BREAKING
Anthropic Publishes Claude Code 'Loops' Guide
What Is a Loop?
Four Loop Types by Use Case
Turn-based
Manual
●
User prompt starts it
●
Short one-off tasks
Goal-based
/goal
●
Verifiable end criterion
●
Evaluator confirms done
Time-based
/loop /schedule
●
Runs at set intervals
●
External monitoring
Proactive
Event
●
Runs without a human
●
Bug triage, updates
Practices for Code Quality
1
Clean codebase
↓
2
Self-verify SKILL.md
↓
3
Second-agent review
↓
4
/code-review skill
Managing Token Usage
1
Cheap models for small tasks
↓
2
Set stop criteria
↓
3
Run small pilots
↓
4
/usage breakdown
Focus Shifts to Designing Loops
AI NEWS BLITZ
Anthropic's Claude Code team just released a design guide for agent loops.