Full assistant response duplicated 11x in terminal after large Agent (Explore) subagent result

Environment

• Claude Code version: 2.1.86 (updated to 2.1.87 after observing bug)
• Platform: Windows 11 Home 10.0.26200
• Shell: bash (Git Bash)
• Interface: Terminal / CLI
• Model: claude-opus-4-6 (1M context)

Description

The full assistant response (including a large Agent tool result) was duplicated approximately 11 times in the terminal output. The user confirmed this was the full response repeating, not just system-reminder blocks.

Steps to Reproduce

1. Start a new conversation in a project with multiple system-reminders at session start:

• SessionStart hook
• Superpowers plugin (many skills listed)
• CLAUDE.md + MEMORY.md loaded
• Deferred tools list

1. User asks a broad question ("show me all work done on this project")
2. Assistant launches an Agent tool with subagent_type: Explore and a "very thorough" exploration prompt
3. The Explore subagent returns a very large result (~5,000+ lines — full file inventory, line counts, feature lists)
4. Assistant formats and outputs the response to the user

Expected Behavior

The response should appear once in the terminal.

Actual Behavior

The full assistant response appeared approximately 11 times in the terminal, wasting an estimated 50,000+ tokens.

Additional Context

• The user reports this has NOT happened before with other large responses or substantial work sessions in the same project
• The distinguishing factor appears to be the combination of: heavy system-reminder context at session start + very large Agent/Explore subagent result
• This was the first message exchange in the conversation