Context limit reached in review agents

[kieranklaassen]

seeing this a lot now with opus 4.6 high effort

[joshRpowell]

I just started experimenting with this in my CLAUDE.md
Will report back.

## Model Usage Policy

**Default Model:** Always use Sonnet for all tasks unless explicitly told otherwise or granted permission to escalate.

**When to ask about Opus:**
Before switching to or recommending Opus, ALWAYS ask the user if the task warrants it. Consider asking when:
- Complex multi-step architectural decisions are needed
- Deep reasoning about system design or tradeoffs
- Debugging particularly gnarly, non-obvious issues
- Tasks involving multiple files with complex interdependencies

**For routine tasks, stick with Sonnet:**
- Feature implementation
- Refactoring
- Writing tests
- Documentation
- Bug fixes with clear causes
- Code reviews

When in doubt, explicitly ask: "This task might benefit from Opus for [specific reason]. Would you like me to switch to Opus, or should I proceed with Sonnet?"

[jimcru21]

Got this issue of context limit too on deep-plan