[BUG] Runaway Ripgrep Processes with Malformed Glob Patterns

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Similarity to Other Reports

Similar 'ripgrep' related bugs have been reported, but they do not consist of the following seen issues I encountered while using the VS Code extension and working on my hyprland config files (and other related configs) in my home directory. I felt it prudent to report the following for additional context in case this is helpful for the ongoing work on fixes.

I would like to also point out: this is my first bug report. Hopefully I'm doing this correctly! I'd appreciate advice or feedback for future reports I make. Thanks!

The following was generated by Claude Code after it analyzed the source of the recurring runaway rg processes.

Summary

Claude Code extension spawns runaway rg (ripgrep) processes that search the entire filesystem from root, causing severe system performance degradation (load average >140, 0% CPU idle).

Additional Context

• Bug occurred during a long conversation about Hyprland configuration optimization
• Multiple file references were made to paths like:
• /home/user /.config/hypr/hyprland.conf
• /home/user /.config/hypr/hyprlock.conf
• /home/user /.config/waybar/style.css
• /home/user /.config/foot/foot.ini
• Each referenced file spawned 1-2 runaway ripgrep processes
• Issue persisted across multiple restarts of VS Code
• Killing all rg processes immediately restored system responsiveness (load dropped from 147 to <1, CPU idle went from 0% to 96%)

Reproducibility

High - Occurs consistently when:

• Long conversations with many file references
• Files referenced using absolute paths
• Files located outside typical code directories (e.g., dotfiles in ~/.config/)

What Should Happen?

Expected Behavior

• Ripgrep should search only within the workspace/project directory
• File indexing should use correct glob patterns relative to workspace root
• System should remain responsive during background indexing

Actual Behavior

• Multiple ripgrep processes spawn and never complete
• Each process runs at 200-300% CPU usage continuously
• System load average climbs to 140+ (on 8-16 core system)
• System becomes completely unresponsive (lag in all applications, even typing)
• Processes must be manually killed with pkill -9 rg

Root Cause Analysis

Malformed Glob Patterns

The ripgrep processes are invoked with glob patterns that start with /, causing them to search from filesystem root:

# Example of malformed command (INCORRECT):
/opt/visual-studio-code/resources/app/node_modules/@vscode/ripgrep/bin/rg \
  --files --hidden --case-sensitive --no-require-git \
  -g /.config/hypr/hyprland.conf \    # ❌ Leading slash causes root search
  -g !**/.git -g !**/.svn -g !**/.hg \
  --no-ignore --follow --no-config --no-ignore-global

What Should Happen

Glob patterns should be relative to workspace or use **/ prefix:

# Correct patterns:
-g .config/hypr/hyprland.conf        # Relative to workspace
-g **/.config/hypr/hyprland.conf     # Match anywhere in workspace tree

Error Messages/Logs

## System Logs
No relevant errors in journalctl. Processes appear to be "working as intended" from their perspective - they're searching the entire filesystem because that's what the malformed glob pattern requested.

### Process List During Bug

$ ps aux | grep "rg " | head -5
user       70835   48256 99 10:57 ?  00:17:54 /opt/visual-studio-code/.../rg --files ... -g /.config/hypr/hyprland.conf ...
user       71506   48256 99 10:58 ?  00:07:51 /opt/visual-studio-code/.../rg --files ... -g /.config/hypr/hyprpaper.conf ...
user       71521   48256 99 10:58 ?  00:09:53 /opt/visual-studio-code/.../rg --files ... -g /.config/hypr/hyprlock.conf ...
user       71688   48256 99 10:59 ?  00:05:34 /opt/visual-studio-code/.../rg --files ... -g /.config/hypr/hyprlock.conf ...
user       71776   48256 99 10:59 ?  00:04:49 /opt/visual-studio-code/.../rg --files ... -g /.config/hypr/hyprlock.conf ...


### System Impact

$ top -b -n 1 | head -12
top - 10:44:22 up 1 day, 34 min,  2 users,  load average: 147.80, 124.31, 82.44
Tasks: 394 total, 3 running, 390 sleep, 1 d-sleep, 0 stopped, 0 zombie
%Cpu(s):  4.9 us, 94.1 sy,  0.0 ni,  0.0 id,  0.0 wa,  1.0 hi,  0.0 si,  0.0 st
MiB Mem :  22834.8 total,    255.9 free,  12429.3 used,  10408.5 buff/cache

    PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
  65179 user      20   0   95180  49352   3244 S 777.4   0.2   6:52.52 rg
  60765 user      20   0   92620  52196   3244 S 210.5   0.2  14:47.79 rg
  61664 user      20   0   95180  60920   3244 S  97.2   0.3  16:18.19 rg
  60459 user      20   0   92620  64212   3244 S  81.0   0.3  23:57.63 rg


**Critical**: Load average of 147 with 0% CPU idle, system completely unusable.

### Process Tree

$ pstree -p | grep "rg("
code(48256)─┬─rg(49097)─┬─{rg}(49098)
            ├─rg(59971)─┬─{rg}(59980)
            ├─rg(60391)─┬─{rg}(60392)
            ├─rg(60459)─┬─{rg}(60460)
            ├─rg(60765)─┬─{rg}(60766)
            ... (15+ total rg processes)


All processes are children of VS Code (PID 48256).

Steps to Reproduce

Trigger Mechanism

The bug appears to trigger when:

1. Claude Code assistant references files using absolute paths in responses
2. VS Code's file indexer attempts to make these files "clickable" for navigation
3. The indexer converts absolute paths to glob patterns incorrectly
4. Leading / in paths like /home/user/.config/... becomes -g /.config/...
5. Ripgrep interprets this as "search from filesystem root for paths matching /.config/..."

Steps to Reproduce

1. Have an extended conversation with Claude Code that references configuration files using absolute paths
2. Use markdown file references in responses (e.g., [hyprlock.conf](file:///home/user/.config/hypr/hyprlock.conf))
3. Allow conversation to continue with multiple file references
4. Monitor system processes: ps aux | grep rg
5. Observe multiple rg processes spawned, each consuming 200-300% CPU

Claude Model

Sonnet (default)

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.0.8

Platform

Anthropic API

Operating System

Other Linux

Terminal/Shell

VS Code integrated terminal

Additional Information

Environment

• OS: Linux (Arch-based, kernel 6.16.10-arch1-1)
• VS Code Version: Latest (visual-studio-code package)
• Claude Code Extension: Latest version installed
• Ripgrep Location: /opt/visual-studio-code/resources/app/node_modules/@vscode/ripgrep/bin/rg
• Working Directory: /home/user
• Session Type: Hyprland (Wayland compositor)

Impact Severity

Critical - System becomes completely unusable:

• User cannot type (input lag in all applications)
• Applications freeze or respond with multi-second delays
• Hard reboot sometimes required if processes spawn faster than they can be killed
• Data loss risk from forced reboots

Workaround

EDIT: previous scripts were failing to find the runaway processes, required better
logic to find them; I run this as a cron task every 3 minutes currently to hunt them
down. Seems to be working for now.

EDIT2: better discovery of rg processes and dropped threshold down to 2 minutes for
runaway detection

#!/usr/bin/env bash
# Kills runaway ripgrep processes from VS Code
# This is a workaround for VS Code indexer bugs

# Find VS Code rg PIDs by full binary path
RG_PIDS=$(pgrep -f "/opt/visual-studio-code/resources/app/node_modules/@vscode/ripgrep/bin/rg")

if [ -z "$RG_PIDS" ]; then
    echo "No VS Code ripgrep processes found"
    echo "Current load:$(uptime | awk -F'load average:' '{print $2}')"
    exit 0
fi

# Check each PID's runtime
RUNAWAY_RG=""
for pid in $RG_PIDS; do
    # Get elapsed time for this PID
    etime=$(ps -p "$pid" -o etime= 2>/dev/null | tr -d ' ')
    if [ -z "$etime" ]; then
        continue
    fi

    # Parse time (MM:SS or HH:MM:SS or DD-HH:MM:SS)
    if [[ "$etime" == *"-"* ]]; then
        # Has days (DD-HH:MM:SS)
        IFS='-:' read -ra time <<< "$etime"
        minutes=$(( 10#${time[0]} * 24 * 60 + 10#${time[1]} * 60 + 10#${time[2]} ))
    else
        # Split by colon
        IFS=':' read -ra time <<< "$etime"
        if [ ${#time[@]} -eq 3 ]; then
            # HH:MM:SS
            minutes=$(( 10#${time[0]} * 60 + 10#${time[1]} ))
        elif [ ${#time[@]} -eq 2 ]; then
            # MM:SS - force base 10 to avoid octal interpretation
            minutes=$((10#${time[0]}))
        else
            # Unknown format, skip
            continue
        fi
    fi

    # If running more than 2 minutes, add to kill list
    if [ "$minutes" -gt 2 ]; then
        RUNAWAY_RG="$RUNAWAY_RG $pid"
    fi
done

if [ -n "$RUNAWAY_RG" ]; then
    COUNT=$(echo "$RUNAWAY_RG" | wc -l)
    echo "Found $COUNT runaway ripgrep process(es), killing..."
    echo "$RUNAWAY_RG" | xargs kill -9 2>/dev/null
    echo "✓ Killed runaway rg processes"
else
    echo "No runaway rg processes found"
fi

# Show current system load
echo "Current load:$(uptime | awk -F'load average:' '{print $2}')"