available for your account.\n\n\u276f /model claude-sonnet-4-6[1m]\n  \u23bf  Sonnet 4.6 with 1M context is not available for your account.\n```\n\n**This is a regression.** The `sonnet[1m]` model was working correctly on my X20 Max subscription for weeks prior to this update. It stopped working immediately after the update to v2.1.45 coinciding with the Sonnet 4.6 / Opus 4.6 release.\n\n## Expected Behavior\n\nMax plan subscribers ($200/mo) should retain access to the 1M context window for Sonnet models, especially since:\n\n1. It was **already working** before the update \u2014 this is a regression, not a missing feature\n2. The Max plan at $200/mo is explicitly designed for power users with heavy context needs\n3. The documentation at `code.claude.com` still references `sonnet[1m]` as a valid option\n\n## Environment\n\n- **Plan:** Claude Max ($200/mo)\n- **Claude Code Version:** v2.1.45\n- **Models affected:** Sonnet 4.6 with `[1m]` suffix (previously Sonnet 4.5 `[1m]` was working)\n- **Platform:** macOS Tahoe 26.2\n- **Terminal/Shell:** zsh (bash)\n\n## Real-World Impact\n\nI work on enterprise-scale modernization projects (Java 17 / Spring Boot / Clean Architecture) with codebases that routinely exceed 200K tokens when including source files, tests, configuration, and documentation. The 1M context window is not a luxury \u2014 it's essential for:\n\n- Full codebase understanding during refactoring sessions\n- Cross-module dependency analysis\n- Large-scale code generation with full project context\n- Multi-agent AI workflows that consume significant context at startup (MCP servers, CLAUDE.md, etc.)\n\nLosing this capability mid-workflow is disruptive and makes the Max plan significantly less valuable.\n\n## Related Issues\n\n- #23822 \u2014 Same error on Max plan ($200/mo), reported Feb 6, 2026\n- #23905 \u2014 Opus 4.6 `[1m]` rejected on Max subscriptions\n- #23700 \u2014 Long context beta unavailable for Max plan with Opus 4.6\n- #15057 \u2014 Sonnet 1M regression on Max 20x subscription (Dec 2025)\n- #10219 \u2014 `sonnet[1m]` doesn't work on Max 20x account\n- #8381 \u2014 Feature request for 1M context on Sonnet 4.5 (open since Sep 2025)\n\n## Additional Context\n\nThe `/status` output confirms the subscription is correctly identified as Claude Max. No `ANTHROPIC_API_KEY` environment variable is set. Standard troubleshooting steps (logout, update, restart terminal, re-authenticate) have all been attempted without success.\n\n### What Should Happen?\n\n## What Should Happen?\n\n1. **Restore 1M context access** for Max plan subscribers on Sonnet 4.6 (and Sonnet 4.5 if still supported)\n2. If this is an intentional change, **communicate it clearly** rather than silently removing access\n3. Do not allow model updates to regress features that were previously available to paying subscribers\n\n\n### Error Messages/Logs\n\n```shell\nclaude\n\n \u2590\u259b\u2588\u2588\u2588\u259c\u258c   Claude Code v2.1.45\n\u259d\u259c\u2588\u2588\u2588\u2588\u2588\u259b\u2598  Sonnet 4.6 \u00b7 Claude Max\n  \u2598\u2598 \u259d\u259d    ~/MY_STUFF/projects\n\u276f /model sonnet[1m]                                                                                                                                                                         \n  \u23bf  Sonnet 4.6 with 1M context is not available for your account. Learn more: https://code.claude.com/docs/en/model-config#extended-context-with-1m                                        \n                                                                                                                                                                                            \n\u276f /model claude-sonnet-4-6[1m]                                                                                                                                                              \n  \u23bf  Sonnet 4.6 with 1M context is not available for your account. Learn more: https://code.claude.com/docs/en/model-config#extended-context-with-1m\n```\n\n### Steps to Reproduce\n\n1. Have an active Max plan ($200/mo) subscription\n2. Update Claude Code to v2.1.45 (or let it auto-update)\n3. Run `/model sonnet[1m]` or `/model claude-sonnet-4-6[1m]`\n4. Observe the error: \"Sonnet 4.6 with 1M context is not available for your account\"\n\n### Claude Model\n\nSonnet (default)\n\n### Is this a regression?\n\nYes, this worked in a previous version\n\n### Last Working Version\n\nv2.1.44\n\n### Claude Code Version\n\nv2.1.45\n\n### Platform\n\nOther\n\n### Operating System\n\nmacOS\n\n### Terminal/Shell\n\nTerminal.app (macOS)\n\n### Additional Information\n\nThe `/status` output confirms the subscription is correctly identified as Claude Max. No `ANTHROPIC_API_KEY` environment variable is set. Standard troubleshooting steps (logout, update, restart terminal, re-authenticate) have all been attempted without success.",
          "state": "closed",
          "age_days": 17,
          "comments": 46,
          "reactions_total": 19,
          "upvotes": 19,
          "downvotes": 0,
          "created_at": "2026-02-17T22:28:33Z",
          "url": "https://github.com/anthropics/claude-code/issues/26428",
          "labels": [
            "bug"
          ],
          "cluster_id": 2
        },
        {
          "number": 30021,
          "title": "\"Create PR\" button missing in Claude Code web UI after push",
          "body": "## Bug report\n\nAfter Claude Code on iOS completes work \u2014 commits and pushes to a remote branch \u2014 the UI shows a section at the bottom that appears to be where a \"Create PR\" button should render, but no button is actually present.\n\n### Steps to reproduce\n\n1. Use Claude Code on iOS\n2. Have Claude make changes, commit, and push to a feature branch\n3. Observe the bottom of the session after the push completes\n\n### Expected behavior\n\nA \"Create PR\" button should appear, allowing the user to quickly open a pull request for the branch that was just pushed.\n\n### Actual behavior\n\nThe area where the button should be is visible but empty \u2014 no button renders. The user has to manually navigate to GitHub to create the PR.\n\n### Screenshot\n\n![image](https://github.com/user-attachments/assets/b19c3dfc-a284-46c2-abcc-4a9052c8e237)\n\n### Environment\n\n- Platform: Claude Code iOS, version 1.260223.3\n- Date: 2026-03-02",
          "state": "open",
          "age_days": 4,
          "comments": 19,
          "reactions_total": 36,
          "upvotes": 36,
          "downvotes": 0,
          "created_at": "2026-03-02T08:27:43Z",
          "url": "https://github.com/anthropics/claude-code/issues/30021",
          "labels": [
            "bug",
            "area:claude-code-web",
            "platform:web",
            "area:ui"
          ],
          "cluster_id": 2
        },
        {
          "number": 28847,
          "title": "Race condition: .claude.json corruption when running multiple instances concurrently",
          "body": "## Description\n\nRunning multiple Claude Code instances simultaneously causes repeated `.claude.json` corruption due to concurrent writes without file locking or atomic write operations.\n\n## Environment\n\n- **Claude Code version:** 2.1.59 (also observed on earlier versions back to at least ~2.1.56)\n- **OS:** Windows 11 Pro 10.0.26200\n- **Shell:** Windows Powershell\n\n## Steps to Reproduce\n\n1. Open 5+ terminal windows\n2. Run `claude` in each\n3. Use them concurrently for normal work\n4. Observe repeated corruption warnings on startup and during sessions\n\n## Observed Behavior\n\nEach instance detects the corruption, backs up the broken file, and recovers \u2014 but this produces:\n\n- **Hundreds of `.claude.json.corrupted.*` backup files** in `~/.claude/backups/` (305 files accumulated in a single day)\n- **Spam warning messages** on every startup:\n  ```\n  Claude configuration file at C:\\Users\\<user>\\.claude.json is corrupted: JSON Parse error: Unexpected EOF\n  The corrupted file has been backed up to: C:\\Users\\<user>\\.claude\\backups\\.claude.json.corrupted.<timestamp>\n  ```\n- The corrupted files show truncated JSON (partial writes caught mid-stream)\n\n## Expected Behavior\n\nMultiple concurrent instances should be able to coexist without corrupting the shared config file. No warning spam, no backup file accumulation.\n\n## Suggested Fix\n\n_Redacted by Claude Code maintainers to avoid confusion as users find this issue_.\n\n## Impact\n\n- **Not data-losing** \u2014 Claude Code recovers gracefully each time\n- **Annoying** \u2014 constant warning messages disrupt workflow\n- **Disk clutter** \u2014 hundreds of small backup files accumulate rapidly\n- **Legitimate workflow** \u2014 running multiple instances in parallel (e.g., one per project/repo) is a common power-user pattern that should be supported",
          "state": "closed",
          "age_days": 9,
          "comments": 26,
          "reactions_total": 26,
          "upvotes": 26,
          "downvotes": 0,
          "created_at": "2026-02-26T04:13:25Z",
          "url": "https://github.com/anthropics/claude-code/issues/28847",
          "labels": [
            "bug",
            "has repro",
            "area:core"
          ],
          "cluster_id": 2
        }
      ]
    },
    {
      "id": 1,
      "issue_count": 295,
      "stats": {
        "total_reactions": 883,
        "total_comments": 1050,
        "avg_age_days": 13.8
      },
      "analysis": {
        "cluster_title": "Keyword Group: Bash Git Tool Plugin Output",
        "summary": "This semantic cluster was defined by the high density of the following keywords: bash, git, tool, plugin, output.",
        "hardest_problems": [
          "[BUG] claude --worktree (-w) does not create a git worktree",
          "SSH remote connection fails on Windows \u2014 hardcoded /usr/bin/ssh path",
          "AskUserQuestion auto-submits without waiting for user input in skill/command context"
        ],
        "hallucination_check": "Local heuristic analysis based on TF-IDF term frequencies and engagement metrics."
      },
      "top_issues": [
        {
          "number": 27044,
          "title": "[BUG] claude --worktree (-w) does not create a git worktree",
          "body": "## Description\n\n`claude -w` (or `claude --worktree`) does not create a new git worktree. After launching with the `-w` flag, `pwd` remains unchanged, `git worktree list` shows only the main working tree, and `.git` is still a directory (not a file, which would indicate a worktree).\n\n## Steps to Reproduce\n\n1. `cd` into a git repository\n2. Run `claude -w`\n3. Inside the Claude session, run `!pwd` and `!git worktree list`\n\n## Expected Behavior\n\nA new git worktree should be created and Claude should operate within it, isolating changes from the main working tree.\n\n## Actual Behavior\n\nNo worktree is created. Claude operates in the original directory as if `-w` was not passed.\n\n## Diagnostic Output\n\n```\npwd: /Users/<redacted>/Dev/<project>\ngit worktree list: /Users/<redacted>/Dev/<project>  e09387f [main]  (only one entry)\ngit rev-parse --show-toplevel: /Users/<redacted>/Dev/<project>\ngit branch --show-current: main\necho $CLAUDE_WORKTREE: (empty / not set)\nls -la .git: regular directory (not a file)\n```\n\nKey observations:\n- Only one worktree exists (the main working tree)\n- `.git` is a directory, not a file (worktrees use a `.git` file pointing to the main repo)\n- `CLAUDE_WORKTREE` environment variable is not set\n- `~/.claude-worktrees` directory does not exist\n\n## Environment\n\n- Claude Code version: 2.1.49\n- OS: macOS\n- Git: installed and functional\n- Executed inside a valid git repository on the `main` branch\n",
          "state": "closed",
          "age_days": 15,
          "comments": 24,
          "reactions_total": 41,
          "upvotes": 41,
          "downvotes": 0,
          "created_at": "2026-02-20T03:32:58Z",
          "url": "https://github.com/anthropics/claude-code/issues/27044",
          "labels": [],
          "cluster_id": 1
        },
        {
          "number": 25659,
          "title": "SSH remote connection fails on Windows \u2014 hardcoded /usr/bin/ssh path",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\nWhen using Claude Code Desktop on Windows and attempting to connect to a remote machine via SSH, it fails with:\n\nFailed to spawn /usr/bin/ssh: spawn /usr/bin/ssh ENOENT \n\nThe SSH spawning logic appears to hardcode the Unix path /usr/bin/ssh, which doesn't exist on Windows. On Windows, SSH is typically located at C:\\Windows\\System32\\OpenSSH\\ssh.exe or C:\\Program Files\\Git\\usr\\bin\\ssh.exe.\n\nExpected behavior: Claude Code should detect the platform and resolve the SSH binary path accordingly (e.g., using where ssh on Windows or respecting the system PATH).\n\nEnvironment:\n- Windows 11 Pro\n- Claude Code Desktop (latest)\n- OpenSSH 10.2p1 installed and working from terminal\n\n### What Should Happen?\n\nClaude Code should detect the operating system platform and resolve the SSH binary path accordingly. On Windows, it should:\n\n1. Use the appropriate SSH executable locations (e.g., C:\\Windows\\System32\\OpenSSH\\ssh.exe or C:\\Program Files\\Git\\usr\\bin\\ssh.exe)\n2. Respect the system PATH environment variable to locate ssh\n3. Use platform detection to determine the correct binary path instead of hardcoding /usr/bin/ssh\n\nThis would allow remote SSH connections to work properly on Windows systems with OpenSSH installed.\n\n### Error Messages/Logs\n\n```shell\n\n```\n\n### Steps to Reproduce\n\n1. Open Claude Code Desktop on Windows 11\n2. Navigate to the SSH feature/settings\n3. Attempt to configure a remote SSH connection to any machine\n4. Try to connect using the SSH remote feature\n\nExpected result: Connection should attempt to establish\nActual result: Connection fails with error \"Failed to spawn /usr/bin/ssh: spawn /usr/bin/ssh ENOENT\"\n\n### Claude Model\n\nNone\n\n### Is this a regression?\n\nNo, this never worked\n\n### Last Working Version\n\n_No response_\n\n### Claude Code Version\n\nlatest\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nWindows\n\n### Terminal/Shell\n\nPowerShell\n\n### Additional Information\n\n_No response_",
          "state": "closed",
          "age_days": 21,
          "comments": 11,
          "reactions_total": 53,
          "upvotes": 34,
          "downvotes": 0,
          "created_at": "2026-02-14T00:55:40Z",
          "url": "https://github.com/anthropics/claude-code/issues/25659",
          "labels": [
            "bug",
            "platform:windows",
            "oncall"
          ],
          "cluster_id": 1
        },
        {
          "number": 29674,
          "title": "AskUserQuestion auto-submits without waiting for user input in skill/command context",
          "body": "# Bug: AskUserQuestion auto-submits without waiting for user input in skill/command context\n\n## Environment\n- **Claude Code version:** 2.1.63\n- **OS:** macOS (Darwin 25.2.0)\n- **Model:** Opus 4.6 (Claude Max)\n- **Shell:** zsh\n\n## Summary\n\n`AskUserQuestion` tool calls auto-submit immediately with empty responses when invoked from within a custom command/skill (via `/command-name`). The tool works correctly in normal conversation \u2014 it renders the interactive UI and blocks for user input. But when the same tool is called inside a command, it returns instantly without displaying the question to the user.\n\n## Regression\n\nThis was working correctly on **2/27/2026**. Broke on **2/28/2026**. Same version (2.1.63), same files, same projects.\n\n## Reproduction Steps\n\n### Working case (normal conversation):\n\n1. Open Claude Code\n2. Type: \"Ask me 3 questions about my favorite food using AskUserQuestion\"\n3. **Result:** Questions render with interactive multi-select UI, Claude waits for input\n\n### Broken case (command context):\n\n1. Create a minimal command file at `.claude/commands/test-ask.md`:\n\n```markdown\n---\ndescription: Test AskUserQuestion in command context\nallowed-tools: AskUserQuestion\n---\n\nAsk the user what their favorite color is using AskUserQuestion with options: Red, Blue, Green.\n\nThen output their answer.\n```\n\n2. Run `/test-ask`\n3. **Expected:** Question renders, Claude waits for user to select an option\n4. **Actual:** \"User answered Claude's questions:\" appears with empty response, Claude immediately continues without waiting\n\n## Evidence from real commands\n\nMultiple custom commands affected \u2014 all exhibit the same behavior:\n\n```\n# /plan-feature output (truncated):\n\u23fa User answered Claude's questions:\n  \u23bf\n\n\u23fa Got it \u2014 Medium size, P1 priority, Beta rollout...\n\n\u23fa User answered Claude's questions:\n  \u23bf\n\n\u23fa Round 2: Visual Identity & Professionalism\n\n\u23fa User answered Claude's questions:\n  \u23bf\n```\n\nEvery `AskUserQuestion` call returns empty and Claude proceeds as if the user answered. This happens for:\n- Commands with explicit `Ask using AskUserQuestion:` formatting with structured options\n- Commands with bullet-point questions and a header instruction to use AskUserQuestion\n- Both custom commands and commands from the official plugin system\n\n## What I've ruled out\n\n| Hypothesis | Test | Result |\n|---|---|---|\n| Custom command file regression | Diffed all command files against last-known-working state | No relevant changes to question-asking logic |\n| Plugin interference (custom matt-toolkit plugin with Stop/SessionStart hooks) | Disabled plugin, retested | Bug persists with plugin disabled |\n| Settings/permissions issue | Reviewed global and project settings.json | No AskUserQuestion-related entries |\n| PreToolUse hook interference | Hook only matches Bash tool, only active during autonomous loops | Not relevant |\n| Project-specific issue | Reproduced across 3 different projects (claude-toolkit, joyride-website, cyoa) | Bug is global |\n\n## Impact\n\nThis breaks any command/skill that relies on interactive discovery or user input. For our toolkit, this makes `/plan-feature` and `/research-feature` completely unusable \u2014 they speed through 8-10 rounds of questions without collecting any user input, then generate specs based on empty answers.",
          "state": "closed",
          "age_days": 6,
          "comments": 12,
          "reactions_total": 30,
          "upvotes": 30,
          "downvotes": 0,
          "created_at": "2026-02-28T21:40:29Z",
          "url": "https://github.com/anthropics/claude-code/issues/29674",
          "labels": [
            "bug",
            "has repro",
            "platform:macos",
            "area:tools",
            "regression",
            "area:skills"
          ],
          "cluster_id": 1
        },
        {
          "number": 28364,
          "title": "[Bug] EINVAL on Windows: Bash commands fail - cannot open task output file",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\n## Description\nAll Bash commands fail with EINVAL error on Windows. Even `echo test` fails.\nThis was working last night, broken this morning without any changes on my end.\n\n## Error\nError: EINVAL: invalid argument, open 'C:\\Users\\hamma\\AppData\\Local\\Temp\\claude\\C--Users-hamma-git-whatap-go-sdk\\tasks\\bwtfhc68s.output'\n\n## Key findings\n- Happens in ALL directories, not project-specific\n- Manually creating files in that path works fine (New-Item succeeds)\n- OS permissions are normal\n- No Windows Defender blocks\n- Disk space sufficient (107GB free)\n- No Windows updates since Feb 11\n\n## Environment\n- Claude Code version: 2.1.53\n- OS: Windows 10 (Build 26100.7840)\n- Shell: PowerShell\n\n## Steps to reproduce\n1. Open any directory on Windows\n2. Run `claude`\n3. Try any Bash command (even `echo test`)\n4. EINVAL error every time\n\n### What Should Happen?\n\n## Description\nAll Bash commands fail with EINVAL error on Windows. Even `echo test` fails.\nThis was working last night, broken this morning without any changes on my end.\n\n## Error\nError: EINVAL: invalid argument, open 'C:\\Users\\hamma\\AppData\\Local\\Temp\\claude\\C--Users-hamma-git-whatap-go-sdk\\tasks\\bwtfhc68s.output'\n\n## Key findings\n- Happens in ALL directories, not project-specific\n- Manually creating files in that path works fine (New-Item succeeds)\n- OS permissions are normal\n- No Windows Defender blocks\n- Disk space sufficient (107GB free)\n- No Windows updates since Feb 11\n\n## Environment\n- Claude Code version: 2.1.53\n- OS: Windows 10 (Build 26100.7840)\n- Shell: PowerShell\n\n## Steps to reproduce\n1. Open any directory on Windows\n2. Run `claude`\n3. Try any Bash command (even `echo test`)\n4. EINVAL error every time\n\n### Error Messages/Logs\n\n```shell\n\n```\n\n### Steps to Reproduce\n\n1. Install Claude Code 2.1.53 on Windows (native installer)\n2. Open PowerShell\n3. cd to any directory (e.g., `mkdir C:\\test-claude && cd C:\\test-claude`)\n4. Run `claude`\n5. Type any command that triggers Bash, e.g., `echo test`\n6. Error occurs:\n   Error: EINVAL: invalid argument, open 'C:\\Users\\<user>\\AppData\\Local\\Temp\\claude\\C--test-claude\\tasks\\<id>.output'\n\nNo specific files or code needed - happens with any Bash command in any directory.\n\nVerified manually that file creation in the same temp path works:\nNew-Item \"C:\\Users\\hamma\\AppData\\Local\\Temp\\claude\\C--test-claude\\tasks\\test.txt\" -Force -Value \"test\"\n# succeeds without error\n\nThe issue appears to be in how Claude Code internally opens the .output file, not an OS-level permission or path problem.\n\n### Claude Model\n\nNone\n\n### Is this a regression?\n\nYes, this worked in a previous version\n\n### Last Working Version\n\n_No response_\n\n### Claude Code Version\n\n2.1.53\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nWindows\n\n### Terminal/Shell\n\nPowerShell\n\n### Additional Information\n\n_No response_",
          "state": "open",
          "age_days": 10,
          "comments": 18,
          "reactions_total": 14,
          "upvotes": 14,
          "downvotes": 0,
          "created_at": "2026-02-25T02:14:52Z",
          "url": "https://github.com/anthropics/claude-code/issues/28364",
          "labels": [
            "bug",
            "duplicate",
            "platform:windows",
            "area:bash"
          ],
          "cluster_id": 1
        },
        {
          "number": 26413,
          "title": "[BUG] Bash tool: all commands fail with Bad file descriptor on Windows (Git Bash stdout broken)",
          "body": "## Bug Description\n\nEvery command run through the Bash tool fails with exit code 1. Stdout file descriptor (fd 1) is broken - bash cannot write any output. PowerShell commands work fine through the same tool.\n\nThis is **not** the same issue as #18469 (which was silent empty stdout with exit code 0). This produces an explicit Bad file descriptor error on fd 1.\n\n## Environment\n\n- **Claude Code version:** 2.1.45\n- **Platform:** Windows 11 (MINGW64_NT-10.0-26200 3.6.4-b9f03e96.x86_64)\n- **Shell:** Git Bash (C:\\Program Files\\Git\\usr\\bin\\bash.exe)\n- **Sandbox:** Disabled (sandbox.enabled: false)\n- **No CLAUDE_CODE_SHELL or CLAUDE_CODE_GIT_BASH_PATH env vars set**\n\n## Reproduction\n\nAny bash command fails:\n\n\nBash(date)        -> Exit code 1 (no output)\nBash(echo hello)  -> Exit code 1 (no output)\nBash(pwd)         -> Exit code 1 (no output)\nBash(env | sort)  -> Exit code 2 (no output)\n\n\nPowerShell wrapper works fine:\n\n\nBash(powershell -Command Get-Date)  -> Tuesday, February 17, 2026 3:46:15 PM\n\n\n## Root Cause (diagnosed)\n\nRunning bash directly via PowerShell with stderr captured reveals the error:\n\npowershell\n& 'C:\\Program Files\\Git\\bin\\bash.exe' -c 'echo hello' 2>&1\n\n\nOutput:\n\n/usr/bin/bash: line 1: echo: write error: Bad file descriptor\n\n\nThe stdout pipe that Claude Code creates for capturing bash output is not valid from MSYS2/Git Bash's perspective. Bash builtins (echo, pwd) fail because fd 1 is broken. PowerShell works because it handles Windows pipe handles natively.\n\n## Diagnostic Steps Taken\n\n1. **Confirmed PATH is correct** - C:\\Program Files\\Git\\usr\\bin is on PATH, bash binary exists and runs\n2. **Removed ~/.bashrc** - no change, still broken (rules out #18469 root cause)\n3. **Checked system bashrc/profile** - standard Git for Windows defaults, nothing unusual\n4. **No CLAUDE env vars** - CLAUDE_CODE_SHELL and CLAUDE_CODE_GIT_BASH_PATH are not set\n5. **Git Bash works normally** outside of Claude Code (user runs it daily)\n\n## Expected Behavior\n\nBash commands should execute and return stdout, as they do in a normal Git Bash terminal.\n\n## Likely Cause\n\nThe child process spawning mechanism is not passing stdout pipe handles in a way that MSYS2's POSIX fd layer can use. This may be related to how child_process.spawn() or similar Node.js APIs create pipes on Windows - the Windows HANDLEs are not being translated to valid POSIX file descriptors by MSYS2's crt0 initialization.\n\n## Related Issues\n\n- #18469 - Bash tool doesn't capture stdout from shell scripts on Windows (different symptoms: silent empty stdout, exit 0; fixed in v2.1.27)\n- #20453 - CLAUDE_CODE_SHELL not respected on Windows (closed as dup)\n- #20034 - Hook stdout routed to stderr in Git Bash (closed as dup of #18469)\n- #25558 - CLAUDE_CODE_SHELL ignored on Windows (open)",
          "state": "closed",
          "age_days": 17,
          "comments": 11,
          "reactions_total": 21,
          "upvotes": 21,
          "downvotes": 0,
          "created_at": "2026-02-17T21:07:05Z",
          "url": "https://github.com/anthropics/claude-code/issues/26413",
          "labels": [],
          "cluster_id": 1
        }
      ]
    },
    {
      "id": 0,
      "issue_count": 273,
      "stats": {
        "total_reactions": 563,
        "total_comments": 1301,
        "avg_age_days": 13.9
      },
      "analysis": {
        "cluster_title": "Keyword Group: Issues Shell _No _No Response_ Response_",
        "summary": "This semantic cluster was defined by the high density of the following keywords: issues, shell, _no, _no response_, response_.",
        "hardest_problems": [
          "[BUG] Cowork: RPC error -1 virtiofs/Plan9 mount fails with \"Plan9 mount failed: bad address\"",
          "[BUG] Claude Desktop cannot install on Windows - redirects to Microsoft Store during installation",
          "Claude Desktop 1.1.4173 crashes on startup \u2014 no window renders, process visible in Task Manager"
        ],
        "hallucination_check": "Local heuristic analysis based on TF-IDF term frequencies and engagement metrics."
      },
      "top_issues": [
        {
          "number": 26554,
          "title": "[BUG] Cowork: RPC error -1 virtiofs/Plan9 mount fails with \"Plan9 mount failed: bad address\"",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\nWindows 11 Pro, Hyper-V enabled, RPC error -1: failed to ensure virtiofs mount: Plan9 mount failed: bad address\" and mention it recurs after every question. Can no longer use Claude\n\n### What Should Happen?\n\nFix bug  please\n\n### Error Messages/Logs\n\n```shell\n\n```\n\n### Steps to Reproduce\n\nOpen Claud cowork. \n\n### Claude Model\n\nNone\n\n### Is this a regression?\n\nYes, this worked in a previous version\n\n### Last Working Version\n\n_No response_\n\n### Claude Code Version\n\n1.1.3363.0\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nWindows\n\n### Terminal/Shell\n\nNon-interactive/CI environment\n\n### Additional Information\n\n_No response_",
          "state": "open",
          "age_days": 16,
          "comments": 49,
          "reactions_total": 56,
          "upvotes": 56,
          "downvotes": 0,
          "created_at": "2026-02-18T11:47:25Z",
          "url": "https://github.com/anthropics/claude-code/issues/26554",
          "labels": [
            "bug",
            "oncall"
          ],
          "cluster_id": 0
        },
        {
          "number": 28892,
          "title": "[BUG] Claude Desktop cannot install on Windows - redirects to Microsoft Store during installation",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\n## What's Wrong?\nClaude Desktop installation fails on Windows. During installation, a popup appears saying \n\"Get an app to open this 'claude' link - Your PC doesn't have an app that can open this link. \nTry looking for a compatible app in the Microsoft Store.\" \nThe popup disappears on its own and the installation does not complete successfully. \nClaude Desktop does not appear in the Start menu after installation.\n\n### What Should Happen?\n\n## Troubleshooting Already Attempted\n- Removed existing Claude packages via PowerShell:\n  Get-AppxPackage -Name \"*Claude*\" -AllUsers | Remove-AppxPackage -AllUsers\n- Enabled Developer Mode in Settings \u2192 System \u2192 For Developers\n- Changed \"Choose where to get apps\" to \"Anywhere\" in Settings \u2192 Apps \u2192 Advanced app settings\n- Tried installing without admin rights (double-click only)\n- Tried installing with Run as administrator\n- Restarted PC between attempts\n- Re-downloaded fresh installer each time\n- None of the above resolved the issue\n\n## Expected Behavior\nClaude Desktop should install successfully and appear in the Start menu.\n\n## Actual Behavior\nInstallation fails silently. Popup redirects to Microsoft Store where Claude Desktop cannot be found.\n\n### Error Messages/Logs\n\n```shell\n\n```\n\n### Steps to Reproduce\n\n## Steps to Reproduce\n1. Download Claude Desktop installer from claude.ai/download\n2. Run installer (tried both normal and Run as administrator)\n3. Popup \"Get an app to open this 'claude' link\" appears during installation\n4. Popup disappears on its own\n5. Claude Desktop not found in Start menu \u2014 installation failed\n\n### Claude Model\n\nNone\n\n### Is this a regression?\n\nNo, this never worked\n\n### Last Working Version\n\n_No response_\n\n### Claude Code Version\n\nDownloaded from claude.ai/download on 26 Feb 2026\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nWindows\n\n### Terminal/Shell\n\nOther\n\n### Additional Information\n\n_No response_",
          "state": "open",
          "age_days": 8,
          "comments": 62,
          "reactions_total": 21,
          "upvotes": 20,
          "downvotes": 0,
          "created_at": "2026-02-26T07:39:00Z",
          "url": "https://github.com/anthropics/claude-code/issues/28892",
          "labels": [
            "invalid"
          ],
          "cluster_id": 0
        },
        {
          "number": 28304,
          "title": "Claude Desktop 1.1.4173 crashes on startup \u2014 no window renders, process visible in Task Manager",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\nTitle: Claude Desktop 1.1.4173 crashes on startup \u2014 no window renders, process visible in Task Manager\nDescription:\nAfter auto-updating to version 1.1.4173, Claude Desktop on Windows no longer launches. The process appears in Task Manager but no window or UI is ever displayed. The application crashes on start.\n\nClaude Desktop version: 1.1.4173\nWindows version: Windows 11 Pro 25H2 (OS Build 26200.7623)\nPrevious version worked: Yes (version unknown \u2014 was working before auto-update)\nClaude Code TUI: Works fine\nImpact: Browser integration (MCP connectors, desktop extensions) requires the Desktop GUI on Windows, so this blocks that functionality entirely\n\nSteps to reproduce:\n\nLet Claude Desktop auto-update to 1.1.4173\nLaunch Claude Desktop\nProcess appears in Task Manager but no window renders\nApplication crashes\n\nWorkaround: None found. Using Claude Code TUI and claude.ai in browser as fallback, but without desktop extension/MCP support.\n\n### What Should Happen?\n\nThe Claude Desktop should show a GUI.\n\n### Error Messages/Logs\n\n```shell\n\n```\n\n### Steps to Reproduce\n\nLet Claude Desktop auto-update to 1.1.4173\nLaunch Claude Desktop\nProcess appears in Task Manager but no window renders\nApplication crashes\n\n### Claude Model\n\nOpus\n\n### Is this a regression?\n\nYes, this worked in a previous version\n\n### Last Working Version\n\n1.1.39\n\n### Claude Code Version\n\n1.1.4173\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nWindows\n\n### Terminal/Shell\n\nOther\n\n### Additional Information\n\n_No response_",
          "state": "open",
          "age_days": 10,
          "comments": 38,
          "reactions_total": 28,
          "upvotes": 28,
          "downvotes": 0,
          "created_at": "2026-02-24T21:38:49Z",
          "url": "https://github.com/anthropics/claude-code/issues/28304",
          "labels": [
            "invalid"
          ],
          "cluster_id": 0
        },
        {
          "number": 28537,
          "title": "[BUG] Hitting much faster usage limits than before",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\nHi Claude Code Team,\n\nI wanted to report an unexpected change in usage behavior that I noticed between yesterday and today.\n\nMy workflow has remained completely the same, but starting today I'm hitting usage limits significantly faster than before. Most notably, this is the **very first time I have ever filled the 5-hour session window in Claude Code Max 200$ sub** which has never happened to me before despite using it regularly with the same workload.\n\nThis strongly suggests something changed on the backend regarding how usage/tokens are being counted or how limits are being applied, rather than anything on my end.\n\nCould you please investigate whether there were any recent changes to usage calculation, rate limiting logic, or session window behavior? Happy to provide more details if helpful.\n\n### What Should Happen?\n\nUsage consumption should remain consistent with previous sessions when the workflow and prompts have not changed. The 5-hour session window should not be exhausted under normal, unchanged usage patterns.\n\n### Error Messages/Logs\n\n```shell\nNo explicit error messages. Symptom: Usage limits reached significantly faster than usual. First time ever hitting the 5-hour session window limit in Claude Code despite identical workflow.\n```\n\n### Steps to Reproduce\n\nUse Claude Code with an unchanged workflow (same types of prompts, same project structure as previous days)\nContinue working as normal throughout a session\nObserve that usage limits are hit significantly faster than on previous days with identical workloads\n\nNote: No specific code or file is needed to reproduce. The issue manifests across the entire session regardless of task type. This is the first time the 5-hour session window has ever been fully consumed despite the workflow being unchanged.\n\n### Claude Model\n\nNone\n\n### Is this a regression?\n\nYes, this worked in a previous version\n\n### Last Working Version\n\n_No response_\n\n### Claude Code Version\n\n2.1.52\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nWindows\n\n### Terminal/Shell\n\nCursor\n\n### Additional Information\n\n_No response_",
          "state": "open",
          "age_days": 9,
          "comments": 38,
          "reactions_total": 19,
          "upvotes": 19,
          "downvotes": 0,
          "created_at": "2026-02-25T14:23:59Z",
          "url": "https://github.com/anthropics/claude-code/issues/28537",
          "labels": [
            "bug",
            "platform:windows",
            "area:cost",
            "area:api",
            "api:anthropic"
          ],
          "cluster_id": 0
        },
        {
          "number": 26224,
          "title": "[BUG] [URGENT!!!] Claude Code is hanging / freezing / stuck on heaps of prompts for 5-20minutes or more.",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\nEver since the release of Opus 4.6 at least, and maybe slightly before (I can't remember exactly when it started), Claude gets stuck \"thinking\" for ~5-20 minutes, and sometimes even longer. Token usage does not go up during this time, and packet inspection shows it hanging on waiting for SSE events from Anthropic end for the given prompt. \n\nSometimes this can be fixed by sending a follow up prompt (doesn't matter what it contains), that kicks it back into action and allows the \"thinking\" prompt to start flowing again. Other times that doesn't work either.\n\nThis isn't Claude doing things behind the scenes, it's literally just stuck / blocking, and doing nothing at all.\n\nIn most cases, EVENTUALLY, it seems to unfreeze itself, somehow. But it tends to take over 5 minutes, sometimes beyond 20 mins.\n\nTheres a lot of other people complaining of similar problems, and I'm sure it is affecting a very large amount of users, but I haven't seen any evidence to see Anthropic is aware of it. \n\n### What Should Happen?\n\nNot freeze.\n\n### Error Messages/Logs\n\n```shell\n\n```\n\n### Steps to Reproduce\n\nI'm not sure how to trigger it deterministically. \n\n### Claude Model\n\nOpus\n\n### Is this a regression?\n\nYes, this worked in a previous version\n\n### Last Working Version\n\n_No response_\n\n### Claude Code Version\n\n2.1.38\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nOther Linux\n\n### Terminal/Shell\n\nWindows Terminal\n\n### Additional Information\n\nI'm using WSL2 Ubuntu through Windows terminal. \n\nI've tried deleting my entire Claude Code installation and starting with completely fresh config with no MCPs, skills, etc. Still same problem.\n\nI'm not using a VPN or anything. It was working fine until either the release of Opus 4.6, or just prior to it. That's when I first started having this problem.\n\nI'm using High thinking mode (but again, this is NOT stuck on actually thinking, token and tool usage is not going up at all).",
          "state": "open",
          "age_days": 18,
          "comments": 22,
          "reactions_total": 30,
          "upvotes": 30,
          "downvotes": 0,
          "created_at": "2026-02-17T02:32:05Z",
          "url": "https://github.com/anthropics/claude-code/issues/26224",
          "labels": [
            "bug"
          ],
          "cluster_id": 0
        }
      ]
    },
    {
      "id": 5,
      "issue_count": 258,
      "stats": {
        "total_reactions": 573,
        "total_comments": 1273,
        "avg_age_days": 13.7
      },
      "analysis": {
        "cluster_title": "Keyword Group: Extension Cowork Bun Vm Desktop",
        "summary": "This semantic cluster was defined by the high density of the following keywords: extension, cowork, bun, vm, desktop.",
        "hardest_problems": [
          "[BUG] Cowork shows \"Virtualization is not enabled\" on Windows 10 Pro despite Hyper-V fully enabled",
          "[BUG] Cowork: \"Failed to start Claude's workspace\" \u2014 VM service not running, persists after reboot",
          "[BUG] Bun v1.3.10 crashes with \"panic: incorrect alignment\" on Windows 11 ARM64 (v2.1.47)"
        ],
        "hallucination_check": "Local heuristic analysis based on TF-IDF term frequencies and engagement metrics."
      },
      "top_issues": [
        {
          "number": 27316,
          "title": "[BUG] Cowork shows \"Virtualization is not enabled\" on Windows 10 Pro despite Hyper-V fully enabled",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\nCowork tab shows \"Virtualization is not enabled. Claude's workspace requires \nhardware virtualization (Hyper-V). Enable virtualization in your computer's \nBIOS/UEFI settings, then restart.\" \u2014 despite all virtualization settings \nbeing fully enabled and confirmed working.\n\nSteps taken:\n- Enabled Intel Virtualization Technology in Dell BIOS (F2)\n- Enabled both Hyper-V sub-options in Windows Features\n- Enabled Virtual Machine Platform and Windows Hypervisor Platform\n- Ran: bcdedit /set hypervisorlaunchtype auto\n- Ran PowerShell commands to enable all three features\n- systeminfo confirms: \"A hypervisor has been detected\"\n- Multiple full restarts performed\n- Running latest Claude Desktop version on Windows 10 Pro (Dell PC)\n\nError persists unchanged after all of the above.\n\n### What Should Happen?\n\nCowork should detect that hardware virtualization is fully enabled at both \nthe BIOS and Windows level, successfully start the VM, and allow access to \nthe Cowork workspace without showing the \"Virtualization is not enabled\" error.\n\n### Error Messages/Logs\n\n```shell\nVirtualization is not enabled\nClaude's workspace requires hardware virtualization (Hyper-V). Enable virtualization in your computer's BIOS/UEFI settings, then restart.\n```\n\n### Steps to Reproduce\n\n1. Install latest Claude Desktop on Windows 10 Pro (Dell PC)\n2. Open Claude Desktop and navigate to the Cowork tab\n3. Observe \"Virtualization is not enabled\" error message\n\nAttempted fixes (error persists after all of these):\n4. Restart PC and enter BIOS via F2 (Dell)\n5. Navigate to Advanced \u2192 CPU Configuration\n6. Enable \"Intel Virtualization Technology\" and \"VT for Direct I/O\"\n7. Save with F10 and restart\n8. Press Windows + R \u2192 type optionalfeatures \u2192 enable:\n   - Hyper-V (both sub-options)\n   - Virtual Machine Platform\n   - Windows Hypervisor Platform\n9. Restart PC\n10. Open PowerShell as Administrator and run:\n    - Enable-WindowsOptionalFeature -Online -FeatureName Microsoft-Hyper-V -All\n    - Enable-WindowsOptionalFeature -Online -FeatureName VirtualMachinePlatform\n    - Enable-WindowsOptionalFeature -Online -FeatureName HypervisorPlatform\n    - bcdedit /set hypervisorlaunchtype auto\n11. Restart PC again\n12. Open Claude Desktop \u2192 Cowork tab\n13. Same \"Virtualization is not enabled\" error still displayed\n14. Confirmed via systeminfo: \"A hypervisor has been detected\"\n\n### Claude Model\n\nNone\n\n### Is this a regression?\n\nI don't know\n\n### Last Working Version\n\n_No response_\n\n### Claude Code Version\n\nClaude 1.1.3918 (a4b368) 2026-02-20T23:13:32.000Z\n\n### Platform\n\nOther\n\n### Operating System\n\nWindows\n\n### Terminal/Shell\n\nOther\n\n### Additional Information\n\n_No response_",
          "state": "closed",
          "age_days": 14,
          "comments": 78,
          "reactions_total": 30,
          "upvotes": 29,
          "downvotes": 0,
          "created_at": "2026-02-21T03:39:42Z",
          "url": "https://github.com/anthropics/claude-code/issues/27316",
          "labels": [
            "bug",
            "oncall"
          ],
          "cluster_id": 5
        },
        {
          "number": 27801,
          "title": "[BUG] Cowork: \"Failed to start Claude's workspace\" \u2014 VM service not running, persists after reboot",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\naWhen opening Cowork in Claude Desktop, the workspace fails to start with the following error:\n\n**\"Failed to start Claude's workspace \u2014 VM service not running. The service failed to start.\"**\n\nThe error persists even after rebooting the computer. The Cowork UI still loads (the prompt input and model selector are visible), but the workspace VM cannot start, making Cowork completely unusable.\n\nThis appears related to the ongoing MCP/Cowork regression introduced in Claude Desktop v1.1.3189 (build 2026-02-14), which has also broken mcp-registry and Claude in Chrome MCP connections (see issue #27492).\n\n### What Should Happen?\n\naThe Cowork workspace VM should start successfully when Cowork is opened. The service should be running reliably and not require workarounds to start.\n\n### Error Messages/Logs\n\n```shell\nError shown in Cowork UI:\n\"Failed to start Claude's workspace\nVM service not running. The service failed to start.\nRestarting Claude or your computer sometimes resolves this. If it persists, you can reinstall the workspace or share your debug logs to help us improve.\"\n\nNote: Error persists after full computer reboot.\n```\n\n### Steps to Reproduce\n\na1. Open Claude Desktop (latest version)\n2. Click on Cowork mode\n3. 3. Observe the error banner: \"Failed to start Claude's workspace \u2014 VM service not running. The service failed to start.\"\n4. 4. Reboot the computer\n5. 5. Re-open Claude Desktop and switch to Cowork\n6. 6. Error banner appears again \u2014 the issue persists after reboot\n\n### Claude Model\n\nOpus\n\n### Is this a regression?\n\nYes, this worked in a previous version\n\n### Last Working Version\n\n_No response_\n\n### Claude Code Version\n\nLatest (post-1.1.3189)\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nmacOS\n\n### Terminal/Shell\n\nTerminal.app (macOS)\n\n### Additional Information\n\nThis issue is related to the ongoing Cowork/MCP regression introduced in Claude Desktop v1.1.3189 (Feb 13-14, 2026). See also issue #27492 which documents mcp-registry and Claude in Chrome MCP connection failures in Cowork.\n\nThe error message shown in Cowork reads:\n- \"Failed to start Claude's workspace\"\n- - \"VM service not running. The service failed to start.\"\n- - - The UI suggests restarting or reinstalling the workspace, but the error persists after a full computer reboot.\nScreenshot of the error attached below.\n\n<img width=\"556\" height=\"287\" alt=\"Image\" src=\"https://github.com/user-attachments/assets/2f4d2b65-39f1-45c7-bc9f-63a5c0e2910b\" />",
          "state": "open",
          "age_days": 12,
          "comments": 28,
          "reactions_total": 20,
          "upvotes": 20,
          "downvotes": 0,
          "created_at": "2026-02-23T04:22:53Z",
          "url": "https://github.com/anthropics/claude-code/issues/27801",
          "labels": [
            "bug"
          ],
          "cluster_id": 5
        },
        {
          "number": 26723,
          "title": "[BUG] Bun v1.3.10 crashes with \"panic: incorrect alignment\" on Windows 11 ARM64 (v2.1.47)",
          "body": "\n\nThis issue will be fixed in the next version of Claude Code.\n\nVia https://github.com/oven-sh/bun/pull/27290\n\nOriginal issue:\n\n\n------\n\n### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\nClaude Code v2.1.47 crashes repeatedly with a Bun runtime panic on Windows 11 ARM64. The crash is 100% reproducible \u2014 it occurs after ~80\u2013150 seconds of runtime regardless of what command is being executed. Even trivial operations trigger it if the session has been running long enough.\n\nThe crash always produces the identical signature:\n- `panic(thread XXXXX): incorrect alignment`\n- RSS at crash: 0.64\u20130.78 GB (machine has 34 GB \u2014 not memory pressure)\n- Crash happened 5 consecutive times in one session\n\n### What Should Happen?\n\nClaude Code should run without crashing on Windows 11 ARM64. Previous versions worked fine on the same machine.\n\n### Error Messages/Logs\n\n```shell\nBun v1.3.10 (1423d3c8) Windows arm64\nWindows v.win11_dt\nCPU: neon fp aes crc32 atomics\nArgs: \"C:\\Users\\...\\claude.exe\"\nFeatures: Bun.stderr(2) Bun.stdin(2) Bun.stdout(2) fetch(91) jsc spawn(64) standalone_executable process_dlopen(3) yaml_parse(4)\nElapsed: 246271ms | User: 19218ms | Sys: 4750ms\nRSS: 0.72GB | Peak: 0.94GB | Commit: 0.99GB | Faults: 1090072 | Machine: 33.90GB\n\npanic(thread 5184): incorrect alignment\noh no: Bun has crashed. This indicates a bug in Bun, not your code.\n\nhttps://bun.report/1.3.10/W_11423d3cmgkgEugogCwmzrfwl0uiBgyvBw32CwmvBo7xB44wCwvhgmCw236lC4i16lCg6xjkBgsjufCYKERNEL32.DLLg0jCCSntdll.dlloqi1BA0eNrLzEvOLypKTS5RSMzJTM/LTc0rAQBMcwep\n```\n\n### Steps to Reproduce\n\n1. Install Claude Code v2.1.47 on Windows 11 ARM64\n2. Run `claude` in a terminal\n3. Execute any command (bash, file read, or even just wait idle)\n4. After ~80\u2013150 seconds of accumulated runtime, Bun panics with \"incorrect alignment\"\n5. Crash is 100% reproducible \u2014 happened 5 consecutive times\n\n### Claude Model\n\nOpus\n\n### Is this a regression?\n\nYes, this worked in a previous version\n\n### Last Working Version\n\nUnknown \u2014 previous version before 2.1.47\n\n### Claude Code Version\n\n2.1.47 (Claude Code)\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nWindows\n\n### Terminal/Shell\n\nWindows Terminal\n\n### Additional Information\n\nBun crash report URL (identical every time):\nhttps://bun.report/1.3.10/W_11423d3cmgkgEugogCwmzrfwl0uiBgyvBw32CwmvBo7xB44wCwvhgmCw236lC4i16lCg6xjkBgsjufCYKERNEL32.DLLg0jCCSntdll.dlloqi1BA0eNrLzEvOLypKTS5RSMzJTM/LTc0rAQBMcwep\n\nThis started recently with the current version \u2014 same machine previously worked without crashes. Likely a regression in the Bun build shipped with Claude Code.",
          "state": "open",
          "age_days": 16,
          "comments": 27,
          "reactions_total": 19,
          "upvotes": 19,
          "downvotes": 0,
          "created_at": "2026-02-18T23:19:19Z",
          "url": "https://github.com/anthropics/claude-code/issues/26723",
          "labels": [
            "bug"
          ],
          "cluster_id": 5
        },
        {
          "number": 26638,
          "title": "[Bug] Claude Code desktop application crashes with unresponsive processes after startup",
          "body": "**Bug Description**\n1YCZHOG claude code on desktop opens and then shows this error code after a few seconds.  once the error shows I can only kill the claude.exe(s) (8 of them) and then restart.  been going on since evening of 17/02/2026\n\n**Environment Info**\n- Platform: win32\n- Terminal: null\n- Version: 2.1.37\n- Feedback ID: 4806c090-f876-4565-bd16-465f5beb9399\n\n**Errors**\n```json\n[{\"error\":\"AbortError: The operation was aborted.\\n    at unknown\\n    at abortChildProcess (node:child_process:947:42)\\n    at onAbortListener2 (node:child_process:35:24)\\n    at abort (unknown)\\n    at <anonymous> (B:/~BUN/root/claude.exe:3458:209)\",\"timestamp\":\"2026-02-18T16:30:11.276Z\"}]\n```\ntried all standard troubleshooting and reinstall, still the same issue.",
          "state": "closed",
          "age_days": 16,
          "comments": 15,
          "reactions_total": 31,
          "upvotes": 31,
          "downvotes": 0,
          "created_at": "2026-02-18T16:54:13Z",
          "url": "https://github.com/anthropics/claude-code/issues/26638",
          "labels": [],
          "cluster_id": 5
        },
        {
          "number": 28333,
          "title": "[Bug] Bash tool execution failing or non-functional",
          "body": "**Bug Description**\nBash tool completely broken\n\n Bash(git status)\n  \u23bf \u00a0Error: EINVAL: invalid argument, open 'C:\\Users\\me\\AppData\\Local\\Temp\\claude\\C--Users-me-Downloads-Prj1\\tasks\\bcyfkqe72.output'\n\nSame for any command.\n\n**Environment Info**\n- Platform: win32\n- Terminal: windows-terminal\n- Version: 2.1.53\n- Feedback ID: 1d914615-c2d4-4f38-aa95-de4b4ecc537c\n\n**Errors**\n```json\n[{\"error\":\"AbortError: The operation was aborted.\\n    at unknown\\n    at abortChildProcess (node:child_process:947:42)\\n    at onAbortListener2 (node:child_process:35:24)\\n    at abort (unknown)\\n    at <anonymous> (B:/~BUN/root/claude.exe:6006:209)\",\"timestamp\":\"2026-02-25T00:25:58.572Z\"},{\"error\":\"RipgrepTimeoutError: Ripgrep search timed out after 20 seconds. The search may have matched files but did not complete in time. Try searching a more specific path or pattern.\\n    at f (B:/~BUN/root/claude.exe:116:189)\\n    at <anonymous> (B:/~BUN/root/claude.exe:116:403)\\n    at exitHandler (node:child_process:109:27)\\n    at errorHandler (node:child_process:118:16)\\n    at emitError (node:events:43:23)\\n    at abortChildProcess (node:child_process:947:17)\\n    at onAbortListener2 (node:child_process:35:24)\\n    at abort (unknown)\\n    at <anonymous> (B:/~BUN/root/claude.exe:6006:209)\",\"timestamp\":\"2026-02-25T00:25:58.573Z\"},{\"error\":\"Error: EINVAL: invalid argument, open 'C:\\\\Users\\\\mpotanin\\\\AppData\\\\Local\\\\Temp\\\\claude\\\\C--Users-mpotanin-Downloads-Reservoir-UI\\\\tasks\\\\bxwcm3ieu.output'\\n    at openSync (unknown)\\n    at $3$ (B:/~BUN/root/claude.exe:1756:2307)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T00:26:24.715Z\"},{\"error\":\"Error: EINVAL: invalid argument, open 'C:\\\\Users\\\\mpotanin\\\\AppData\\\\Local\\\\Temp\\\\claude\\\\C--Users-mpotanin-Downloads-Reservoir-UI\\\\tasks\\\\bcyfkqe72.output'\\n    at openSync (unknown)\\n    at $3$ (B:/~BUN/root/claude.exe:1756:2307)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T00:26:27.273Z\"},{\"error\":\"Error: Request was aborted.\\n    at makeRequest (B:/~BUN/root/claude.exe:326:3940)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T00:26:29.089Z\"}]\n```\n",
          "state": "closed",
          "age_days": 10,
          "comments": 15,
          "reactions_total": 22,
          "upvotes": 22,
          "downvotes": 0,
          "created_at": "2026-02-25T00:28:11Z",
          "url": "https://github.com/anthropics/claude-code/issues/28333",
          "labels": [
            "bug",
            "platform:windows",
            "area:bash"
          ],
          "cluster_id": 5
        }
      ]
    },
    {
      "id": 4,
      "issue_count": 56,
      "stats": {
        "total_reactions": 477,
        "total_comments": 388,
        "avg_age_days": 8.5
      },
      "analysis": {
        "cluster_title": "Keyword Group: Remote Control Remote Control Session Max",
        "summary": "This semantic cluster was defined by the high density of the following keywords: remote, control, remote control, session, max.",
        "hardest_problems": [
          "[BUG] `claude remote-control` shows misleading \"Contact your administrator\" error for personal (Pro/Max) accounts",
          "[Bug] Remote Control unavailable despite Pro plan authentication",
          "[BUG] \"Remote Control environments are not available for your account.\" for Claude Code Pro Plan user"
        ],
        "hallucination_check": "Local heuristic analysis based on TF-IDF term frequencies and engagement metrics."
      },
      "top_issues": [
        {
          "number": 28098,
          "title": "[BUG] `claude remote-control` shows misleading \"Contact your administrator\" error for personal (Pro/Max) accounts",
          "body": "### Bug Description\n\n`claude remote-control` subcommand (added in v2.1.51) fails with the error:\n\n```\nError: Remote Control is not enabled for your account. Contact your administrator.\n```\n\nThis error message is misleading for **personal account users (Pro/Max)** \u2014 there is no \"administrator\" to contact. The message appears to be written for Enterprise/Teams accounts only.\n\n### Expected Behavior\n\nOne of:\n1. The feature should work for personal accounts as the changelog states \"enabling local environment serving for **all users**\"\n2. If the feature is plan-restricted, the error should clearly state which plan is required (e.g., \"Remote Control requires an Enterprise plan\")\n3. For personal accounts, \"Contact your administrator\" should not appear \u2014 instead provide actionable guidance\n\n### Steps to Reproduce\n\n1. Use a personal (Pro or Max) account\n2. Install Claude Code v2.1.51\n3. Run `claude remote-control`\n4. Observe the error message\n\n### Error Output\n\n```\n$ claude remote-control\nError: Remote Control is not enabled for your account. Contact your administrator.\n```\n\n### Questions\n\n1. Is `remote-control` intended for all users as the changelog states, or is it plan-restricted?\n2. If plan-restricted, which plans are supported?\n3. How do personal account users enable this feature?\n\n### Environment\n\n- Claude Code version: 2.1.51\n- OS: macOS 15.5 (arm64)\n- Account type: Personal\n- Installation: Native\n\n### Related\n\n- #28038 \u2014 CLI reference missing `claude remote-control` documentation\n- Changelog entry: \"Added `claude remote-control` subcommand for external builds, enabling local environment serving for all users.\"",
          "state": "open",
          "age_days": 11,
          "comments": 73,
          "reactions_total": 79,
          "upvotes": 77,
          "downvotes": 0,
          "created_at": "2026-02-24T05:08:39Z",
          "url": "https://github.com/anthropics/claude-code/issues/28098",
          "labels": [
            "bug",
            "platform:macos",
            "area:auth",
            "oncall",
            "area:cli"
          ],
          "cluster_id": 4
        },
        {
          "number": 28817,
          "title": "[Bug] Remote Control unavailable despite Pro plan authentication",
          "body": "**Bug Description**\nIssue: Remote Control not available despite being on Pro plan\n\nError Message:\n\nError: Remote Control is not yet available on your plan.\nSteps Attempted:\n\nRan claude auth logout then claude auth login to re-authenticate\nUsed /login command to verify claude.ai account authentication\nUpdated Claude Code to the latest version\nEnvironment:\n\nPlan: Pro\nAuthentication: claude.ai account (not API key)\nExpected Behavior: Remote Control should be available for Pro plan users according to documentation.\n\n**Environment Info**\n- Platform: darwin\n- Terminal: iTerm.app\n- Version: 2.1.59\n- Feedback ID: 8f350ff8-a614-4350-abd2-2222b5ee12f9\n\n**Errors**\n```json\nError: Remote Control is not yet available on your plan.\n```\n",
          "state": "open",
          "age_days": 9,
          "comments": 37,
          "reactions_total": 59,
          "upvotes": 58,
          "downvotes": 0,
          "created_at": "2026-02-26T02:39:40Z",
          "url": "https://github.com/anthropics/claude-code/issues/28817",
          "labels": [
            "bug",
            "duplicate",
            "area:docs"
          ],
          "cluster_id": 4
        },
        {
          "number": 29449,
          "title": "[BUG] \"Remote Control environments are not available for your account.\" for Claude Code Pro Plan user",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\nAfter answering `y` to enable Remote Control, the CLI immediately returns `Remote Control environments are not available for your account.` All subsequent runs skip the prompt and repeat the same message, leaving the user with no way to retry or understand why.\n\n### What Should Happen?\n\nRemote Control enables successfully\n\n\n### Steps to Reproduce\n\n\n1. Open VS Code integrated terminal on macOS.\n2. Run `claude remote-control`.\n3. The CLI displays the Remote Control prompt:\n```\n   Remote Control lets you access this CLI session from the web (claude.ai/code)\n   or the Claude app, so you can pick up where you left off on any device.\n   You can disconnect remote access anytime by running /remote-control again.\n   Enable Remote Control? (y/n)\n```\n\n4. Type `y` and press Enter.\n5. The CLI responds: `Remote Control environments are not available for your account.`\n\n**What happens next (secondary run)**\n\n6. Run `claude remote-control` again in the same or any new terminal session.\n7. The prompt is skipped entirely \u2014 the CLI immediately prints: `Remote Control environments are not available for your account.`\n\nNo y/n prompt is shown.\n\n### Claude Model\n\nOther\n\n### Is this a regression?\n\nNo, this never worked\n\n### Claude Code Version\n\n2.1.62 (Claude Code)\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nmacOS\n\n### Terminal/Shell\n\nVS Code integrated terminal\n\n### Additional Information\n\nAfter answering `y` to enable Remote Control, the CLI returns `Remote Control environments are not available for your account.`\n`claude auth status` confirms I am a Pro user, `claude --version` confirms I am up to date.\nI have logged out and logged back in several times via `/logout` and `/login` and re-authenticated.\n\nRemote Control has been buggy for many Pro/Max users (Related GitHub issues: #28816, #28884, #29037, #29164, #29185, #29430), but `Remote Control environments are not available for your account.` after being prompted to answer (`y/n`) is a distinct message I cannot find in any of those reports, so this does not necessarily qualify as a duplicate.",
          "state": "open",
          "age_days": 7,
          "comments": 20,
          "reactions_total": 22,
          "upvotes": 22,
          "downvotes": 0,
          "created_at": "2026-02-27T23:38:54Z",
          "url": "https://github.com/anthropics/claude-code/issues/29449",
          "labels": [
            "bug",
            "duplicate",
            "platform:macos",
            "area:auth",
            "platform:vscode",
            "area:cli"
          ],
          "cluster_id": 4
        },
        {
          "number": 28951,
          "title": "[BUG] Remote control (/rc) not supported in VS Code extension",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\nThe Claude Code VS Code extension does not support the /rc (remote control) feature. When remote control is enabled by default in the CLI configuration, it is ignored when the CLI is invoked through the VS Code extension. The extension only exposes a subset of slash commands and /rc is not among them.\n\n### What Should Happen?\n\nThe VS Code extension should support remote control sessions, either via the /rc slash command or by respecting the CLI's default remote control configuration. Since the extension already uses the CLI under the hood, remote control functionality should be passed through to the extension.\n\n### Error Messages/Logs\n\n```shell\n\n```\n\n### Steps to Reproduce\n\n1. Enable remote control by default in Claude Code CLI configuration\n2. Open a terminal and run claude \u2014 remote control session is created as expected\n3. Open the Claude Code VS Code extension panel\n4. Note that no remote control session is created\n5. Try typing /rc in the extension \u2014 command is not available\n\n### Claude Model\n\nOpus\n\n### Is this a regression?\n\nNo, this never worked\n\n### Last Working Version\n\n_No response_\n\n### Claude Code Version\n\n2.1.59 (Claude Code)\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nmacOS\n\n### Terminal/Shell\n\nCursor\n\n### Additional Information\n\n_No response_",
          "state": "open",
          "age_days": 8,
          "comments": 27,
          "reactions_total": 14,
          "upvotes": 14,
          "downvotes": 0,
          "created_at": "2026-02-26T11:03:35Z",
          "url": "https://github.com/anthropics/claude-code/issues/28951",
          "labels": [
            "bug",
            "platform:macos",
            "area:ide",
            "platform:vscode"
          ],
          "cluster_id": 4
        },
        {
          "number": 29214,
          "title": "Remote Control: mobile app shows permission prompts despite --dangerously-skip-permissions",
          "body": "## Bug Report\n\n### Description\nWhen starting a Claude Code session with `--dangerously-skip-permissions` and enabling Remote Control (`/rc`), the Claude mobile app still shows permission prompts for every file edit and bash command. The mobile interface should inherit the permission mode of the local session.\n\n### Steps to Reproduce\n1. Start Claude Code with `~/.local/bin/claude --dangerously-skip-permissions`\n2. Enable Remote Control via `/rc`\n3. Connect to the session from the Claude mobile app\n4. Ask Claude to make a file change or run a command\n\n### Expected Behavior\nNo permission prompts \u2014 the mobile interface should mirror the local session's permission mode (`--dangerously-skip-permissions`).\n\n### Actual Behavior\nMobile app displays a permission prompt for every change, requiring manual approval each time.\n\n### Environment\n- Platform: WSL Ubuntu-24.04\n- Claude Code: latest\n- Mobile app: Claude iOS/Android\n- Plan: Max\n\n### Additional Context\nRemote Control is supposed to mirror the local session exactly. Permission mode is a local session property and should not be overridden by the remote interface.",
          "state": "open",
          "age_days": 8,
          "comments": 10,
          "reactions_total": 19,
          "upvotes": 18,
          "downvotes": 0,
          "created_at": "2026-02-27T04:20:09Z",
          "url": "https://github.com/anthropics/claude-code/issues/29214",
          "labels": [
            "bug",
            "platform:wsl",
            "area:permissions"
          ],
          "cluster_id": 4
        }
      ]
    },
    {
      "id": 7,
      "issue_count": 114,
      "stats": {
        "total_reactions": 342,
        "total_comments": 320,
        "avg_age_days": 15.4
      },
      "analysis": {
        "cluster_title": "Keyword Group: Feature Context _No _No Response_ Response_",
        "summary": "This semantic cluster was defined by the high density of the following keywords: feature, context, _no, _no response_, response_.",
        "hardest_problems": [
          "Migrate Claude Cowork from Electron to Electrobun",
          "[FEATURE] Support multiple Connector accounts (same connector, different accounts) in Claude and Claude Code on the web (claude.ai/code)",
          "[FEATURE] Max plan usage limits silently reduced since Claude 4.6 release"
        ],
        "hallucination_check": "Local heuristic analysis based on TF-IDF term frequencies and engagement metrics."
      },
      "top_issues": [
        {
          "number": 26648,
          "title": "Migrate Claude Cowork from Electron to Electrobun",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single feature request (not multiple features)\n\n### Problem Statement\n\n## Today with Electron\n\nI downloaded Claude Cowork Desktop today. \n\n- The DMG is 241.6MB\n- Once I put it in the /Applications folder the app is 574.9MB\n- If you're using differential updates, Electron's uses bucket hashes such that a single character change in the beginning of a file affects the bucket position and hashes of the rest of the file, so all your users have to redownload the whole 250MB.\n- Electron's `<webview>` tag [has been deprecated for a year](https://blackboard.sh/blog/building-a-better-oopif/) \n- You're using node.js which is slow when you just acquired Bun\n- You have to wait for Electron to update Chromium even with urgent security issues\n\n## Tomorrow With Electrobun\n- Your DMG could be as small as 14MB, most of that is the bun runtime using electrobun's Zstd bundles\n- Built-in differential updates using a custom optimized zig-bsdiff that generates patch files as small as 4KB so you can ship as often as the web without you or your users paying the bandwidth tax\n- You can choose to use the system webview or bundleCEF and change the CEF version whenever you want. In the future ladybird and Servo will be drop in alternatives when they're ready.\n- Our `<electrobun-webview>` tag is re-implemented from the ground up and webview agnostic\n- When bundling CEF your app can open both Webkit and Chromium webviews and OOPIFs side-by-side\n- Our OOPIFs are actually isolated, no dangerous node integration, with encrypted-per-webview, fully typed RPC that is **fast**\n- The Electrobun cli handles bundling, code signing, and notarization for you, just provide the ENV vars.\n- Electrobun has built in Update api and generates artifacts that can be put on any static file host (R2, S3, Github Releases)\n- Every feature you ship to Bun's runtime, you get in Electrobun.\n\n## Why now\n- Electrobun just passed v1 (stable architecture) a week ago\n- Developers are already shipping production apps with Electrobun\n- It is gaining over 1,000 github stars per day\n- It is a next-gen desktop app framework\n- developers just write typescript for both the main and browser processes so you can move fast and share code.\n\n## The Next Steps\nRight now, you are shipping a heavy, slow Node.js/Electron environment to your users literally weeks after acquiring the fastest JavaScript runtime on the planet.\n\nI\u2019d love to help the Anthropic team fix this. I see two paths forward:\n\n1. The Migration: I am happy to prioritize whatever specific APIs the Claude Desktop team needs in Electrobun over the next sprint to make a 14MB footprint a reality.\n2. The Status Quo: You stick with Electron, and keep paying the massive bandwidth tax and UX penalty of shipping Chromium and Node when you already own Bun.\n\n\n## References\n### Claude now:\n<img width=\"283\" height=\"563\" alt=\"Image\" src=\"https://github.com/user-attachments/assets/77125d25-d499-4017-913d-cbf0d98d6793\" />\n\n### Electrobun v1: \n\nhttps://x.com/YoavCodes/status/2020348890525155621?s=20\nhttps://github.com/blackboardsh/electrobun\n\n<img width=\"1176\" height=\"741\" alt=\"Image\" src=\"https://github.com/user-attachments/assets/e6d90427-0200-47f6-9a62-c9108cdccb68\" />\n\n### Video Demos\n\n- Co(lab): a hybrid web browser, code editor, and PTY terminal for deep work. https://youtu.be/WWTCqGmE86w \n- Audio TTS: a vibe coded app I built with Claude Code in Colab using Electrobun, the audio for both videos generated using it. https://youtu.be/Z4dNK1d6l6E\n\n\n\n\n\n### Proposed Solution\n\nMigrate to Electrobun.\n\n### Alternative Solutions\n\n_No response_\n\n### Priority\n\nCritical - Blocking my work\n\n### Feature Category\n\nPerformance and speed\n\n### Use Case Example\n\n_No response_\n\n### Additional Context\n\n_No response_",
          "state": "closed",
          "age_days": 16,
          "comments": 13,
          "reactions_total": 94,
          "upvotes": 60,
          "downvotes": 0,
          "created_at": "2026-02-18T18:19:23Z",
          "url": "https://github.com/anthropics/claude-code/issues/26648",
          "labels": [
            "enhancement"
          ],
          "cluster_id": 7
        },
        {
          "number": 27302,
          "title": "[FEATURE] Support multiple Connector accounts (same connector, different accounts) in Claude and Claude Code on the web (claude.ai/code)",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single feature request (not multiple features)\n\n### Problem Statement\n\n**Note:** This is just a duplicate of Issue [#15581](https://github.com/anthropics/claude-code/issues/15581) which was closed due to inactivity. The github-actions bot suggests we open a new issue if this is still relevant.\n\nAs mentioned in another comment, this a more general issue:\n\n> This applies to all connectors, not just GitHub. Can't have multiple Gmail connected at the same time, which forces separate subscriptions. Profile-scoped connectors would solve it.\n\nCurrently, Claude Code on the web (claude.ai/code) only allows connecting a single GitHub account through the Connectors settings (Settings \u2192 Connectors \u2192 GitHub).\n\nAs a developer, I maintain multiple GitHub accounts:\n\n* Personal account for side projects\n* * Work/organization account for professional work\n    When I connect my personal GitHub account, I cannot access repositories from my work account, and vice versa. The only workaround is to disconnect and reconnect with a different GitHub account each time, which is cumbersome and disrupts workflow.\n\n**Note:** Issue [#261](https://github.com/anthropics/claude-code/issues/261) addressed multiple accounts for the CLI version using `CLAUDE_CONFIG_DIR` environment variable, but this solution doesn't apply to the web interface (claude.ai/code).\n\n### Proposed Solution\n\nAllow users to connect multiple GitHub accounts in Claude.ai Settings \u2192 Connectors, similar to how other development tools (VS Code, JetBrains IDEs) handle multiple accounts.\n\n**Ideal implementation:**\n\n1. Add ability to connect additional GitHub accounts in Connectors settings\n2. Show repositories from all connected accounts in the repository selector dropdown\n3. Clearly indicate which account each repository belongs to in the UI (e.g., \"username/repo-name\")\n4. Allow users to set a default account or choose per-session\n\n### Alternative Solutions\n\n_No response_\n\n### Priority\n\nMedium - Would be very helpful\n\n### Feature Category\n\nMCP server integration\n\n### Use Case Example\n\n_No response_\n\n### Additional Context\n\n_No response_",
          "state": "open",
          "age_days": 14,
          "comments": 23,
          "reactions_total": 37,
          "upvotes": 22,
          "downvotes": 0,
          "created_at": "2026-02-21T02:14:14Z",
          "url": "https://github.com/anthropics/claude-code/issues/27302",
          "labels": [
            "enhancement"
          ],
          "cluster_id": 7
        },
        {
          "number": 28848,
          "title": "[FEATURE] Max plan usage limits silently reduced since Claude 4.6 release",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single feature request (not multiple features)\n\n### Problem Statement\n\n## Problem\n\nSince the Claude 4.6 release, I've been burning through my weekly usage cap dramatically faster on the Max plan ($200/month). This is not increased token consumption on my end \u2014 I audited my entire tooling stack looking for rogue API calls or background token drain. There are none. The caps themselves were lowered without a word.\n\nThere is zero transparency around this:\n\n1. **The actual limits are never published** \u2014 just a vague percentage with no denominator\n2. **No notification when they change** \u2014 you find out by slamming into a wall mid-week\n3. **No changelog, no announcement, nothing** \u2014 paying customers are left guessing\n\n## Impact\n\nI'm paying $200/month. That's not a casual subscription. Silently pulling back what that buys is not okay.\n\n- I wasted hours investigating whether my own tooling was causing excess usage \u2014 it wasn't. It was a quiet cap reduction.\n- I can't scope my work for the week because I don't know what \"100%\" means anymore\n- I can't trust that the workflow I rely on today will still be viable next week\n- Every major model release apparently comes with a stealth nerf to what Max subscribers get\n\n\n### Proposed Solution\n\n- **Publish concrete rate limits** for each plan tier. Stop hiding behind percentages.\n- **Notify users when limits change** \u2014 email, dashboard banner, anything. Silence is disrespectful.\n- **Show actual usage numbers, not just percentages** \u2014 \"75% of ?\" is useless\n- **Changelog for limit adjustments** \u2014 especially after model releases when costs change\n- **If you need to reduce limits, say so upfront** \u2014 don't let paying customers discover it by getting cut off\n\n### Alternative Solutions\n\n_No response_\n\n### Priority\n\nHigh - Significant impact on productivity\n\n### Feature Category\n\nDocumentation\n\n### Use Case Example\n\n_No response_\n\n### Additional Context\n\n_No response_",
          "state": "open",
          "age_days": 9,
          "comments": 24,
          "reactions_total": 19,
          "upvotes": 19,
          "downvotes": 0,
          "created_at": "2026-02-26T04:16:06Z",
          "url": "https://github.com/anthropics/claude-code/issues/28848",
          "labels": [
            "invalid"
          ],
          "cluster_id": 7
        },
        {
          "number": 29200,
          "title": "[FEATURE]  Allow customizing the animated thinking/status words",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single feature request (not multiple features)\n\n### Problem Statement\n\n\nClaude Code shows \"fun\" animated status words while thinking (e.g., \"Cogitating...\", \"Discombobulating...\", \"Flibbertigibbeting...\"). It would be great to let users customize this word list.\n\n\nWhy:\n\n1. It's a low-risk, fun personalization option\n2. The word list is currently hardcoded in the compiled binary with no way to override it\n3. Users who spend a lot of time in the CLI would enjoy making it their own\n4. Sometimes I would rather not _flibbertigibbet_ and would rather Claude didn't either.\n5. *@#$!\n\n### Proposed Solution\n\nA config file like ~/.claude/thinking-words.json or a key in ~/.claude/settings.json:\n\n```json\n{\n  \"thinkingWords\": [\n    \"Cussing\",\n    \"Swearing\",\n    \"Drinking\",\n    \"Fuming\",\n    \"Into the void staring\",\n    \"Career changing\",\n    \"Goat farming\"\n  ]\n}\n```\nIf present, use the custom list instead of the built-in defaults. If absent, fall back to the current words.\n\n### Alternative Solutions\n\nI tried monkey patching but even if it were javascript rather than binary it still would just be overwritten next update.\n\n### Priority\n\nLow - Nice to have\n\n### Feature Category\n\nOther\n\n### Use Case Example\n\n1. I am writing code and / or interacting with claude\n2. It keeps talking about how it is discombobulating\n3. I would like to have it use less cutesy words\n4. The don't have to be swears but sometimes... it would be nice\n\n### Additional Context\n\n_No response_",
          "state": "open",
          "age_days": 8,
          "comments": 6,
          "reactions_total": 18,
          "upvotes": 10,
          "downvotes": 0,
          "created_at": "2026-02-27T02:27:29Z",
          "url": "https://github.com/anthropics/claude-code/issues/29200",
          "labels": [
            "enhancement",
            "area:tui"
          ],
          "cluster_id": 7
        },
        {
          "number": 27263,
          "title": "[FEATURE] Preview: Allow configurable external URL whitelist for OAuth and other third-party required flows",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single feature request (not multiple features)\n\n### Problem Statement\n\nThe new App Preview feature in Claude Code on desktop restricts all navigation to localhost URLs only. When a locally running web application initiates an OAuth flow that requires navigating to a third-party identity provider (e.g., `idbroker.webex.com`, `accounts.google.com`, `login.microsoftonline.com`), the Preview browser blocks the outbound navigation with the message:\n\n> \u26a0\ufe0f Link to webexapis.com was blocked. Preview only supports localhost URLs.\n\nThis makes it impossible to develop or test any application that relies on third-party OAuth/SSO authentication flows within the Preview panel.\n\n### Proposed Solution\n\nAdd a configurable **external URL allowlist** for the Preview browser. This could be specified at the project level (e.g., in `.claude/settings.local.json` or a dedicated preview config) or via the desktop app settings.\n\n### Alternative Solutions\n\n| Alternative | Why It's Not Ideal |\n| --- | --- |\n| **Use Claude in Chrome (`--chrome`)** | Loses the inline Preview experience; requires switching to an external browser window. Defeats the purpose of the integrated Preview panel. |\n| **Dev-only login bypass route** | Requires code changes to the application under development; doesn't test the actual auth flow; not always feasible with third-party SDKs. |\n| **Open OAuth URL in external browser** | Fragile; callback URL may not route back correctly; breaks the seamless Preview workflow. |\n\n\n### Priority\n\nMedium - Would be very helpful\n\n### Feature Category\n\nConfiguration and settings\n\n### Use Case Example\n\n1. I'm working on a SaaS app that uses OAuth login as a core flow (like login with Google)\n2. I open settings and add the Google OAuth URLs to the Preview URL whitelist\n3. I login to my application within App Preview window without external URL errors\n\n### Additional Context\n\n_No response_",
          "state": "open",
          "age_days": 14,
          "comments": 1,
          "reactions_total": 23,
          "upvotes": 23,
          "downvotes": 0,
          "created_at": "2026-02-20T22:57:32Z",
          "url": "https://github.com/anthropics/claude-code/issues/27263",
          "labels": [
            "enhancement"
          ],
          "cluster_id": 7
        }
      ]
    },
    {
      "id": 3,
      "issue_count": 84,
      "stats": {
        "total_reactions": 235,
        "total_comments": 305,
        "avg_age_days": 14.7
      },
      "analysis": {
        "cluster_title": "Keyword Group: Mcp Server Tool Tools Json",
        "summary": "This semantic cluster was defined by the high density of the following keywords: mcp, server, tool, tools, json.",
        "hardest_problems": [
          "Claude Desktop: MCP servers fail with Protocol instance reuse error",
          "[BUG] Windows MSIX: \"Edit Config\" opens wrong claude_desktop_config.json \u2014 MCP servers silently fail to load",
          "[BUG] Notion MCP connector serializes JSON object parameters as strings, breaking update/move/create operations"
        ],
        "hallucination_check": "Local heuristic analysis based on TF-IDF term frequencies and engagement metrics."
      },
      "top_issues": [
        {
          "number": 27390,
          "title": "Claude Desktop: MCP servers fail with Protocol instance reuse error",
          "body": "## Bug Description\n\nClaude Desktop v1.1.3918 on macOS shows error toasts on every launch:\n\n- \"Could not connect to MCP server mcp-registry\"\n- \"Could not connect to MCP server Claude in Chrome\"\n\n## Root Cause\n\nThe actual error from `claude.ai-web.log` is identical for both servers:\n\n```\n[MCP] Could not connect to MCP server mcp-registry Error: Error invoking remote method 'connect-to-mcp-server': Error: Already connected to a transport. Call close() before connecting to a new transport, or use a separate Protocol instance per connection.\n```\n\n```\n[MCP] Could not connect to MCP server Claude in Chrome Error: Error invoking remote method 'connect-to-mcp-server': Error: Already connected to a transport. Call close() before connecting to a new transport, or use a separate Protocol instance per connection.\n```\n\nThe app connects three MCP servers on startup (Filesystem, mcp-registry, Claude in Chrome). The Filesystem extension connects successfully, but the other two fail because the Protocol instance is being reused instead of creating separate instances per connection.\n\nThis is a race condition \u2014 all three connection requests fire at the same timestamp, and the shared Protocol instance can only handle one.\n\n## Steps to Reproduce\n\n1. Install Claude Desktop v1.1.3918 on macOS\n2. Have the Filesystem extension installed (from the extension registry)\n3. Launch the app\n4. Observe error toasts for mcp-registry and Claude in Chrome\n\n## Environment\n\n- **App version**: 1.1.3918 (commit a4b368d308b05c425b74c1c19ed47a572d3a90f6)\n- **Platform**: macOS (darwin arm64, Apple M2)\n- **OS**: Darwin 25.2.0\n\n## Notes\n\n- The errors are **not** caused by misconfiguration \u2014 `mcpServers` in `claude_desktop_config.json` is empty. Both mcp-registry and Claude in Chrome are built-in servers.\n- Disabling the Chrome extension in settings and removing the `chromeExtension` pairing data from config does not fix it.\n- The Filesystem MCP server works correctly despite the errors.\n- The errors reproduce on every single launch and on every new conversation.",
          "state": "closed",
          "age_days": 13,
          "comments": 22,
          "reactions_total": 59,
          "upvotes": 59,
          "downvotes": 0,
          "created_at": "2026-02-21T11:20:00Z",
          "url": "https://github.com/anthropics/claude-code/issues/27390",
          "labels": [
            "oncall"
          ],
          "cluster_id": 3
        },
        {
          "number": 26073,
          "title": "[BUG] Windows MSIX: \"Edit Config\" opens wrong claude_desktop_config.json \u2014 MCP servers silently fail to load",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\n## Summary\n\nOn Windows, Claude Desktop installed via the official installer (claude.ai/download) uses MSIX packaging. The app's **\"Edit Config\"** button in Developer settings opens a config file at `%APPDATA%\\Claude\\claude_desktop_config.json`, but the app actually reads from a **different location** inside the MSIX virtualized filesystem. This causes MCP server configurations to be silently ignored with no error messages, no logs, and no indication that anything is wrong.\n\n## Environment\n\n- **OS:** Windows 11 Pro (latest version, Build 26100)\n- **Claude Desktop version:** 1.1.3189.0 (MSIX package: `Claude_pzs8sxrjxfjjc`)\n- **Installation method:** Downloaded from claude.ai/download (standard installer)\n- **Hyper-V:** Enabled\n- **Intel Virtualization (VT-x):** Enabled\n- **Node.js:** v24.13.1 (installed via vfox version manager)\n- **npm/npx:** v11.8.0\n\n## Installation Details\n\nThe app installs as an MSIX package at:\n\n```\nC:\\Program Files\\WindowsApps\\Claude_1.1.3189.0_x64__pzs8sxrjxfjjc\\app\\Claude.exe\n```\n\nNote: A second package also exists at `Anthropic.ClaudeDesktop_h6f0761` which may be related to the confusion.\n\n## Root Cause\n\nThis is an **Electron + MSIX filesystem virtualization** issue.\n\nMSIX apps on Windows run inside a lightweight container with a virtualized filesystem. When the Claude Desktop app (running inside the MSIX container) accesses `%APPDATA%\\Claude\\`, Windows silently **redirects** that read/write to the virtualized path:\n\n```\n%LOCALAPPDATA%\\Packages\\Claude_pzs8sxrjxfjjc\\LocalCache\\Roaming\\Claude\\\n```\n\nHowever, the \"Edit Config\" button likely uses an Electron API like `shell.openPath()` or `shell.openExternal()` to launch the user's text editor. This API call resolves to the **real, non-virtualized** `%APPDATA%\\Claude\\` path \u2014 bypassing the MSIX redirection.\n\nThe result: the app reads from the virtualized path, but tells the user to edit a completely different file at the real path. These are two separate files on disk that are never synchronized.\n\n## Bug Description\n\n### The Problem\n\nThere are **two** `claude_desktop_config.json` files on the system:\n\n1. **The file \"Edit Config\" opens** (the wrong one):\n\n   ```\n   C:\\Users\\<username>\\AppData\\Roaming\\Claude\\claude_desktop_config.json\n   ```\n\n   \u2192 This is the standard `%APPDATA%\\Claude\\` path referenced in all official documentation.\n\n2. **The file the app actually reads** (the correct one):\n\n   ```\n   C:\\Users\\<username>\\AppData\\Local\\Packages\\Claude_pzs8sxrjxfjjc\\LocalCache\\Roaming\\Claude\\claude_desktop_config.json\n   ```\n\n   \u2192 This is the MSIX virtualized filesystem path.\n\nWhen a user adds MCP server configuration to file #1 (as instructed by the UI and documentation), the app reads file #2 and finds no MCP servers configured. The MCP servers never start.\n\n### What Makes This Hard to Debug\n\n- **No error messages** are displayed in the app\n- **No logs folder** is created at `%APPDATA%\\Claude\\logs\\` (because the app isn't reading from there)\n- **The logs folder exists** in the virtualized path but users don't know to look there\n- **Developer settings** show no indication that the config was not loaded\n- **\"Edit Config\" button** confidently opens the wrong file, leading users to believe they're editing the right config\n- **All official documentation** references `%APPDATA%\\Claude\\` as the config location\n\n## Steps to Reproduce\n\n1. Install Claude Desktop on Windows from claude.ai/download (results in MSIX installation)\n\n2. Open Claude Desktop \u2192 Settings \u2192 Developer \u2192 click \"Edit Config\"\n\n3. Add any MCP server configuration to the file that opens, for example:\n\n   ```json\n   {\n     \"mcpServers\": {\n       \"filesystem\": {\n         \"command\": \"npx\",\n         \"args\": [\"-y\", \"@modelcontextprotocol/server-filesystem\", \"C:\\\\Users\\\\username\\\\Desktop\"]\n       }\n     }\n   }\n   ```\n\n4. Save the file and restart Claude Desktop (or use Developer \u2192 Reload MCP Configuration)\n\n5. Observe: No MCP server icon appears, no tools are listed, no logs are generated\n\n## Expected Behavior\n\n- \"Edit Config\" should open the config file that the app actually reads\n- OR the app should read from the standard `%APPDATA%\\Claude\\` path\n- OR at minimum, the app should display a warning/log when no MCP servers are found despite a config file existing in the standard location\n- Documentation should reference the correct path for MSIX installations\n\n## Actual Behavior\n\n- \"Edit Config\" opens a file the app does not read\n- MCP servers silently fail to load\n- No errors, no warnings, no logs at the expected location\n- Users can spend hours troubleshooting (verifying JSON syntax, testing Node.js, checking paths) without finding the issue\n\n## Workaround\n\nEdit the config file at the MSIX virtualized path directly:\n\n```\n%LOCALAPPDATA%\\Packages\\Claude_pzs8sxrjxfjjc\\LocalCache\\Roaming\\Claude\\claude_desktop_config.json\n```\n\nNote: This file already contains app preferences (e.g., `coworkScheduledTasksEnabled`, `sidebarMode`), so MCP config must be merged into the existing JSON structure:\n\n```json\n{\n  \"preferences\": {\n    \"coworkScheduledTasksEnabled\": false,\n    \"sidebarMode\": \"chat\"\n  },\n  \"mcpServers\": {\n    \"your-server\": {\n      \"command\": \"...\",\n      \"args\": [\"...\"]\n    }\n  }\n}\n```\n\nAlternatively, create a symbolic link from the virtualized path to the standard path (requires admin privileges).\n\n## Impact\n\nThis affects **all Windows users** who install Claude Desktop via the official installer (which now defaults to MSIX packaging). Any user following the official MCP setup documentation will hit this issue. The silent failure with no error messages makes it extremely difficult to diagnose without deep knowledge of Windows MSIX filesystem virtualization.\n\n## Suggested Fix\n\nOne or more of the following:\n\n1. **Make \"Edit Config\" open the correct file** \u2014 before calling `shell.openPath()`, resolve the actual virtualized path the app reads from (e.g., using `process.env.LOCALAPPDATA + '\\Packages\\...'` or Windows `GetPackagePath` APIs) so the editor opens the same file the app uses\n2. **Read from the non-virtualized path explicitly** \u2014 bypass MSIX redirection by using the real `%APPDATA%\\Claude\\` path with full filesystem APIs, ensuring the app and \"Edit Config\" both use the same file\n3. **Add logging/diagnostics** \u2014 surface a warning in Developer settings when the config file has no `mcpServers` key or when MCP server startup fails. Generate logs even when no MCP servers are configured, so users have something to debug with\n4. **Update documentation** \u2014 mention the MSIX virtualized path for Windows users and note that \"Edit Config\" may open the wrong file on MSIX installations\n\n### What Should Happen?\n\n## Bug Description\n\n### The Problem\n\nThere are **two** `claude_desktop_config.json` files on the system:\n\n1. **The file \"Edit Config\" opens** (the wrong one):\n\n   ```\n   C:\\Users\\<username>\\AppData\\Roaming\\Claude\\claude_desktop_config.json\n   ```\n\n   \u2192 This is the standard `%APPDATA%\\Claude\\` path referenced in all official documentation.\n\n2. **The file the app actually reads** (the correct one):\n\n   ```\n   C:\\Users\\<username>\\AppData\\Local\\Packages\\Claude_pzs8sxrjxfjjc\\LocalCache\\Roaming\\Claude\\claude_desktop_config.json\n   ```\n\n   \u2192 This is the MSIX virtualized filesystem path.\n\nWhen a user adds MCP server configuration to file #1 (as instructed by the UI and documentation), the app reads file #2 and finds no MCP servers configured. The MCP servers never start.\n\n### What Makes This Hard to Debug\n\n- **No error messages** are displayed in the app\n- **No logs folder** is created at `%APPDATA%\\Claude\\logs\\` (because the app isn't reading from there)\n- **The logs folder exists** in the virtualized path but users don't know to look there\n- **Developer settings** show no indication that the config was not loaded\n- **\"Edit Config\" button** confidently opens the wrong file, leading users to believe they're editing the right config\n- **All official documentation** references `%APPDATA%\\Claude\\` as the config location\n\n## Actual Behavior\n\n- \"Edit Config\" opens a file the app does not read\n- MCP servers silently fail to load\n- No errors, no warnings, no logs at the expected location\n- Users can spend hours troubleshooting (verifying JSON syntax, testing Node.js, checking paths) without finding the issue\n\n## Workaround\n\nEdit the config file at the MSIX virtualized path directly:\n\n```\n%LOCALAPPDATA%\\Packages\\Claude_pzs8sxrjxfjjc\\LocalCache\\Roaming\\Claude\\claude_desktop_config.json\n```\n\nNote: This file already contains app preferences (e.g., `coworkScheduledTasksEnabled`, `sidebarMode`), so MCP config must be merged into the existing JSON structure:\n\n```json\n{\n  \"preferences\": {\n    \"coworkScheduledTasksEnabled\": false,\n    \"sidebarMode\": \"chat\"\n  },\n  \"mcpServers\": {\n    \"your-server\": {\n      \"command\": \"...\",\n      \"args\": [\"...\"]\n    }\n  }\n}\n```\n\nAlternatively, create a symbolic link from the virtualized path to the standard path (requires admin privileges).\n\n## Impact\n\nThis affects **all Windows users** who install Claude Desktop via the official installer (which now defaults to MSIX packaging). Any user following the official MCP setup documentation will hit this issue. The silent failure with no error messages makes it extremely difficult to diagnose without deep knowledge of Windows MSIX filesystem virtualization.\n\n## Suggested Fix\n\nOne or more of the following:\n\n1. **Make \"Edit Config\" open the correct file** \u2014 before calling `shell.openPath()`, resolve the actual virtualized path the app reads from (e.g., using `process.env.LOCALAPPDATA + '\\Packages\\...'` or Windows `GetPackagePath` APIs) so the editor opens the same file the app uses\n2. **Read from the non-virtualized path explicitly** \u2014 bypass MSIX redirection by using the real `%APPDATA%\\Claude\\` path with full filesystem APIs, ensuring the app and \"Edit Config\" both use the same file\n3. **Add logging/diagnostics** \u2014 surface a warning in Developer settings when the config file has no `mcpServers` key or when MCP server startup fails. Generate logs even when no MCP servers are configured, so users have something to debug with\n4. **Update documentation** \u2014 mention the MSIX virtualized path for Windows users and note that \"Edit Config\" may open the wrong file on MSIX installations\n\n### Error Messages/Logs\n\n```shell\n\n```\n\n### Steps to Reproduce\n\n## Steps to Reproduce\n\n1. Install Claude Desktop on Windows from claude.ai/download (results in MSIX installation)\n\n2. Open Claude Desktop \u2192 Settings \u2192 Developer \u2192 click \"Edit Config\"\n\n3. Add any MCP server configuration to the file that opens, for example:\n\n   ```json\n   {\n     \"mcpServers\": {\n       \"filesystem\": {\n         \"command\": \"npx\",\n         \"args\": [\"-y\", \"@modelcontextprotocol/server-filesystem\", \"C:\\\\Users\\\\username\\\\Desktop\"]\n       }\n     }\n   }\n   ```\n\n4. Save the file and restart Claude Desktop (or use Developer \u2192 Reload MCP Configuration)\n\n5. Observe: No MCP server icon appears, no tools are listed, no logs are generated\n\n## Expected Behavior\n\n- \"Edit Config\" should open the config file that the app actually reads\n- OR the app should read from the standard `%APPDATA%\\Claude\\` path\n- OR at minimum, the app should display a warning/log when no MCP servers are found despite a config file existing in the standard location\n- Documentation should reference the correct path for MSIX installations\n\n### Claude Model\n\nNone\n\n### Is this a regression?\n\nYes, this worked in a previous version\n\n### Last Working Version\n\n_No response_\n\n### Claude Code Version\n\nClaude Desktop\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nWindows\n\n### Terminal/Shell\n\nWindows Terminal\n\n### Additional Information\n\n_No response_",
          "state": "open",
          "age_days": 18,
          "comments": 16,
          "reactions_total": 15,
          "upvotes": 15,
          "downvotes": 0,
          "created_at": "2026-02-16T10:15:23Z",
          "url": "https://github.com/anthropics/claude-code/issues/26073",
          "labels": [
            "bug",
            "invalid"
          ],
          "cluster_id": 3
        },
        {
          "number": 25865,
          "title": "[BUG] Notion MCP connector serializes JSON object parameters as strings, breaking update/move/create operations",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\nWhen using the official Notion MCP connector in Cowork mode, all tool calls that require a JSON **object** parameter fail because the MCP framework serializes the object into a JSON **string** before passing it to the tool.\n\nThree tools are affected:\n\n| Tool | Broken Parameter | Expected Type |\n|------|-----------------|---------------|\n| `notion-update-page` | `data` | object |\n| `notion-move-pages` | `new_parent` | object |\n| `notion-create-pages` | `parent` | object |\n\nThe parameter value `{\"type\": \"page_id\", \"page_id\": \"abc123\"}` arrives at the Notion MCP server as the string `\"{\\\"type\\\": \\\"page_id\\\", \\\"page_id\\\": \\\"abc123\\\"}\"` \u2014 causing Zod validation to reject it.\n\nThis makes it impossible to update page content, move pages, or create pages under a parent. The only working write path is `notion-create-pages` without a `parent` parameter (which creates orphan pages at workspace root), requiring users to manually drag pages into position in Notion afterward.\n\nTools with string-typed parameters (`notion-fetch`, `notion-search`, `notion-create-pages` without parent) work fine \u2014 suggesting the framework JSON.stringify()'s all parameter values regardless of whether the tool schema expects a string or object.\n\n### What Should Happen?\n\nObject parameters should be passed as actual JSON objects to the MCP tool, preserving their type. Tools like `notion-move-pages`, `notion-update-page`, and `notion-create-pages` (with parent) should work as documented.\n\n### Error Messages/Logs\n\n```shell\n**notion-move-pages:**\n\nMCP error -32602: Invalid arguments for tool notion-move-pages: [\n  {\n    \"code\": \"invalid_union\",\n    \"unionErrors\": [\n      {\n        \"issues\": [\n          {\n            \"code\": \"invalid_type\",\n            \"expected\": \"object\",\n            \"received\": \"string\",\n            \"path\": [\"new_parent\"],\n            \"message\": \"Expected object, received string\"\n          }\n        ],\n        \"name\": \"ZodError\"\n      }\n    ],\n    \"path\": [\"new_parent\"],\n    \"message\": \"Invalid input\"\n  }\n]\n\n\n**notion-update-page:**\n\nMCP error -32602: Invalid arguments for tool notion-update-page: [\n  {\n    \"code\": \"invalid_type\",\n    \"expected\": \"object\",\n    \"received\": \"string\",\n    \"path\": [\"data\"],\n    \"message\": \"Expected object, received string\"\n  }\n]\n\n\n**notion-create-pages (with parent):**\n\nMCP error -32602: Invalid arguments for tool notion-create-pages: [\n  {\n    \"code\": \"invalid_type\",\n    \"expected\": \"object\",\n    \"received\": \"string\",\n    \"path\": [\"parent\"],\n    \"message\": \"Expected object, received string\"\n  }\n]\n```\n\n### Steps to Reproduce\n\n1. Open Claude Desktop \u2192 Cowork mode\n2. Connect the official Notion MCP connector\n3. Have at least two Notion pages shared with the integration (a parent page and a child page)\n4. Ask Claude: \"Move page [child-page-url] under [parent-page-url]\"\n5. Claude will call `notion-move-pages` with:\n   ```json\n   {\n     \"page_or_database_ids\": [\"<child-page-id>\"],\n     \"new_parent\": {\"type\": \"page_id\", \"page_id\": \"<parent-page-id>\"}\n   }\n   ```\n6. Observe `Expected object, received string` error on `new_parent`\n7. Repeat with \"Update the content of [page-url]\" \u2192 same error on `data` param of `notion-update-page`\n8. Repeat with \"Create a new page under [parent-page-url]\" \u2192 same error on `parent` param of `notion-create-pages`\n\n**Environment:**\n- Platform: Claude Desktop (Cowork mode)\n- OS: macOS\n- Notion MCP connector ID: mcp__2a4c8006-4047-4496-8cd4-0d762c457ab2\n- Date: February 15, 2026\n\n### Claude Model\n\nOpus\n\n### Is this a regression?\n\nYes, this worked in a previous version\n\n### Last Working Version\n\nWorking as of February 13, 2026. Broke by February 15, 2026. Exact version unknown \u2014 Claude Desktop / Cowork mode auto-updates.\n\n### Claude Code Version\n\nClaude Desktop (Cowork mode) \u2014 version unknown, auto-updated between Feb 13\u201315, 2026\n\n### Platform\n\nOther\n\n### Operating System\n\nmacOS\n\n### Terminal/Shell\n\nTerminal.app (macOS)\n\n### Additional Information\n\nBug occurs in Cowork mode (not Claude Code CLI). The Notion MCP connector is the official hosted connector (ID: mcp__2a4c8006-4047-4496-8cd4-0d762c457ab2), not a self-hosted notion-mcp-server. The Terminal/Shell field is not applicable \u2014 this is entirely within the Cowork UI, no terminal involved. ",
          "state": "open",
          "age_days": 20,
          "comments": 10,
          "reactions_total": 15,
          "upvotes": 15,
          "downvotes": 0,
          "created_at": "2026-02-15T05:52:55Z",
          "url": "https://github.com/anthropics/claude-code/issues/25865",
          "labels": [],
          "cluster_id": 3
        },
        {
          "number": 27625,
          "title": "Claude Desktop: Built-in MCP servers 'Claude in Chrome' and 'mcp-registry' fail to connect (Failed to get account UUID)",
          "body": "## Description\n\nClaude Desktop v1.1.3647 on macOS consistently fails to connect its two built-in MCP servers: **\"Claude in Chrome\"** and **\"mcp-registry\"**. These errors appear as toast notifications every time extensions are installed or conversations start. User-installed extensions (Control Chrome, iMessage, Desktop Commander) all connect successfully.\n\n## Error Details\n\n### Claude in Chrome\n```\n[chrome-mcp] Failed to get account UUID\n[Claude in Chrome] No user ID available after Xms\nchrome_bridge_connection_failed | error_type: 'no_user_id'\n[Claude in Chrome] Bridge reconnecting in 3000ms (attempt 2)\n```\n\nThe bridge registers successfully on startup:\n```\n[Chrome MCP] Bridge feature flag: chrome_ext_bridge_enabled_desktop=true, IS_NEST_BUILD=false\n[Claude in Chrome] MCP server registered\n```\n\nBut when a conversation requests a connection, the server never launches. The `main.log` shows `MCP Server connection requested for: Claude in Chrome` but no corresponding `Launching MCP Server` entry appears (unlike user-installed extensions which do launch).\n\n### mcp-registry\nSame pattern: `MCP Server connection requested for: mcp-registry` appears in logs but the server never launches. No explicit error logged.\n\n## Environment\n- **Claude Desktop**: v1.1.3647 (commit 8f7c53e009d62927d53b39524d8fd204b2bb8b10)\n- **macOS**: Darwin 25.3.0 (Sequoia)\n- **Chrome extension**: Installed (ID: fcoeoabgfenejglbffodgkkbkcdhcgfn, v1.0.54)\n- **Chrome native host**: Working (socket created at `/tmp/claude-mcp-browser-bridge-*/`, listening for connections)\n- **Node.js**: v25.4.0\n\n## Key Observations\n\n1. **OAuth works for conversations** \u2014 the user is signed in and can chat normally. `[oauth] using cached token for orgId=...` appears in logs.\n2. **The account UUID retrieval specifically fails** \u2014 even though OAuth tokens are valid, the Chrome MCP bridge component cannot get the account UUID.\n3. **Native messaging infrastructure is healthy** \u2014 the `chrome-native-host` binary creates sockets, listens, but Claude Desktop never connects to them (no \"Accepted new MCP connection\" in native host logs).\n4. **User-installed extensions work fine** \u2014 Control Chrome, iMessage, and Desktop Commander all show \"Server started and connected successfully\" in their logs.\n5. **The error is persistent** \u2014 occurs across multiple app restarts, sign out/sign in cycles, and Chrome extension re-pairing.\n6. **Wait times vary wildly** \u2014 from 43ms to 992272ms (16+ minutes), suggesting the bridge genuinely waits for auth but never gets it.\n\n## Steps to Reproduce\n1. Install Claude Desktop v1.1.3647 on macOS\n2. Sign in to Claude Desktop\n3. Install the Claude Chrome extension and pair it\n4. Install any extension from the marketplace (e.g., Control Chrome)\n5. Observe toast notifications: \"Could not connect to MCP server mcp-registry\" and \"Could not connect to MCP server Claude in Chrome\"\n\n## What Was Tried\n- Signing out and back into Claude Desktop\n- Re-pairing the Chrome extension\n- Restarting both Chrome and Claude Desktop\n- Reinstalling extensions\n\nNone of these resolved the issue.\n\n## Expected Behavior\nBuilt-in MCP servers should connect without errors, or at minimum fail silently without showing error toasts to the user (since user-installed extensions work fine).",
          "state": "open",
          "age_days": 12,
          "comments": 10,
          "reactions_total": 14,
          "upvotes": 13,
          "downvotes": 0,
          "created_at": "2026-02-22T10:15:29Z",
          "url": "https://github.com/anthropics/claude-code/issues/27625",
          "labels": [],
          "cluster_id": 3
        },
        {
          "number": 26094,
          "title": "[BUG] Claude Desktop 1.1.3189 (Cowork) serializes MCP object parameters as strings, breaking Notion MCP writes",
          "body": "### Preflight Checklist\n\n- [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\n- [x] This is a single bug report (please file separate reports for different bugs)\n- [x] I am using the latest version of Claude Code\n\n### What's Wrong?\n\nAfter updating Claude Desktop to 1.1.3189, all Notion MCP tools that accept JSON object parameters fail with ZodError \"Expected object, received string\". Tools with only flat string/array parameters continue to work fine.\n\nAffected tools:\n- `notion-create-pages` \u2014 `parent` parameter fails\n- `notion-move-pages` \u2014 `new_parent` parameter fails\n- `notion-update-page` \u2014 `data` parameter fails\n\nWorking tools (flat parameters only):\n`notion-fetch`, `notion-search`, `notion-get-comments`, `notion-create-comment`, `notion-get-users`, `notion-get-teams`, `notion-query-database-view`\n\n### What Should Happen?\n\nObject parameters should be passed as parsed JSON objects to MCP tools, not as serialized JSON strings. All three tools worked correctly in a Cowork session approximately 24 hours before the update.\n\n### Error Messages/Logs\n\n```shell\nMCP error -32602: Invalid arguments for tool notion-create-pages: [\n  {\n    \"code\": \"invalid_union\",\n    \"unionErrors\": [\n      {\n        \"issues\": [\n          {\n            \"code\": \"invalid_type\",\n            \"expected\": \"object\",\n            \"received\": \"string\",\n            \"path\": [\"parent\"],\n            \"message\": \"Expected object, received string\"\n          }\n        ]\n      }\n    ],\n    \"path\": [\"parent\"],\n    \"message\": \"Invalid input\"\n  }\n]\n```\n\n### Steps to Reproduce\n\n1. Open Cowork mode in Claude Desktop 1.1.3189\n2. Connect the built-in Notion MCP connector (remote MCP)\n3. Ask Claude to create a page in a Notion database \u2014 this triggers `notion-create-pages` with a `parent` object parameter like `{\"data_source_id\": \"some-uuid\"}`\n4. Observe ZodError: \"Expected object, received string\" on the `parent` parameter\n5. Repeat with `notion-update-page` (data parameter) or `notion-move-pages` (new_parent parameter) \u2014 same error\n6. Verify that read-only tools (`notion-fetch`, `notion-search`) still work \u2014 they do, because they use flat string parameters\n\n### Claude Model\n\nOpus\n\n### Is this a regression?\n\nYes, this worked in a previous version\n\n### Last Working Version\n\nUnknown exact version \u2014 writes were working in a Cowork session on Feb 15, 2026 (one day before updating to 1.1.3189)\n\n### Claude Code Version\n\nClaude Desktop 1.1.3189 (1b7b58), build 2026-02-14T00:09:46.000Z (Cowork mode, not CLI)\n\n### Platform\n\nAnthropic API\n\n### Operating System\n\nmacOS\n\n### Terminal/Shell\n\nOther\n\n### Additional Information\n\nRelated Notion MCP issue: https://github.com/makenotion/notion-mcp-server/issues/208\nMultiple users report the same symptoms there, including with non-Claude MCP clients.\n\nThe pattern is consistent: every MCP tool parameter typed as a JSON object gets serialized as a string before reaching the MCP server. Parameters typed as plain strings or arrays work fine. This suggests the serialization issue is in the MCP client layer, not the server.",
          "state": "open",
          "age_days": 18,
          "comments": 10,
          "reactions_total": 14,
          "upvotes": 13,
          "downvotes": 0,
          "created_at": "2026-02-16T12:10:01Z",
          "url": "https://github.com/anthropics/claude-code/issues/26094",
          "labels": [
            "bug",
            "invalid"
          ],
          "cluster_id": 3
        }
      ]
    },
    {
      "id": 6,
      "issue_count": 87,
      "stats": {
        "total_reactions": 112,
        "total_comments": 289,
        "avg_age_days": 14.8
      },
      "analysis": {
        "cluster_title": "Keyword Group: Root Bunfs Bunfs Root 39 Native",
        "summary": "This semantic cluster was defined by the high density of the following keywords: root, bunfs, bunfs root, 39, native.",
        "hardest_problems": [
          "[Bug] Anthropic API Error: Long context beta access revoked for eligible subscription",
          "API Error: Rate limit reached",
          "[Bug] Anthropic API Error: 500 Internal Server Error with intermittent failures"
        ],
        "hallucination_check": "Local heuristic analysis based on TF-IDF term frequencies and engagement metrics."
      },
      "top_issues": [
        {
          "number": 27943,
          "title": "[Bug] Anthropic API Error: Long context beta access revoked for eligible subscription",
          "body": "**Bug Description**\ni was able to access 1M context opus model with my plan but all of a sudden its not working. its giving me this error: API Error: 400 {\"type\":\"error\",\"error\":{\"type\":\"invalid_request_error\",\"message\":\"The long context beta is not yet available for this subscription.\"},\"request_id\":\"req_011CYRQCm6X1keZLRqVa5Pgq\"}  \n\n**Environment Info**\n- Platform: darwin\n- Terminal: vscode\n- Version: 2.1.50\n- Feedback ID: 9c9f09c9-ced6-4899-bb3f-3f10d7b1dc5f\n\n**Errors**\n```json\n[{\"error\":\"Error: NON-FATAL: Lock acquisition failed for /Users/nikhilsharma/.local/share/claude/versions/2.1.50 (expected in multi-process scenarios)\\n    at OzR (/$bunfs/root/claude:2600:2098)\\n    at SbA (/$bunfs/root/claude:2600:1209)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-23T17:32:20.293Z\"},{\"error\":\"Error: Plugin pyright-lsp has an invalid manifest file at /Users/nikhilsharma/.claude/plugins/cache/claude-plugins-official/pyright-lsp/1.0.0/.claude-plugin/plugin.json.\\n\\nValidation errors: : Unrecognized keys: \\\"source\\\", \\\"category\\\", \\\"strict\\\"\\n\\nPlease fix the manifest or remove it. The plugin cannot load with an invalid manifest.\\n    at UUR (/$bunfs/root/claude:2501:11033)\\n    at async e2B (/$bunfs/root/claude:2509:378)\\n    at async LA8 (/$bunfs/root/claude:2509:8500)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-23T17:32:20.310Z\"},{\"error\":\"Error: 400 {\\\"type\\\":\\\"error\\\",\\\"error\\\":{\\\"type\\\":\\\"invalid_request_error\\\",\\\"message\\\":\\\"The long context beta is not yet available for this subscription.\\\"},\\\"request_id\\\":\\\"req_011CYRQBGrce8JwDqeeZphMU\\\"}\\n    at generate (/$bunfs/root/claude:262:28230)\\n    at makeRequest (/$bunfs/root/claude:301:5435)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-23T17:32:23.918Z\"},{\"error\":\"Error: Failed to delete keychain entry\\n    at eUB (/$bunfs/root/claude:2601:6910)\\n    at async nRD (/$bunfs/root/claude:6103:685)\\n    at async lzA (/$bunfs/root/claude:6103:589)\\n    at async qET (/$bunfs/root/claude:2111:2222)\\n    at async startOAuthFlow (/$bunfs/root/claude:2111:3618)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-23T17:32:38.606Z\"},{\"error\":\"Error: 400 {\\\"type\\\":\\\"error\\\",\\\"error\\\":{\\\"type\\\":\\\"invalid_request_error\\\",\\\"message\\\":\\\"The long context beta is not yet available for this subscription.\\\"},\\\"request_id\\\":\\\"req_011CYRQCm6X1keZLRqVa5Pgq\\\"}\\n    at generate (/$bunfs/root/claude:262:28230)\\n    at makeRequest (/$bunfs/root/claude:301:5435)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-23T17:32:44.020Z\"}]\n```\n",
          "state": "closed",
          "age_days": 11,
          "comments": 10,
          "reactions_total": 12,
          "upvotes": 12,
          "downvotes": 0,
          "created_at": "2026-02-23T17:34:37Z",
          "url": "https://github.com/anthropics/claude-code/issues/27943",
          "labels": [],
          "cluster_id": 6
        },
        {
          "number": 26496,
          "title": "API Error: Rate limit reached",
          "body": "**Bug Description**\n\n\n**Environment Info**\n- Platform: darwin\n- Terminal: tmux\n- Version: 2.1.45\n- Feedback ID: 69b46464-12e0-4bb8-9d1a-710144248c02\n\n**Errors**\n```json\n[{\"error\":\"Error: NON-FATAL: Lock acquisition failed for /Users/storm/.local/share/claude/versions/2.1.45 (expected in multi-process scenarios)\\n    at azR (/$bunfs/root/claude:3442:2098)\\n    at BjA (/$bunfs/root/claude:3442:1209)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T05:14:30.464Z\"},{\"error\":\"Error: LSP server plugin:php-lsp:intelephense failed to start: Executable not found in $PATH: \\\"intelephense\\\"\\n    at start (/$bunfs/root/claude:2317:22314)\\n    at async H (/$bunfs/root/claude:2317:24762)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T05:14:32.999Z\"},{\"error\":\"Error: LSP server plugin:typescript-lsp:typescript failed to start: Executable not found in $PATH: \\\"typescript-language-server\\\"\\n    at start (/$bunfs/root/claude:2317:22314)\\n    at async H (/$bunfs/root/claude:2317:24762)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T05:14:32.999Z\"},{\"error\":\"Error: LSP server plugin:pyright-lsp:pyright failed to start: Executable not found in $PATH: \\\"pyright-langserver\\\"\\n    at start (/$bunfs/root/claude:2317:22314)\\n    at async H (/$bunfs/root/claude:2317:24762)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T05:14:32.999Z\"},{\"error\":\"Error: LSP server plugin:lua-lsp:lua failed to start: Executable not found in $PATH: \\\"lua-language-server\\\"\\n    at start (/$bunfs/root/claude:2317:22314)\\n    at async H (/$bunfs/root/claude:2317:24762)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T05:14:32.999Z\"},{\"error\":\"Error: Executable not found in $PATH: \\\"intelephense\\\"\\n    at spawn (unknown)\\n    at spawn (node:child_process:667:35)\\n    at spawn (node:child_process:14:39)\\n    at start (/$bunfs/root/claude:2317:20844)\\n    at H (/$bunfs/root/claude:2317:24762)\\n    at _ (/$bunfs/root/claude:2317:32538)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T05:14:32.999Z\"},{\"error\":\"Error: Executable not found in $PATH: \\\"typescript-language-server\\\"\\n    at spawn (unknown)\\n    at spawn (node:child_process:667:35)\\n    at spawn (node:child_process:14:39)\\n    at start (/$bunfs/root/claude:2317:20844)\\n    at H (/$bunfs/root/claude:2317:24762)\\n    at _ (/$bunfs/root/claude:2317:32538)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T05:14:32.999Z\"},{\"error\":\"Error: Executable not found in $PATH: \\\"pyright-langserver\\\"\\n    at spawn (unknown)\\n    at spawn (node:child_process:667:35)\\n    at spawn (node:child_process:14:39)\\n    at start (/$bunfs/root/claude:2317:20844)\\n    at H (/$bunfs/root/claude:2317:24762)\\n    at _ (/$bunfs/root/claude:2317:32538)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T05:14:32.999Z\"},{\"error\":\"Error: Executable not found in $PATH: \\\"lua-language-server\\\"\\n    at spawn (unknown)\\n    at spawn (node:child_process:667:35)\\n    at spawn (node:child_process:14:39)\\n    at start (/$bunfs/root/claude:2317:20844)\\n    at H (/$bunfs/root/claude:2317:24762)\\n    at _ (/$bunfs/root/claude:2317:32538)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T05:14:32.999Z\"},{\"error\":\"Error: Failed to start LSP server plugin:php-lsp:intelephense: Executable not found in $PATH: \\\"intelephense\\\"\\n    at <anonymous> (/$bunfs/root/claude:2317:32560)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T05:14:32.999Z\"},{\"error\":\"Error: Failed to start LSP server plugin:typescript-lsp:typescript: Executable not found in $PATH: \\\"typescript-language-server\\\"\\n    at <anonymous> (/$bunfs/root/claude:2317:32560)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T05:14:32.999Z\"},{\"error\":\"Error: Failed to start LSP server plugin:pyright-lsp:pyright: Executable not found in $PATH: \\\"pyright-langserver\\\"\\n    at <anonymous> (/$bunfs/root/claude:2317:32560)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T05:14:32.999Z\"},{\"error\":\"Error: Failed to start LSP server plugin:lua-lsp:lua: Executable not found in $PATH: \\\"lua-language-server\\\"\\n    at <anonymous> (/$bunfs/root/claude:2317:32560)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T05:14:32.999Z\"},{\"error\":\"Error: 429 {\\\"type\\\":\\\"error\\\",\\\"error\\\":{\\\"type\\\":\\\"rate_limit_error\\\",\\\"message\\\":\\\"Extra usage is required for long context requests.\\\"},\\\"request_id\\\":\\\"req_011CYExsNitfPvqgdFYrKXJa\\\"}\\n    at generate (/$bunfs/root/claude:327:9006)\\n    at makeRequest (/$bunfs/root/claude:366:5435)\\n    at\u2026\n```\n\n**Note:** Content was truncated.\n\n<img width=\"682\" height=\"556\" alt=\"Image\" src=\"https://github.com/user-attachments/assets/097b8bd2-b685-4c76-9414-3f5282190741\" />\n",
          "state": "open",
          "age_days": 17,
          "comments": 9,
          "reactions_total": 6,
          "upvotes": 6,
          "downvotes": 0,
          "created_at": "2026-02-18T05:22:57Z",
          "url": "https://github.com/anthropics/claude-code/issues/26496",
          "labels": [],
          "cluster_id": 6
        },
        {
          "number": 28623,
          "title": "[Bug] Anthropic API Error: 500 Internal Server Error with intermittent failures",
          "body": "**Bug Description**\nI'm getting repeated 500 API errors. It's stopping my work in its track, sometimes a few seconds into doing something and sometimes a few minutes. \n\n\u00a0API Error: 500 {\"type\":\"error\",\"error\":{\"type\":\"api_error\",\"message\":\"Internal server\n     error\"},\"request_id\":\"req_011CYVACPyuYrPgfVu7crA2V\"}\n\n\n\n**Environment Info**\n- Platform: darwin\n- Terminal: iTerm.app\n- Version: 2.1.56\n- Feedback ID: 16c47f90-85f3-4976-96cf-f88861ca1f5f\n\n**Errors**\n```json\n[{\"error\":\"Error: 500 {\\\"type\\\":\\\"error\\\",\\\"error\\\":{\\\"type\\\":\\\"api_error\\\",\\\"message\\\":\\\"Internal server error\\\"},\\\"request_id\\\":\\\"req_011CYVACPyuYrPgfVu7crA2V\\\"}\\n    at generate (/$bunfs/root/claude:267:28561)\\n    at makeRequest (/$bunfs/root/claude:306:5435)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T17:11:26.424Z\"}]\n```\n",
          "state": "closed",
          "age_days": 9,
          "comments": 3,
          "reactions_total": 11,
          "upvotes": 10,
          "downvotes": 0,
          "created_at": "2026-02-25T17:15:03Z",
          "url": "https://github.com/anthropics/claude-code/issues/28623",
          "labels": [
            "bug",
            "duplicate",
            "platform:macos",
            "external",
            "area:api",
            "api:anthropic"
          ],
          "cluster_id": 6
        },
        {
          "number": 28608,
          "title": "[Bug] Anthropic API Error: Internal server error (500)",
          "body": "**Bug Description**\ngot this error more than 10 times in the last minutes, I can't work like this. fix this please!\n\nAPI Error: 500 {\"type\":\"error\",\"error\":{\"type\":\"api_error\",\"message\":\"Internal server\n     error\"},\"request_id\":\"req_011CYV9R1djajLXXiiej6tXu\"}\n\n**Environment Info**\n- Platform: darwin\n- Terminal: iTerm.app\n- Version: 2.1.56\n- Feedback ID: b89c1772-e3bb-467b-9ca2-d66889fc3b0f\n\n**Errors**\n```json\n[{\"error\":\"Error: LSP server plugin:typescript-lsp:typescript failed to start: Executable not found in $PATH: \\\"typescript-language-server\\\"\\n    at start (/$bunfs/root/claude:2638:22314)\\n    at async H (/$bunfs/root/claude:2638:24624)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T16:43:51.852Z\"},{\"error\":\"Error: Executable not found in $PATH: \\\"typescript-language-server\\\"\\n    at spawn (unknown)\\n    at spawn (node:child_process:667:35)\\n    at spawn (node:child_process:14:39)\\n    at start (/$bunfs/root/claude:2638:20844)\\n    at H (/$bunfs/root/claude:2638:24624)\\n    at _ (/$bunfs/root/claude:2638:32726)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T16:43:51.852Z\"},{\"error\":\"Error: Failed to start LSP server plugin:typescript-lsp:typescript: Executable not found in $PATH: \\\"typescript-language-server\\\"\\n    at <anonymous> (/$bunfs/root/claude:2638:32748)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T16:43:51.852Z\"},{\"error\":\"Error: 500 {\\\"type\\\":\\\"error\\\",\\\"error\\\":{\\\"type\\\":\\\"api_error\\\",\\\"message\\\":\\\"Internal server error\\\"},\\\"request_id\\\":\\\"req_011CYV8HjYJT3wwDCoY5z7Jw\\\"}\\n    at generate (/$bunfs/root/claude:267:28561)\\n    at makeRequest (/$bunfs/root/claude:306:5435)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T16:46:24.937Z\"},{\"error\":\"Error: Request was aborted.\\n    at makeRequest (/$bunfs/root/claude:306:3940)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T16:51:52.773Z\"},{\"error\":\"Error: 500 {\\\"type\\\":\\\"error\\\",\\\"error\\\":{\\\"type\\\":\\\"api_error\\\",\\\"message\\\":\\\"Internal server error\\\"},\\\"request_id\\\":\\\"req_011CYV8k2QnW1w7d3UnF4gKh\\\"}\\n    at generate (/$bunfs/root/claude:267:28561)\\n    at makeRequest (/$bunfs/root/claude:306:5435)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T16:52:21.618Z\"},{\"error\":\"Error: 500 {\\\"type\\\":\\\"error\\\",\\\"error\\\":{\\\"type\\\":\\\"api_error\\\",\\\"message\\\":\\\"Internal server error\\\"},\\\"request_id\\\":\\\"req_011CYV986HZYu1SwfG3F72Hz\\\"}\\n    at generate (/$bunfs/root/claude:267:28561)\\n    at makeRequest (/$bunfs/root/claude:306:5435)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T16:57:21.017Z\"},{\"error\":\"Error: 500 {\\\"type\\\":\\\"error\\\",\\\"error\\\":{\\\"type\\\":\\\"api_error\\\",\\\"message\\\":\\\"Internal server error\\\"},\\\"request_id\\\":\\\"req_011CYV9PFrhJdKUT9acgdHRq\\\"}\\n    at generate (/$bunfs/root/claude:267:28561)\\n    at makeRequest (/$bunfs/root/claude:306:5435)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T17:00:46.928Z\"},{\"error\":\"Error: 500 {\\\"type\\\":\\\"error\\\",\\\"error\\\":{\\\"type\\\":\\\"api_error\\\",\\\"message\\\":\\\"Internal server error\\\"},\\\"request_id\\\":\\\"req_011CYV9QP1afZkXmUGjAk8Ta\\\"}\\n    at generate (/$bunfs/root/claude:267:28561)\\n    at makeRequest (/$bunfs/root/claude:306:5435)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T17:01:01.940Z\"},{\"error\":\"Error: 500 {\\\"type\\\":\\\"error\\\",\\\"error\\\":{\\\"type\\\":\\\"api_error\\\",\\\"message\\\":\\\"Internal server error\\\"},\\\"request_id\\\":\\\"req_011CYV9R1djajLXXiiej6tXu\\\"}\\n    at generate (/$bunfs/root/claude:267:28561)\\n    at makeRequest (/$bunfs/root/claude:306:5435)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-25T17:01:10.466Z\"}]\n```\n",
          "state": "closed",
          "age_days": 9,
          "comments": 14,
          "reactions_total": 0,
          "upvotes": 0,
          "downvotes": 0,
          "created_at": "2026-02-25T17:04:34Z",
          "url": "https://github.com/anthropics/claude-code/issues/28608",
          "labels": [],
          "cluster_id": 6
        },
        {
          "number": 26606,
          "title": "[Bug] Anthropic API Error: Internal Server Error (500)",
          "body": "**Bug Description**\nAPI Error: 500 {\"type\":\"error\",\"error\":{\"type\":\"api_error\",\"message\":\"Internal server error\"},\"request_id\":\"req_011CYFmrpsCfZVCj4YKZhnV5\"}\n\n**Environment Info**\n- Platform: linux\n- Terminal: vte-based\n- Version: 2.1.45\n- Feedback ID: 6aa292b8-0bca-4c12-a1ad-3b7bb71823b7\n\n**Errors**\n```json\n[{\"error\":\"Error: File does not exist.\\n    at call (/$bunfs/root/claude:1292:6219)\\n    at async hE1 (/$bunfs/root/claude:3304:14119)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T14:17:28.168Z\"},{\"error\":\"Error\\n    at iR (/$bunfs/root/claude:42:1144)\\n    at <anonymous> (/$bunfs/root/claude:43:10097)\\n    at emit (node:events:92:22)\\n    at endReadableNT (internal:streams/readable:861:50)\\n    at processTicksAndRejections (native:7:39)\\n    at request (/$bunfs/root/claude:45:2149)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T14:26:00.888Z\"},{\"error\":\"MaxFileReadTokenExceededError: File content (34780 tokens) exceeds maximum allowed tokens (25000). Please use offset and limit parameters to read specific portions of the file, or use the GrepTool to search for specific content.\\n    at UwI (/$bunfs/root/claude:1283:7592)\\n    at async XwI (/$bunfs/root/claude:1287:2464)\\n    at async call (/$bunfs/root/claude:1292:5924)\\n    at async <anonymous> (/$bunfs/root/claude:5777:2040)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T14:36:40.723Z\"},{\"error\":\"Error: Request was aborted.\\n    at jJ$ (/$bunfs/root/claude:920:49642)\\n    at next (native:1:11)\\n    at zrD (/$bunfs/root/claude:5522:5173)\\n    at next (native:1:11)\\n    at NrD (/$bunfs/root/claude:5527:9028)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T14:37:07.110Z\"},{\"error\":\"MaxFileReadTokenExceededError: File content (34780 tokens) exceeds maximum allowed tokens (25000). Please use offset and limit parameters to read specific portions of the file, or use the GrepTool to search for specific content.\\n    at UwI (/$bunfs/root/claude:1283:7592)\\n    at async XwI (/$bunfs/root/claude:1287:2464)\\n    at async call (/$bunfs/root/claude:1292:5924)\\n    at async <anonymous> (/$bunfs/root/claude:5777:2040)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T14:38:15.832Z\"},{\"error\":\"Error: File does not exist.\\n    at call (/$bunfs/root/claude:1292:6219)\\n    at async <anonymous> (/$bunfs/root/claude:5777:2040)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T15:15:11.299Z\"},{\"error\":\"Error: File does not exist.\\n    at call (/$bunfs/root/claude:1292:6219)\\n    at async <anonymous> (/$bunfs/root/claude:5777:2040)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T15:15:11.303Z\"},{\"error\":\"Error: File does not exist.\\n    at call (/$bunfs/root/claude:1292:6219)\\n    at async <anonymous> (/$bunfs/root/claude:5777:2040)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T15:15:11.305Z\"},{\"error\":\"Error: File does not exist.\\n    at call (/$bunfs/root/claude:1292:6219)\\n    at async <anonymous> (/$bunfs/root/claude:5777:2040)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T15:15:11.312Z\"},{\"error\":\"Error: File does not exist.\\n    at call (/$bunfs/root/claude:1292:6219)\\n    at async <anonymous> (/$bunfs/root/claude:5777:2040)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T15:15:11.319Z\"},{\"error\":\"Error: File does not exist.\\n    at call (/$bunfs/root/claude:1292:6219)\\n    at async <anonymous> (/$bunfs/root/claude:5777:2040)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T15:15:15.731Z\"},{\"error\":\"TelemetrySafeError: ### Error\\nError: page.goto: net::ERR_INVALID_AUTH_CREDENTIALS at https://autoretainer.uwservices.cz/\\nCall log:\\n\\u001b[2m  - navigating to \\\"https://autoretainer.uwservices.cz/\\\", waiting until \\\"domcontentloaded\\\"\\u001b[22m\\n\\n    at vbI (/$bunfs/root/claude:1836:15652)\\n    at async T_9 (/$bunfs/root/claude:1836:13148)\\n    at async call (/$bunfs/root/claude:1836:31433)\\n    at async hE1 (/$bunfs/root/claude:3304:14119)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T15:16:06.839Z\"},{\"error\":\"Error: File does not exist.\\n    at call (/$bunfs/root/claude:1292:6219)\\n    at async <anonymous> (/$bunfs/root/claude:5777:2040)\\n    at processTicksAndRejections (native:7:39)\",\"timestamp\":\"2026-02-18T15:20:24.629Z\"},{\"error\":\"Error: 1P event logging: 66 events failed to export (code=ECONNABORTED, timeout of 10000ms exceeded)\\n    at queueFailedEvents (/$bunfs/root/claude:2098:2350)\\n    at async doExport (/$bunfs/root/claude:2098:1257)\\n    at processTicksAndRejec\u2026\n```\n\n**Note:** Content was truncated.\n",
          "state": "closed",
          "age_days": 16,
          "comments": 5,
          "reactions_total": 8,
          "upvotes": 8,
          "downvotes": 0,
          "created_at": "2026-02-18T15:31:49Z",
          "url": "https://github.com/anthropics/claude-code/issues/26606",
          "labels": [],
          "cluster_id": 6
        }
      ]
    }
  ]
}