Tool calls issues on macOS — page 2
Issues the maintainers labelled both platform:macos and area:tools.
1,780 issues · 157 open · 1,623 resolved (91%) · first seen Mar 3, 2025
Is this getting better or worse?
This class of problem is converging. 194 new reports in the last 90 days vs 301 in the 90 before — -36%. The open backlog peaked at 446 in 2025-12 and sits at 157 today.
Derived by counting each matching issue from its creation date until it closed. First-party
computation over the anthropics/claude-code tracker — the trend is not published
anywhere upstream.
Which builds report this
811 of these reports name the Claude Code build they were running, spanning 189 releases. Heaviest reporters:
- Claude Code v2.0.76
- Claude Code v2.1.19
- Claude Code v2.1.34
- Claude Code v2.1.12
- Claude Code v2.1.63
- Claude Code v2.1.29
- Claude Code v2.1.17
- Claude Code v2.1.39
Report volume, not a defect count: an issue can mention a build without that build having caused the failure. Versions come from the version string in the bug-report template, matched against real published tags.
Do these get fixed?
Median time to close is 36 days across 1,623 closures. Of the 1,623 closures with a recorded reason, 25% were closed as completed and 1,217 as not-planned or duplicate. The most recent completed fix landed in 2026-10.
Workarounds reported by the community
471 of these issues post a workaround someone says worked. The highest-engagement ones:
- API Error: 400 due to tool use concurrency issues. Run /rewind to recover the conversation. - [Bug] Anthropic API Error: Unexpected 400 Bad Request Response
- …67 (my report) 8887 8893 8894 8895 8897 8903 EDIT (Oct 14): Workaround for tool use/tool result 400 errors: A research-backed solution was posted on Oct 10: Add the provided instructions to your global CLAUDE.md (~/.clau…
Found in the thread of #8763 · resolved - [FEATURE] Allow users to configure which shell the Bash tool uses or inherit initial shell
- …t shell environment. Alternative Solutions I've tried every workaround I can think of other than because it's not an option. I spent $20 on a claude code session trying to find workarounds, nothing worked. Priority Mediu…
Found in the description of #7490 · still open - [BUG] LSP servers not loading due to race condition between LSP Manager initialization and plugin loading
- …LSP Manager initializes with 0 servers before plugins load Workaround: Downgrade to version 2.0.67 with Claude Model None Is this a regression? Yes, this worked in a previous version Last Working Version 2.0.67 Claude C…
Found in the description of #13952 · resolved - [Feature Request] Support selecting base branch when creating git worktree
- …wrong branch or require manual rebasing after the fact. The workaround of ensuring the main worktree has checked out before invoking Claude is fragile in parallel/concurrent setups. A flag (or equivalent config in agent…
Found in the thread of #23622 · still open - Skills/subagents do not inherit user-level permissions from settings.json
- …macOS 26.2 (25C56) - Settings file: ~/.claude/settings.json Workaround Select option 2 to create workspace-level permissions, but this defeats the purpose of user-level permissions. Impact Significantly degrades user exp…
Found in the description of #18950 · still open
Excerpts are quoted from the linked issue so you can judge relevance before opening it. Threads that only ask for a workaround, or report that there is none, are excluded. These are community suggestions, not official guidance from Anthropic — read the full thread before applying one.
Matching issues 101–200
Ranked by community engagement (reactions weighted over comments). Listing the top 300 of 1,780 matches; the statistics above cover all 1,780.
- LSP plugins not registering servers - LSP Manager initializes before plugins load
- Tool permissions not recognized when running commands in git worktrees
- [Feature Request] Reduce verbosity in plan mode to minimize token consumption
- Claude Code cannot read claude.ai artifact links (sandboxed frame blocks text extraction, scrolling, and export)
- AskUserQuestion auto-completes with empty answers when listed in command allowed-tools
- Hook executor doesn't set CLAUDE_PLUGIN_ROOT environment variable
- [Bug] TaskOutput returns raw JSONL transcript instead of subagent summary (regression in 2.0.77)
- [BUG] Tool-call results empty in live UI then flush late/out-of-order (Bash, Read, MCP, advisor) — macOS, Opus 4.8, parallel batches
- ReferenceError: classifyHandoffIfNeeded is not defined when Task agent completes
- [BUG] Version 2.1.20 hides tool usage details (what files are being read, what search patterns are being used) behind Ctrl + O
- [BUG] Agent system and WebFetch fail with Bedrock EU regions - model region not respected
- [BUG] claude-shell-snapshot-* has a syntax error when using Oh My Zsh
- [BUG] Bash(git commit:*) in allowedTools during claude -p mode isn't respected.
- [Bug] Excessive token usage for short messages (4% usage for "hi")
- [BUG] --enable-lsp-logging option not found
- Validation Failure: Tool Name Exceeds 64 Character Limit
- [BUG] Thinking mode 400 error on FIRST tool call in brand new conversation (Opus 4.5)
- [BUG] API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"Could not process image"},"request_id":"req_011CVaC1ZBLb4PLudLYZ6RZR"}
- Orphaned subagent process leaks memory when parent terminal session terminated
- [BUG] Claude code unusable on MacOS, can't run any bash command
- Tab path autocomplete silently fails with too many files in the project
- Custom agents cannot use deferred MCP tools even when declared in mcpServers/tools frontmatter
- Claude in Chrome: 'Browser extension is not connected' despite correct setup (v2.1.39 + ext v1.0.47)
- [Bug] Sandbox blocks uv cache access during pytest execution
- [BUG] Plugin marketplace removal doesn't clean up settings.json, causing marketplace to reinstall
- [BUG] USE DATETIME when Claude Code does Web Search
- [Bug] API Error 400 - Tool Use Concurrency Issues - Widespread Regression Affecting All Platforms
- Permission bypass when commands are chained with &&
- [Bug] Background subagent tool calls exposed in parent context window
- Crash When Attempting Empty File Write
- Path-based rules (.claude/rules/ with paths: frontmatter) are not loaded on Write tool — only on Read
- [BUG] ExitPlanMode tool fails with "AbortError" - hangs indefinitely when MCP servers active
- [BUG] each tool_use must have a single result. Found multiple `tool_result` blocks
- Subagent launch fails with 'Tool names must be unique' when MCP server is configured
- [BUG] "invalid_request_error" `tool_use` ids were found without `tool_result` blocks immediately after
- [BUG] [1.0.64-] authentification to subagents are no longer propagated
- [BUG] API 400: thinking block ordering error during conversation with Task/Explore agent
- [BUG] Hooks fail with posix_spawn '/bin/sh' ENOENT due to sandbox restrictions
- [BUG] Tool/Agent progress lines repeat hundreds of times instead of updating in place (v2.1.2, macOS, iTerm2)
- [BUG] AskUserQuestion tool misinterprets numeric input
- Skill loading executes inline bash patterns from documentation examples
- [BUG] CLAUDE_CODE_EXTRA_BODY thinking config breaks WebSearch and WebFetch on Opus 4.7 [1m]
- EnterWorktree tool creates worktrees inside .claude/ directory, breaking slash command and skill discovery
- Agent Teams: Teammates don't poll inbox — messages never delivered (tmux backend, macOS)
- Sandbox: zsh writes to /tmp/claude-501/cwd-* causing 'operation not permitted' errors
- [Bug] Anthropic API Error: 400 Tool Concurrency Issues with Opus Model
- [BUG] Continuous "API Error: 400 due to tool use concurrency issues" in Claude Code in VS Code
- [BUG] Prompt-based Stop hook cannot response correct JSON
- File Edit Failure: String Not Found in Target File
- Add an opt-out for the built-in find→bfs / grep→ugrep shadow functions injected into the Bash tool shell
- [BUG] Regression in 2.1.158: claude spirals into redundant/invented tool calls during file reads (2.1.157 clean)
- [BUG] Opus 4.8 in Claude Code declares work "verified" / "done" without running the canonical build — false-green regression vs. Opus 4.7
- WebFetch can fabricate confident factual claims by misreading page boilerplate as live status
- Claude Code freezes during API call ("Befuddling" spinner), terminal becomes completely unresponsive
- [BUG] Claude Code reads .env files and hardcodes secrets into inline scripts
- [BUG] Plugin-defined subagents cannot access MCP tools - breaks plugin ecosystem
- [BUG] Background task queue operations after completion signal cause session hang (stop_reason: null)
- [BUG] WebFetch freezes Claude Code with no way to interrupt besides exiting and restarting, losing all context
- [BUG] Subagent spawning fails with marketplace permission error for Amazon Bedrock Haiku 4.5 despite automatic model access
- [BUG] input_schema does not support oneOf, allOf, or anyOf at the top level
- Inefficient Tool Invocation Causing Significant Development Productivity Loss
- [BUG] Bash tool execution consistently fails with temporary script file error on macOS
- API Error 400 'Could not process image' kills entire session with no recovery
- Claude in Chrome MCP server fails to connect to native host bridge socket
- Claude Code MAX — Web Fetch returns 401 API Error: “OAuth authentication is currently not supported.”
- Unhandled Promise Rejection: String Not Found During File Edit Operation
- [BUG]Token compensation request — Auto Mode classifier outage
- [BUG] - Bash tool commands extremely slow (~30s) despite being instant in normal terminal
- [Suspicious Behavior]: Hidden <system-reminder> 10,000+ injections consuming 15%+ of context window without user knowledge or consent
- [BUG] LSP tool doesn't recognize plugin-based LSP server configurations
- [BUG] API Error 400 "no low surrogate in string" when Bash output contains invalid Unicode
- [BUG] Native installation / binary return API errors
- Multiple Agent Workflow Fails to Automatically Parallelize Tasks
- Programmatic Task Tool: Missing Progress Feedback in SDK Mode
- [Report] File path completion may fail unless USE_BUILTIN_RIPGREP is enabled
- Worktree creation writes an ABSOLUTE core.hooksPath into config.worktree, so worktrees run the MAIN checkout's hooks
- Worktree edits can silently land in the parent repo checkout (nested .claude/worktrees/ + no out-of-worktree write guard)
- [BUG] LSP Tool returns "No LSP server available" despite correct plugin configuration (v2.0.76)
- [BUG] /context not working for Claude Code Bedrock users anymore
- [BUG] SSL certificate configuration don't work with web page `Fetch` (but works with MCP)
- Shell conflict with SCM Breeze causes '_safe_eval: command not found' error
- [BUG] PermissionRequest: Allow does not exit plan mode for `ExitPlanMode` tool.
- Bug: Custom agent body content (instructions) not injected when subagent is spawned via Task tool
- Skill tool incorrectly parses markdown content as Bash command
- [BUG] API Error: 400 error
- [BUG] WebFetch downloads content successfully but returns 404 errors to assistant when using Vertex AI
- [Bug] Anthropic API Error: No assistant message found
- [BUG] Bash tool is completely unresponsive - no commands are executing or returning output
- [BUG] Edit tool applies only one replacement when diff shows multiple
- Bug: `SendUserFile` reports success but delivers nothing (VS Code extension)
- [BUG] Skill frontmatter `model:` field documented but not working
- [BUG] Search/Grep permissions in settings.json not being respected
- [BUG] Claude should not clone claude-plugins-official.git with SSH
- Bash permission pattern with wildcard not matching commands with redirects
- [BUG] Bash tool hangs indefinitely with pipe characters in commands
- [BUG] Claude Code Misinterprets <error> and <name> XML Tags When Reading Files
- Allow running sub-processes under a Seatbelt policy
- WebFetch summarizer emits <system-reminder> blocks in tool output, violating harness trust boundary
- Allow opting out of the per-Read "malware" system-reminder — defense is porous and cost is non-consensual
- Resume hangs indefinitely on sessions with large tool outputs