CLI session unexpectedly resets when spawning 3 parallel Agent subagents

Description

While using Claude Code CLI in the terminal (NOT the desktop app), my session unexpectedly reset/dropped in the middle of an active debugging session. There was no context limit warning — the conversation simply restarted as if /clear was called.

Environment

• Claude Code CLI (terminal, not desktop app)
• macOS Darwin 25.3.0
• Model: Opus 4.6 (1M context)
• Shell: zsh

Steps to reproduce

1. Start a conversation with several tool calls (Slack MCP read, git commands)
2. In a single response, spawn 3 parallel Agent subagents (all subagent_type=Explore) simultaneously
3. The session drops/resets — all prior conversation context is lost

What happened

• I was in the middle of a multi-step debugging workflow (reading Slack threads, checking PRs, inspecting code)
• When 3 parallel agents were launched in a single response, the session reset
• All prior context was lost — the next message appeared as a fresh conversation
• The [Request interrupted by user for tool use] markers appeared in the conversation history, suggesting the agents were killed mid-flight

Expected behavior

The session should survive parallel agent spawning. If there's a resource limit, it should gracefully degrade or warn — not silently reset the entire session.

Impact

• High — lost all debugging context mid-investigation
• No crash logs found in ~/Library/Logs/DiagnosticReports/, dmesg, or ~/.claude/
• No way to recover the previous session

Additional context

• No OOM kills visible in dmesg
• No Claude-specific crash logs found anywhere on the system
• The CLI appears to not persist any crash/error logs that could aid debugging