Bundled ripgrep binary loses execute permission on macOS (breaks command/skill discovery)

+1

Confirmed on Linux VM running with HyperV on Win11 (Debian 13, x64) — not macOS-only

Same issue here on a fresh npm install -g @anthropic-ai/claude-code@2.1.89 on Debian (ext4 filesystem). The bundled vendor/ripgrep/x64-linux/rg binary is installed with 644 permissions instead of 755.

Symptoms:

• Custom agents, skills and commands silently disappear from /agents menu
• Glob and Grep tools fail with EACCES errors
• No visible error message to the user

Quick fix (breaks after every update):

chmod +x $(dirname $(which claude))/../lib/node_modules/@anthropic-ai/claude-code/vendor/ripgrep/x64-linux/rg

⚠️ This needs to be re-applied after every npm install -g @anthropic-ai/claude-code until the bug is fixed.

Permanent workaround — use system ripgrep:

Install ripgrep system-wide

sudo apt install ripgrep # Debian/Ubuntu
or: brew install ripgrep (macOS)

Tell Claude Code to use it

echo 'export USE_BUILTIN_RIPGREP=0' >> ~/.zshrc # or ~/.bashrc
source ~/.zshrc or source ~/.bashrc

This survives Claude Code updates — no need to chmod +x after every npm install -g.

(found and reproduced by Claude.ai - wow!)

Docker reproduction (100% reliable)

This is trivially reproducible in any Docker container using npm install -g. Every fresh install produces a non-executable rg binary, making this a guaranteed repro — no flaky timing or platform-specific conditions.

Dockerfile

FROM node:20
RUN npm install -g @anthropic-ai/claude-code
RUN useradd -m -s /bin/bash claude
RUN git config --system safe.directory '*'
WORKDIR /workspace
USER claude
ENTRYPOINT ["claude"]

Verify broken permissions

docker build -t claude-cli -f Dockerfile.claude .
docker run --rm --entrypoint bash claude-cli -c \
  "ls -la /usr/local/lib/node_modules/@anthropic-ai/claude-code/vendor/ripgrep/*/rg"

Output:

-rw-r--r-- 1 root root 4658376 ... /usr/local/lib/node_modules/@anthropic-ai/claude-code/vendor/ripgrep/arm64-linux/rg

No execute bit.

Symptoms

• Custom skills from .claude/commands/ don't appear in autocomplete
• Invoking them manually returns Unknown skill: <name>
• No user-facing error — skills silently don't load
• CLAUDE.md, git operations, and everything else work fine, making it very hard to diagnose

Debug evidence

Running with --debug --debug-to-stderr reveals the problem:

[DEBUG] Loading skills from: managed=/etc/claude-code/.claude/skills, user=/home/claude/.claude/skills, project=[]
[DEBUG] Loaded 0 unique skills (0 unconditional, 0 conditional, managed: 0, user: 0, project: 0, additional: 0, legacy commands: 0)
...
[ERROR] Error: spawn .../vendor/ripgrep/arm64-linux/rg EACCES

Internally, baz() → Wr("commands", projectRoot) → OS1() correctly resolves the .claude/commands/ directory and finds the files. But the subsequent gitignore-aware file scan via ripgrep fails with EACCES, the error is caught internally, and an empty array is returned — zero skills loaded, zero diagnostics shown to the user.

Workaround

RUN npm install -g @anthropic-ai/claude-code \
    && chmod +x /usr/local/lib/node_modules/@anthropic-ai/claude-code/vendor/ripgrep/*/rg

Note on silent failure

Even with --debug --debug-to-stderr, the EACCES error only appears much later in the log (associated with plugin refresh timing, not the skill loading phase). The skill scan catches the error and returns [] with no diagnostic at all. A startup warning like ⚠ Bundled ripgrep not executable — skill discovery disabled would have saved hours of investigation.

Environment: Claude Code v2.1.89, Docker with node:20 base image (Debian), arm64-linux.

This was fixed in v2.1.91 — The bundled ripgrep binary now ships with execute permissions on macOS, restoring command and skill discovery. If you're still seeing this in the latest version, please comment with your version and repro and we'll reopen.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.