Strategic Compact

1---2name: strategic-compact3description: Suggests manual context compaction at logical intervals to preserve context through task phases rather than arbitrary auto-compaction.4origin: ECC5---6 7# Strategic Compact Skill8 9Suggests manual `/compact` at strategic points in your workflow rather than relying on arbitrary auto-compaction.10 11## When to Activate12 13- Running long sessions that approach context limits (200K+ tokens)14- Working on multi-phase tasks (research → plan → implement → test)15- Switching between unrelated tasks within the same session16- After completing a major milestone and starting new work17- When responses slow down or become less coherent (context pressure)18 19## Why Strategic Compaction?20 21Auto-compaction triggers at arbitrary points:22- Often mid-task, losing important context23- No awareness of logical task boundaries24- Can interrupt complex multi-step operations25 26Strategic compaction at logical boundaries:27- **After exploration, before execution** — Compact research context, keep implementation plan28- **After completing a milestone** — Fresh start for next phase29- **Before major context shifts** — Clear exploration context before different task30 31## How It Works32 33The `suggest-compact.js` script runs on PreToolUse (Edit/Write) and:34 351. **Tracks tool calls** — Counts tool invocations in session362. **Threshold detection** — Suggests at configurable threshold (default: 50 calls)373. **Periodic reminders** — Reminds every 25 calls after threshold38 39## Hook Setup40 41Add to your `~/.claude/settings.json`:42 43```json44{45  "hooks": {46    "PreToolUse": [47      {48        "matcher": "Edit",49        "hooks": [{ "type": "command", "command": "node ~/.claude/skills/strategic-compact/suggest-compact.js" }]50      },51      {52        "matcher": "Write",53        "hooks": [{ "type": "command", "command": "node ~/.claude/skills/strategic-compact/suggest-compact.js" }]54      }55    ]56  }57}58```59 60## Configuration61 62Environment variables:63- `COMPACT_THRESHOLD` — Tool calls before first suggestion (default: 50)64 65## Compaction Decision Guide66 67Use this table to decide when to compact:68 69| Phase Transition | Compact? | Why |70|-----------------|----------|-----|71| Research → Planning | Yes | Research context is bulky; plan is the distilled output |72| Planning → Implementation | Yes | Plan is in TodoWrite or a file; free up context for code |73| Implementation → Testing | Maybe | Keep if tests reference recent code; compact if switching focus |74| Debugging → Next feature | Yes | Debug traces pollute context for unrelated work |75| Mid-implementation | No | Losing variable names, file paths, and partial state is costly |76| After a failed approach | Yes | Clear the dead-end reasoning before trying a new approach |77 78## What Survives Compaction79 80Understanding what persists helps you compact with confidence:81 82| Persists | Lost |83|----------|------|84| CLAUDE.md instructions | Intermediate reasoning and analysis |85| TodoWrite task list | File contents you previously read |86| Memory files (`~/.claude/memory/`) | Multi-step conversation context |87| Git state (commits, branches) | Tool call history and counts |88| Files on disk | Nuanced user preferences stated verbally |89 90## Best Practices91 921. **Compact after planning** — Once plan is finalized in TodoWrite, compact to start fresh932. **Compact after debugging** — Clear error-resolution context before continuing943. **Don't compact mid-implementation** — Preserve context for related changes954. **Read the suggestion** — The hook tells you *when*, you decide *if*965. **Write before compacting** — Save important context to files or memory before compacting976. **Use `/compact` with a summary** — Add a custom message: `/compact Focus on implementing auth middleware next`98 99## Token Optimization Patterns100 101### Trigger-Table Lazy Loading102Instead of loading full skill content at session start, use a trigger table that maps keywords to skill paths. Skills load only when triggered, reducing baseline context by 50%+:103 104| Trigger | Skill | Load When |105|---------|-------|-----------|106| "test", "tdd", "coverage" | tdd-workflow | User mentions testing |107| "security", "auth", "xss" | security-review | Security-related work |108| "deploy", "ci/cd" | deployment-patterns | Deployment context |109 110### Context Composition Awareness111Monitor what's consuming your context window:112- **CLAUDE.md files** — Always loaded, keep lean113- **Loaded skills** — Each skill adds 1-5K tokens114- **Conversation history** — Grows with each exchange115- **Tool results** — File reads, search results add bulk116 117### Duplicate Instruction Detection118Common sources of duplicate context:119- Same rules in both `~/.claude/rules/` and project `.claude/rules/`120- Skills that repeat CLAUDE.md instructions121- Multiple skills covering overlapping domains122 123### Context Optimization Tools124- `token-optimizer` MCP — Automated 95%+ token reduction via content deduplication125- `context-mode` — Context virtualization (315KB to 5.4KB demonstrated)126 127## Related128 129- [The Longform Guide](https://x.com/affaanmustafa/status/2014040193557471352) — Token optimization section130- Memory persistence hooks — For state that survives compaction131- `continuous-learning` skill — Extracts patterns before session ends