[BUG] Memory Leaks with Opus 4.6 CLI - massive sudden spikes in RAM/SWAP usage

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?

Environment:

• Claude Code (Max Plan)
• Model: Opus 4.6 (did NOT experience this on Opus 4.5)
• Platform: GCP VM, Ubuntu Linux, 15GB RAM, 16GB swap
• VS Code SSH Remote for terminal access
• Project type: Next.js

Frequency: 4 incidents in ~24 hours

Behavior:
Memory spikes suddenly (not gradual accumulation). Typical pattern:

• Normal operation at ~1GB RAM per session with occasional small spikes
• Issue simple command (e.g., "fix the font size on this badge")
• RAM spikes to 6-14GB within 1 minute (yes within 1 minute, not a typo)
• Status shows "0 tokens used"
• No tool calls visible in output
• Session appears to be "spinning" but never does anything
• Happened within main session as well as subagent session

Recovery:

• Pressing ESC immediately releases the memory IF caught as it's happening
• If not caught in time, swap fills up and VM becomes unresponsive
• First incident required a complete restart of the VM
• Second incident I was able to slowly and painfully access browser-based SSH and kill the Claude PID hogging 25GB of RAM+SWAP
• Third and fourth time I was able to hit ESC as it was happening after I installed a resource monitor

Possible trigger:
In one incident, the CC session self-diagnosed: "Yeah that was the system reminder trying to get me to use task
tools." The <system-reminder> about task tools may be triggering expensive processing.

Key observations:

1. Spike happens BEFORE any action is actually taken by Claude Code (0 tokens)
2. No tool calls executed — bug is in local preprocessing
3. ESC releases memory instantly — suggests active allocation, not cached data
4. Only happening on Opus 4.6, not 4.5
5. All incidents on same Next.js project (not Python/Streamlit project on same VM, but I haven't tried Opus 4.6 on the python project yet)

Impact: VM crashes or becomes unresponsive if not caught quickly. Requires constant memory monitoring to use Claude Code reliably.

What Should Happen?

No good reason for Claude Code to hog 10GB+ RAM/SWAP within 1 minute without actually doing anything.

Error Messages/Logs

Steps to Reproduce

Unfortunately I can't replicate this on demand. It's happening randomly within a regular working session. This morning it happened after a fresh start (session had been compacted and restarted) on the second command.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

Version: 2.1.37

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

VS Code integrated terminal

Additional Information

_No response_