Allow configuring model for built-in agents without replacing entire agent definition

Problem

Built-in agents like Explore have a hardcoded model (e.g., haiku). If a user doesn't have access to that specific model, the agent fails with:

There's an issue with the selected model (claude-haiku-4-5). It may not exist or you may not have access to it.

The environment variable CLAUDE_CODE_SUBAGENT_MODEL only affects agents without an explicit model defined - it doesn't override the built-in agents' model settings.

Current Workaround

Users must create their own agent file (e.g., ~/.claude/agents/Explore.md) that completely replaces the built-in agent. This means:

• Losing the carefully crafted prompts from Anthropic
• Having to guess at the original agent's behavior
• Missing out on future improvements to the built-in agents

Proposed Solution

Add a way to override only the model for built-in agents while keeping their original prompts and configuration. Options:

1. Model-specific env vars: CLAUDE_CODE_EXPLORE_MODEL=sonnet
2. Override priority: Make CLAUDE_CODE_SUBAGENT_MODEL override even explicit model settings
3. Config file option: Allow settings.json to specify model overrides per agent name

Use Case

• Users on API plans without access to certain models (e.g., no Haiku access)
• Teams wanting to use higher-quality models for exploration tasks
• Cost optimization by using cheaper models for simple agents

Environment

• Claude Code version: 2.1.41
• OS: Windows 11