Claude Code Settings & Configuration Maintenance

Claude Code Settings & Configuration Maintenance

Purpose: Best practices for maintaining Claude Code settings.json and agent files to avoid validation errors and ensure proper configuration.

🚨 Critical: Always Consult Official Documentation

MANDATORY PROTOCOL: When uncertain about configuration format, ALWAYS web search official Claude Code documentation first.

Documentation Search Strategy

1. Use WebFetch tool to retrieve latest official docs

2. Primary documentation URLs:

   • https://code.claude.com/docs/en/ - Main documentation hub
   • https://code.claude.com/docs/en/hooks - Hooks documentation
   • https://code.claude.com/docs/en/agents - Agents documentation
   • https://code.claude.com/docs/en/settings - Settings reference

3. Search pattern:

   WebFetch(url="https://code.claude.com/docs/en/hooks",
            prompt="What is the correct format for hook matchers?")
   

📋 Hooks Configuration Format

✅ Correct Format (String Matchers)

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "echo 'Running pre-tool hook'",
            "description": "Example hook"
          }
        ]
      },
      {
        "matcher": "Write",
        "hooks": [
          {
            "type": "command",
            "command": "echo 'Before write operation'",
            "description": "Pre-write hook"
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "command": "echo 'After bash command'",
            "description": "Post-bash hook"
          }
        ]
      }
    ],
    "UserPromptSubmit": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "echo 'User submitted prompt'",
            "description": "Prompt submission hook"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "echo 'Session stopping'",
            "description": "Stop hook"
          }
        ]
      }
    ]
  }
}

❌ Incorrect Format (Object Matchers - OLD FORMAT)

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": {"tools": ["*"]},  // ❌ WRONG - Object format
        "hooks": [...]
      },
      {
        "matcher": {"tools": ["Write"]},  // ❌ WRONG - Object format
        "hooks": [...]
      }
    ]
  }
}

Matcher Types

Matcher Pattern	Description	Example
"*"	Match all tools	"matcher": "*"
"Write"	Match specific tool	"matcher": "Write"
`"Edit	Write"`	Match multiple tools (regex)
"Bash(git:*)"	Match specific bash commands	"matcher": "Bash(git:*)"
""	Empty matcher (for non-tool hooks)	"matcher": ""

Note: Matcher patterns accept raw regular expressions. Use the pipe (|) for alternation without escaping (e.g., "Edit|Write").

Hook Event Types

• PreToolUse: Runs before tool execution (requires matcher)
• PostToolUse: Runs after tool execution (requires matcher)
• UserPromptSubmit: Runs when user submits prompt (use empty matcher "")
• SessionStart: Runs at session start (use empty matcher "")
• Stop: Runs when session stops (use empty matcher "")

🤖 Agent File Frontmatter Format

✅ Correct Format (Unquoted Values)

---
name: my-agent
description: A specialized agent for specific tasks with detailed expertise
---

# Agent Content
Your agent instructions here...

❌ Incorrect Format (Quoted Values)

---
name: "my-agent"  # ❌ WRONG - Quoted
description: "A specialized agent..."  # ❌ WRONG - Quoted
---

Required Frontmatter Fields

Field	Required	Format	Example
name	✅ Yes	Unquoted string	name: code-review
description	✅ Yes	Unquoted string	description: Expert code reviewer

Agent Naming Best Practices

• Use kebab-case: code-review, test-runner, security-audit
• Be descriptive: Name should indicate agent's purpose
• Avoid generic names: Prefer python-test-runner over tester
• No quotes: YAML values should be unquoted

🔍 Validation Protocol

1. Use /doctor Command

ALWAYS run /doctor after configuration changes:

/doctor

Expected clean output:

✅ Diagnostics
 └ Currently running: npm-global (2.0.43)
 └ Settings: Valid
 └ Agents: All parsed successfully
 └ Hooks: All registered correctly

2. Common Validation Errors

Error	Cause	Fix
matcher: Expected string, but received object	Using {"tools": [...]} format	Change to string: "*" or "Write"
Missing required "description" field	Agent frontmatter missing description	Add description: ... to frontmatter
Missing required "name" field	Agent frontmatter missing name	Add name: ... to frontmatter
Invalid frontmatter	Quoted values in YAML	Remove quotes from name/description

3. Pre-Commit Checklist

Before committing settings changes:

• Run /doctor to validate configuration
• Check hooks section uses string matchers
• Verify all agent files have required frontmatter
• Ensure agent frontmatter uses unquoted values
• Test hooks execute correctly (if applicable)

🛠️ Troubleshooting Workflow

Issue: Hooks Not Working

1. Check matcher format: Ensure using string matchers, not objects
2. Verify hook syntax: Confirm JSON structure is valid
3. Test command: Run hook command manually to ensure it works
4. Check permissions: Ensure hook script files are executable

Issue: Agent Parse Errors

1. Check frontmatter: Verify both name and description fields present
2. Remove quotes: Ensure values are unquoted (YAML format)
3. Validate YAML: Ensure frontmatter block starts/ends with ---
4. Check indentation: YAML is indent-sensitive (use spaces, not tabs)

Issue: Settings Not Loading

1. Validate JSON: Use jq or JSON validator to check syntax
2. Check file location: Ensure settings.json is in correct directory
   • Global: ~/.claude/settings.json
   • Project: <project>/.claude/settings.json
3. Restart Claude Code: Configuration changes may require restart

📚 Documentation Reference Quick Links

Topic	URL
Hooks	https://code.claude.com/docs/en/hooks
Agents	https://code.claude.com/docs/en/agents
Settings	https://code.claude.com/docs/en/settings
MCP Servers	https://code.claude.com/docs/en/mcp
Permissions	https://code.claude.com/docs/en/permissions

🎯 Best Practices Summary

1. Always consult official docs when uncertain about format
2. Use string matchers for hooks (not object format)
3. Use unquoted values in agent frontmatter
4. Run /doctor after every configuration change
5. Test hooks manually before committing
6. Keep settings.json valid - use JSON validator
7. Document custom configurations in project README
8. Version control all .claude/ directory files
9. Use descriptive names for agents and hooks
10. Follow principle of least privilege for permissions

⚠️ Common Pitfalls to Avoid

Pitfall	Impact	Prevention
Using old object matcher format	Hooks fail validation	Always use string matchers
Quoting agent frontmatter values	Agent parse errors	Use unquoted YAML values
Missing description field	Agent not loaded	Always include name + description
Invalid JSON syntax	Settings not loaded	Validate JSON before commit
Not running /doctor	Deploy with broken config	Run /doctor before every commit

🔄 Migration Guide: Old to New Hook Format

Step 1: Identify Old Format

Search for object matchers:

grep -n '"matcher": {' .claude/settings.json

Step 2: Convert to String Format

Old:

"matcher": {"tools": ["*"]}
"matcher": {"tools": ["Write"]}
"matcher": {"tools": ["Bash(git:*)"]}

New:

"matcher": "*"
"matcher": "Write"
"matcher": "Bash(git:*)"

Step 3: Handle Non-Tool Hooks

For UserPromptSubmit, Stop, SessionStart:

{
  "UserPromptSubmit": [
    {
      "matcher": "",  // Empty string for non-tool hooks
      "hooks": [...]
    }
  ]
}

Step 4: Validate

/doctor

📝 Example: Complete Valid Configuration

{
  "env": {
    "BASH_MAX_OUTPUT_LENGTH": "5000"
  },
  "permissions": {
    "allow": ["Bash(git:*)"]
  },
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "echo 'Pre-tool execution'",
            "description": "Log before tool use"
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "Write",
        "hooks": [
          {
            "type": "command",
            "command": "echo 'File written'",
            "description": "Log file writes"
          }
        ]
      }
    ],
    "UserPromptSubmit": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "echo 'Prompt submitted'",
            "description": "Log prompt submission"
          }
        ]
      }
    ]
  }
}

🚀 When to Web Search Official Docs

ALWAYS search official docs when:

• Unsure about configuration format
• Encountering validation errors
• Implementing new features
• Migrating from old formats
• Debugging hook execution issues
• Adding new agent types
• Configuring MCP servers
• Setting up permissions

Search Example:

WebFetch({
  url: "https://code.claude.com/docs/en/hooks",
  prompt: "What is the correct format for hook matchers? Show examples."
})

✅ Success Criteria

Configuration is correct when:

• /doctor shows no errors
• All hooks execute as expected
• All agents parse successfully
• Settings.json is valid JSON
• Agent frontmatter is valid YAML
• Matchers use string format (not objects)
• Frontmatter values are unquoted

---

Last Updated: 2025-11-17 Applies To: Claude Code 2.0+ Official Docs: https://code.claude.com/docs/en/