Improve an existing agent based on recent performance:
- Analyze recent uses of: $ARGUMENTS
- Identify patterns in:
- Failed tasks
- User corrections
- Suboptimal outputs
- Update the agent's prompt with:
- New examples
- Clarified instructions
- Additional constraints
- Test on recent scenarios
- Save improved version