history-pruner

# history-pruner - Conversation History Cleanup

Automatically prunes old Claude Code conversation history files to prevent them from growing too large.

## When to Use This Skill

Invoke this skill when:
- User mentions disk space issues with Claude Code
- User asks to clean up or prune conversation history
- Project files in `~/claudesync/projects/` exceed 50MB
- User wants to do maintenance on Claude Code storage
- Before syncing if files are getting too large for GitHub

## How It Works

Project files (`~/claudesync/projects/**/*.jsonl`) store full conversation history including:
- Messages (keep)
- Summaries (keep)
- Tool results with full output (PRUNE - this is the bulk of the data)
- Thinking blocks (prune old ones)
- File history snapshots (prune old ones)

## Pruning Strategy

### Conservative (default)
- Keep last 7 days of full history
- For older entries: truncate `tool_result` content to 500 chars
- Keep all summaries and message structure intact

### Aggressive
- Keep last 3 days of full history
- For older entries: remove `tool_result` content entirely (keep metadata)
- Remove thinking blocks older than 7 days
- Remove file-history-snapshots older than 14 days

## Commands

### Check current sizes
```bash
du -sh ~/claudesync/projects/-Users-andrewwilkinson/*.jsonl | sort -hr | head -10
```

### Run pruning script
```bash
~/claudesync/skills/history-pruner/prune-history.sh [conservative|aggressive]
```

### Dry run (see what would be pruned)
```bash
~/claudesync/skills/history-pruner/prune-history.sh --dry-run
```

## Manual Pruning Steps

If the script isn't available, prune manually:

1. **Identify large files:**
   ```bash
   du -sh ~/claudesync/projects/**/*.jsonl | sort -hr | head -5
   ```

2. **Backup before pruning:**
   ```bash
   cp large-file.jsonl large-file.jsonl.bak
   ```

3. **Prune with jq (truncate old tool results):**
   ```bash
   # Keep entries from last 7 days, truncate older tool_results
   CUTOFF=$(date -v-7d +%s)000
   jq -c '