EISDIR: illegal operation on a directory — page 3
A directory path was handed to a call that expects a file.
495 issues · 9 open · 486 resolved (98%) · first seen Apr 24, 2025
First response · local
What to do now
Most likely: The operating system rejected a file or directory operation.
- Read the first path in the error and verify that it exists and belongs to the expected user.
- Check disk space, permissions, symlinks and whether another process holds the target.
- Reproduce with the smallest affected path before changing project-wide permissions.
These are conservative triage steps, not an official Anthropic fix. Use the issue and workaround evidence below before making a destructive configuration change.
Is this getting better or worse?
This class of problem is converging. 6 new reports in the last 90 days vs 181 in the 90 before — -97%. The open backlog peaked at 80 in 2025-12 and sits at 9 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
304 of these reports name the Claude Code build they were running, spanning 95 releases. Heaviest reporters:
- Claude Code v2.0.76
- Claude Code v2.1.119
- Claude Code v2.1.92
- Claude Code v2.1.114
- Claude Code v2.1.81
- Claude Code v2.0.73
- Claude Code v2.1.118
- Claude Code v2.1.71
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 4 days across 486 closures. Of the 486 closures with a recorded reason, 13% were closed as completed and 422 as not-planned or duplicate. The most recent completed fix landed in 2026-08.
Workarounds reported by the community
18 of these issues post a workaround someone says worked. The highest-engagement ones:
- [Bug] Anthropic API Error: Internal Server Error (500)
- This error appears to occur with Sonnet. I was able to work around it by using Haiku or Opus with the /model command.
Found in the thread of #11827 · resolved - [Bug] Multiple Claude Code instances freeze on macOS due to file locking conflict
- …lock warning) if useful for the team's investigation. Local workaround: Smart launcher that detects concurrent sessions and uses with a minimal MCP config for secondary sessions, reducing startup surface.
Found in the thread of #13287 · still open - [Bug] Weekly usage reset time displayed incorrectly for Pro plan users
- …gent: While our Support team is unable to manually reset or work around usage limits, you can learn about best practices here ( If you’ve hit a message limit, you’ll need to wait until the reset time, or you can consider…
Found in the thread of #51222 · resolved - Bug: /resume command shows incomplete session list
- …ct directory 4. Many sessions will be missing from the list Workaround: Resume by direct session ID: claude --resume <session-id Suggested fix: Either rebuild the index from .jsonl files when /resume is called, or remove…
Found in the description of #23302 · resolved - [Bug] Status line disappears during "accept edits?" prompt
- …{ "type": "command", "command": "~/.claude/statusline.sh" } Workaround: None available - CLI controls when statusline command is invoked Impact: User loses visibility into context usage, cost, and session stats during in…
Found in the description of #20002 · resolved
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 201–300
Ranked by community engagement (reactions weighted over comments). Listing the top 300 of 495 matches; the statistics above cover all 495.
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant messages
- [Bug] Compaction functionality not working
- [Bug] Stop hook hangs for 1-2 minutes after response completion, blocking subsequent messages
- [Feature Request] Provide claude-3-5-opus model option or revert to previous version
- [Bug] Claude loses context of previous question when asked to simplify explanation
- [Feature Request] Implement background message compaction to prevent workflow interruptions
- [Bug] Claude Opus model compatibility issue
- [Feature Request] Restore cmd+delete to delete single line instead of entire prompt
- [Bug] Build status reports success for failed builds
- [Bug] Auto-Compact triggers prematurely at 24% token usage
- [Bug] Content filtering policy blocking legitimate repository setup operations
- [Feature Request] Add ability to import existing plans across chat sessions
- Desktop app on Windows should execute natively in WSL for projects on \wsl.localhost\ paths
- [Bug] Session Usage Data Disappears After New Session Initialization
- [Bug] Feedback buffer reserved even when auto-compact disabled
- [Bug] Non-Opus models consume tokens when only Opus 4.6 specified
- I need a bug report or feature request description to generate a GitHub issue title. Please provide details about the problem you'd like to report.
- [Bug] Memory write returns corrupted/garbled output
- [Feature Request] Add option to disable survey prompt
- [Bug] Claude Code generates task graphs with lost reasoning context and ignores provided implementation details
- You've hit your limit · resets 4pm (Asia/Tbilisi)
- [Bug] Excessive token consumption and quality degradation in recent builds
- [Bug] Remote control causes Claude to freeze in infinite loop
- [Bug] Claude loses tool awareness and attempts nonsensical tools on private repositories
- [Bug] Rate limit reset time changed without user action
- [Bug] Low quality code analysis and diagnosis accuracy compared to Cursor Auto model
- [Bug] Agent modifies code in plan mode
- I don't have a bug report to work with. Please provide the bug report details so I can generate an appropriate GitHub issue title.
- You've hit your limit · resets 4pm (America/Chicago)
- [Bug] Audio processing latency: iOS 300ms and Whisper 1s delays
- [Bug] API response quality degradation and rate limiting issues
- [Bug] Unexpected high token usage on minimal "continue" command
- [Feature Request] Cross-directory session resumption support
- [Bug] Sonnet 4.6 model regression: internal details leakage before failure
- [Feature Request] Add optional mode for suggesting tests and best practices
- [Feature Request] Fix footer input field to remain visible when scrolling up in chat history
- [Bug] Agent writes files while in plan mode
- [Feature Request] Support fetching claude.ai shared conversation links in Claude Code
- [Feature Request] Built-in Slash Command Descriptions Should Support Localization
- [Feature Request] Add internationalization (i18n) support for CLI interface
- [Bug] Session name lost when executing plan with "clear context and bypass permissions"
- [Bug] Excessive token consumption for single file modifications
- [Feature Request] Add support for license compatibility queries in code analysis
- [Bug Report] Unable to process request - insufficient context provided
- [Bug] Claude Code does not display proposed plan/changes before implementation
- Bug Report
- [Bug] Text truncation and character corruption when pasting from documents
- [Bug] Anthropic API Rate Limit: Extended Throttling (1+ hours)
- [Bug] Claude Code memory bleeding across unrelated project sessions
- [Bug] Auto-compact not triggering despite being enabled, causing context limit errors
- [Feature Request] Add Japanese language support for status messages
- [Feature Request] Add ability to comment on file edit diffs without closing modal
- [Bug] Claude 4.6 Opus ignores explicit user rules and instructions
- [Bug] Context limit reached error triggers prematurely before auto-compact activates
- [Bug] Anthropic API Error: 400 Bad Request with extended thinking enabled
- [Bug] Context limit reached during plan writing after extended exploration
- [Bug] Anthropic API Error: Conversation too long during compaction
- [Bug] Agent unable to read PDF schematics and technical documentation
- [Bug] Image pasting causes context limit errors and breaks session recovery
- [Bug] Claude fails to complete all tasks in multi-section instruction files on first pass
- [Bug] Tool permissions silently denied without user prompt despite allow-list configuration
- [Bug] Claude Code returning responses from unrelated context with non-existent filenames
- [Bug] Claude Code ignores explicit user instructions and modifies files without permission
- [Bug] Conversation too long error during compaction prevents continuation
- [Bug] PostToolUse functionality completely non-functional
- [Bug] Background agents fail with "classifyHandoffIfNeeded is not defined" after completing work (v2.1.37)
- [Feature Request] Separate command history per Claude Code instance in multi-terminal sessions
- [Bug] Auto-Compact not triggering before context overflow
- [Bug] File modifications occurring during plan mode
- I am getting this in all of my requests: API Error: 500...
- Bug Report
- [Bug] Subagents complete multiple times causing duplicate processing and redundant summaries
- [Bug] Anthropic API Error: Cannot modify thinking blocks in assistant messages
- [Bug] Login blocked: CSP frame-ancestors directive violation for claude.ai domain
- [Bug] Anthropic API Error: Cannot Modify Thinking Blocks in Assistant Messages
- [Bug] claude-in-chrome MCP connection failure after Chrome force quit and system memory error
- [Bug] Anthropic API Error: Stream Connection Stalls with Zero Token Output
- [Bug] Clipboard Paste Truncation: Partial Text Copy from Notepad
- [Feature Request] Expose quota/rate limit data to status line JSON
- [Bug] Background task output files persist across sessions and become stale
- [Bug] No cursor displayed in feedback input box
- [Bug] PreToolUse hook blockOnFailure bypassed by dangerouslyDisableSandbox parameter
- [Bug] Inspector.cs Main() entry point prevents Program.cs execution in dotnet run
- [Bug] Stale build artifact served despite recent refactors - build cache or process mismatch
- [Bug] Auto-compression misinterprets context references as approval to continue coding
- [Bug] Agent ignores bash execution restrictions and config settings
- [Bug] Bash command context overflow: `! ls -la` in large directory exceeds token limit
- [Bug] Temporary files accumulating in temp directory with only directory paths
- [BUG] EISDIR - "illegal operation on a directory, read"
- [Bug] Session URL displayed incorrectly in /tasks command
- [Bug] Bug report screen exits on "n" key press
- [Bug] Claude Code executes shell commands 'rm' without permission verification
- [BUG] Intermittent EISDIR error when spawning subagents via Task tool
- [Bug] Status line displays incorrect model across multiple Claude Code sessions
- [Bug] Image size validation prevents file upload with unclear error message
- [Bug] Windows Auto-Compact triggers prematurely with "Conversation too long" error
- [Bug] Clickable file link in tool output resolves to wrong file when duplicate filenames exist
- [Feature Request] Display session name in statusline for multi-instance identification
- [Bug] Claude misinterprets component-specific prompts, removes unintended similar UI elements
- [Bug] Claude falls back to lazy implementation patterns, ignoring explicit instructions and stubbing features instead of completing work