Enhancement Request: Add --quiet flag to suppress tool call output

Summary

Add a --quiet (or --minimal-output) CLI flag to suppress tool call output in interactive mode, showing only the agent's final responses.

Use Case

Advisory agents that use wake-up protocols to load session state currently produce verbose output that obscures the clean summary users expect:

Current Behavior (verbose):

> wake up

⏺ my-executive-coach-aget v2.5.0 (Advisory)
  Co-Active Coaching • Strategic Thinking • Executive Leadership

⏺ Search(pattern: ".aget/sessions/SESSION_*.md")
  ⎿  Found 1 file (ctrl+o to expand)

⏺ Search(pattern: ".aget/commitments/active.yaml")
  ⎿  Found 1 file (ctrl+o to expand)

⏺ Search(pattern: ".aget/client_progress/*.yaml")
  ⎿  Found 1 file (ctrl+o to expand)

⏺ Read(.aget/sessions/SESSION_2025-10-08_10-00.md)
  ⎿  Read 38 lines

⏺ Read(.aget/commitments/active.yaml)
  ⎿  Read 21 lines

⏺ Read(.aget/client_progress/demo_user.yaml)
  ⎿  Read 23 lines

⏺ 📍 Last session: 2025-10-08 (2 days ago)
  📋 Active commitments: 2
     • Delegate architecture decision (due 10/15)
  📊 Progress: 1 session

  Ready for coaching session.

Desired Behavior (with --quiet):

> wake up

⏺ my-executive-coach-aget v2.5.0 (Advisory)
  Co-Active Coaching • Strategic Thinking • Executive Leadership

  📍 Last session: 2025-10-08 (2 days ago)
  📋 Active commitments: 2
     • Delegate architecture decision (due 10/15)
  📊 Progress: 1 session

  Ready for coaching session.

Problem

• Tool call output adds 6+ lines of noise before the actual response
• Users must scroll past technical details to see the summary
• Particularly problematic for:
  • Demo/presentation scenarios
  • Advisory agents with initialization routines
  • Production use by non-technical users
  • Multi-agent workflows where tool verbosity compounds

Proposed Solution

Add a CLI flag:

claude --quiet           # Suppress tool call output
claude --minimal-output  # Alternative flag name

Behavior:

• Suppress all tool call lines (Search, Read, Glob, etc.)
• Show only agent text responses (text blocks)
• Keep error messages visible (tool failures should still show)
• Apply only to interactive mode (--print mode unaffected)

Configuration:
Could also support:

// ~/.config/claude/settings.json
{
  "outputMode": "quiet"  // or "verbose" (default)
}

Alternatives Considered

1. Agent-level suppression - Not possible; tool output is CLI behavior
2. Custom output formatter - No hook exists for this
3. Accept verbosity - Reduces UX quality for production agents
4. Screenshot workarounds - Manual, doesn't help actual users

Impact

Benefits:

• Cleaner UX for advisory agents
• Better demo/presentation experience
• Reduced cognitive load for non-technical users
• Opt-in (doesn't affect existing workflows)

Risks:

• Users might miss important tool failures (mitigated by showing errors)
• Less transparency about what agent is doing (user's choice via flag)

Related Issues

Similar to #8036 (feedback prompt complaints) - users want control over UI noise.

Environment

• Claude Code v2.0.14
• macOS/Linux (likely affects all platforms)
• Use case: AGET framework advisory agents with internal state management

Would this enhancement be considered? Happy to provide additional context or participate in design discussion.