---

name: julien-cmd-claude-json-cleanup description: "Clean up oversized ~/.claude.json file to prevent token limit errors. Use when Read tool fails with 'exceeds maximum allowed tokens' or to maintain Claude config performance." license: Apache-2.0 metadata: author: "Julien" version: "1.0.0" category: "maintenance" last_updated: "2026-01" triggers:

Keywords

• "claude.json"
• "claude json"
• "token limit"

Action phrases

• "cleanup claude"
• "nettoyer claude"
• "clean config"
• "reduce config size"
• "claude config cleanup"
• "settings cleanup"
• "shrink claude config"
• "réduire la taille de claude"

Problem phrases

• "exceeds maximum"
• "file too large"
• "fichier trop gros"
• "claude.json trop gros"
• "can't read claude.json"
• "token limit exceeded"
• "25000 tokens"
• "config file too big"

---

Claude JSON Cleanup

Manage and clean up oversized ~/.claude.json configuration files.

Observability

First: Display activation message:

Copy🔧 Skill "julien-cmd-claude-json-cleanup" activated

Problem

The ~/.claude.json file stores Claude Code configuration including:

• Project settings per directory
• MCP server configurations
• Conversation memory references
• Session data

Over time, this file can exceed the 25,000 token limit, causing errors like:

CopyError: File content (26095 tokens) exceeds maximum allowed tokens (25000)

Quick Command

Run the cleanup script:

Copynode ~/.claude/scripts/cleanup-claude-json.js

Or from marketplace:

Copynode "$HOME/OneDrive/Coding/_Projets de code/2025.11 Claude Code MarketPlace/skills/julien-cmd-claude-json-cleanup/scripts/cleanup-claude-json.js"

What It Does

The script automatically:

1. Creates backup → ~/.claude.json.backup
2. Removes stale projects → Entries for deleted directories
3. Cleans github paths → Non-existent repo references
4. Reports savings → Before/after size and token estimates

Manual Analysis

If you need to investigate further:

Copy# Check file size
wc -c ~/.claude.json

# Analyze sections by size
node -e "
const fs = require('fs');
const os = require('os');
const d = JSON.parse(fs.readFileSync(os.homedir() + '/.claude.json'));
Object.entries(d).forEach(([k, v]) => {
    const size = JSON.stringify(v).length;
    const type = Array.isArray(v) ? 'array' : typeof v;
    if (type === 'object') console.log(k + ': ' + Object.keys(v).length + ' entries (' + size + ' bytes)');
    else if (type === 'array') console.log(k + ': ' + v.length + ' items (' + size + ' bytes)');
});
"

Restoring Backup

If something goes wrong:

Copycp ~/.claude.json.backup ~/.claude.json

Troubleshooting

Script fails with "ENOENT"

Copy# File doesn't exist - this is normal for fresh installs
ls ~/.claude.json || echo "No config file yet"

Script fails with "EACCES"

Copy# Permission issue - check file ownership
ls -la ~/.claude.json
# Fix permissions if needed
chmod 644 ~/.claude.json

File still too large after cleanup

If projects are all valid but file is still > 25,000 tokens:

1. Check mcpServers section - remove unused servers
2. Check skillUsage section - can be cleared safely
3. Consider moving rarely-used MCP configs to .mcp.json per-project

Restore from backup

Copycp ~/.claude.json.backup ~/.claude.json

Prevention

Add to your workflow:

1. Run cleanup monthly
2. Keep project list tidy (remove old projects)
3. Use .mcp.json per-project instead of global MCP configs

Skill Chaining

Skills Required Before

• None

Input Expected

• None (operates on ~/.claude.json)

Output Produced

• Format: Cleaned ~/.claude.json, console report
• Side effects:
  • Creates backup: ~/.claude.json.backup
• Duration: < 5 seconds

Compatible Skills After

• None (standalone maintenance task)

Called By

• Direct invocation when token limit error occurs
• Periodic maintenance

Tools Used

• Bash (usage: run node script)