Configuration & Settings

Pro Tip: Put team conventions in .claude/settings.json (committed to git) and personal preferences in .claude/settings.local.json (gitignored).

🧠 Knowledge Check

Q: If you set model: opus in your user settings and model: sonnet in your project's shared settings, which model will Claude use?

Sonnet. Project settings override user settings. The hierarchy is: Managed > CLI > Local > Shared Project > User. Since shared project settings are more specific than user settings, sonnet wins.

Model Configuration

Available Model Aliases

Switch mid-session:

Set via CLI:

Set via environment:

Effort Level (Opus 4.6)

Controls how deeply Claude reasons before responding:

Gotcha: Effort level only applies to Opus 4.6's adaptive reasoning. Adjust in /model using left/right arrow keys.

✍️ Recall Exercise

Fill in the blanks:

• The hybrid model that uses Opus for planning and Sonnet for execution: ___
• Environment variable to set the default model: ANTHROPIC_____
• To switch models during a session: /___

• opusplan
• ANTHROPIC_MODEL
• /model

Permission Modes

Permission Rules

Fine-grained control over what Claude can do:

Rule Syntax

Important: Rules evaluate in order: deny → ask → allow. Deny always wins.

🧠 Knowledge Check

Q: You have allow: ["Bash(git *)"] and deny: ["Bash(git push *)"] in your settings. What happens when Claude tries to run git push origin main?

The command is denied. Deny rules always take precedence over allow rules, regardless of order. The evaluation chain is: deny → ask → allow.

Environment Variables

Core Variables

Model Override Variables

Prompt Caching

Set Variables in Settings

Extended Thinking

Extended thinking lets Claude reason through complex problems before responding. It's enabled by default.

Press Ctrl+O to toggle verbose mode and see Claude's thinking as gray italic text.

Gotcha: With Opus 4.6, thinking uses adaptive reasoning — the model dynamically allocates thinking tokens based on your effort level. MAX_THINKING_TOKENS is ignored on Opus 4.6 (except when set to 0 to disable entirely).

🧠 Knowledge Check

Q: How do you disable extended thinking entirely on Opus 4.6?

Set MAX_THINKING_TOKENS=0. This is the only value that works on Opus 4.6, since it uses adaptive reasoning and ignores other values. You can also toggle it with Option+T / Alt+T mid-session.

Working Directories

By default, Claude has access to the directory where it was launched. Extend this:

CLI Flags Quick Reference

✍️ Recall Exercise

Fill in the blanks:

• Settings file for personal prefs not committed to git: .claude/settings.___
• Flag to set a spending limit in headless mode: --max-___-usd
• Effort level environment variable: CLAUDE_CODE____LEVEL

• .claude/settings.local.json
• --max-budget-usd
• CLAUDE_CODE_EFFORT_LEVEL

🔨 Try It Yourself

1. Open ~/.claude/settings.json (create it if it doesn't exist)
2. Add your preferred model:

   {
     "model": "sonnet"
   }
   

3. Start a Claude session and run /status to verify the model
4. Try /model opus to switch mid-session
5. Run /cost to compare token usage between models

Reflection: Did you notice a difference in response quality or speed between models?

🤔 What Would You Do?

Scenario: Your team wants everyone to use Sonnet by default and block rm -rf commands, but you personally prefer Opus for complex architecture work.

How would you set this up?

1. Team settings (.claude/settings.json, committed to git):

   {
     "model": "sonnet",
     "permissions": {
       "deny": ["Bash(rm -rf *)"]
     }
   }
   

2. Your personal override (.claude/settings.local.json, gitignored):

   {
     "model": "opus"
   }
   

Local settings override shared project settings, so you'll get Opus while the team defaults to Sonnet. The deny rule applies to everyone (including you) since deny rules cascade.

📋 Key Takeaways (Review These Tomorrow)

Next up: Memory & CLAUDE.md System → — Learn how to give Claude persistent instructions and let it build its own memory.