---

name: history-pruner description: Prunes old Claude Code conversation history to save disk space. Use when project files get large (>50MB) or periodically for maintenance.

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

Copydu -sh ~/claudesync/projects/-Users-andrewwilkinson/*.jsonl | sort -hr | head -10

Run pruning script

Copy~/claudesync/skills/history-pruner/prune-history.sh [conservative|aggressive]

Dry run (see what would be pruned)

Copy~/claudesync/skills/history-pruner/prune-history.sh --dry-run

Manual Pruning Steps

If the script isn't available, prune manually:

1. Identify large files:

   Copydu -sh ~/claudesync/projects/**/*.jsonl | sort -hr | head -5
   

2. Backup before pruning:

   Copycp large-file.jsonl large-file.jsonl.bak
   

3. Prune with jq (truncate old tool results):

   Copy# Keep entries from last 7 days, truncate older tool_results
   CUTOFF=$(date -v-7d +%s)000
   jq -c 'if .timestamp and .timestamp < '$CUTOFF' and .type == "tool_result" then .content = (.content[:500] + "... [truncated]") else . end' file.jsonl > file.pruned.jsonl
   

4. Verify and replace:

   Copywc -l file.jsonl file.pruned.jsonl  # Should have same line count
   mv file.pruned.jsonl file.jsonl
   

Automation

To run weekly cleanup, add to your personal crontab if desired:

Copy# Add to crontab for weekly Sunday 3am cleanup
0 3 * * 0 ~/.claude/plugins/marketplaces/Folly/skills/history-pruner/prune-history.sh conservative

Note: Component-sync runs automatically at session start via the SessionStart hook.

Important Notes

• Always backup before aggressive pruning
• Pruning removes ability to see full old tool outputs
• Summaries are preserved so conversation context remains
• This won't break Claude Code - it handles missing/truncated data gracefully