strategic-compact

Strategic Compact Skill

This skill helps maintain context coherence by suggesting manual compaction at strategic points during long coding sessions.

Overview

AI coding assistants have limited context windows. When auto-compaction occurs mid-task, critical context can be lost. This skill tracks session activity and suggests compaction at logical phase transitions rather than arbitrary points.

Why Strategic Compaction?

When to Activate

This skill activates via the Claude Code PreToolUse hook when:

• Tool calls reach the configured threshold (default: 50)
• Matches Edit or Write tool operations
• Session is in a stable state for compaction

How It Works

PreToolUse Hook → suggest-compact.sh
                         ↓
              Increment tool counter
                         ↓
              Check against threshold
                         ↓
         (if threshold reached)
                         ↓
    Output: "[StrategicCompact] Consider /compact"

Counter Mechanism

The script maintains a per-session counter in /tmp/claude-tool-count-$$:

1. Initial threshold (50 calls): First suggestion to compact
2. Every 25 calls after: Periodic reminders if context feels stale

# First suggestion at 50 tool calls
[StrategicCompact] 50 tool calls reached - consider /compact if transitioning phases

# Subsequent suggestions every 25 calls
[StrategicCompact] 75 tool calls - good checkpoint for /compact if context is stale

Configuration

Environment Variables

Set custom threshold:

export COMPACT_THRESHOLD=100

Hook Setup

Add to ~/.claude/settings.json:

{
  "hooks": {
    "PreToolUse": [{
      "matcher": "Edit|Write",
      "hooks": [{
        "type": "command",
        "command": "~/.claude/skills/strategic-compact/suggest-compact.sh"
      }]
    }]
  }
}

Matcher Options

Best Practices

Ideal Compaction Points

1. After exploration, before implementation

   • Research phase complete
   • Plan finalized
   • Ready to start coding

2. After completing a milestone

   • Feature implemented
   • Tests passing
   • Ready for next feature

3. When switching contexts

   • Moving to different component
   • Changing focus area
   • Starting new task

Signs Context is Stale

• AI repeating previous explanations
• Confusion about recent changes
• Asking for information already provided
• Inconsistent responses

Output Messages

Related Skills

• continuous-learning/ - Extract patterns from sessions
• tdd-workflow/ - Test-driven development workflow