Loop Engineering Skill

Standalone skill for designing, auditing, and critiquing closed-loop LLM agent systems: propose -> check -> measure -> feedback -> refine.

The entry point is SKILL.md. It routes into the knowledge base:

• knowledge_base/04_design_guide.md for composing a new loop
• knowledge_base/05_audit_checklist.md for debugging or optimizing a loop
• knowledge_base/06_antipatterns.md for known failure modes

Core principle: a loop improves only when the checker has an edge the generator lacked. Start by naming that verifier edge.

Contents

• SKILL.md - skill frontmatter and routing entry point
• knowledge_base/01_inventory.md - source inventory
• knowledge_base/02_patterns.md - design axes and evidence map
• knowledge_base/03_critique.md - rules and evidence-weighted verdicts
• knowledge_base/04_design_guide.md - compose procedure
• knowledge_base/05_audit_checklist.md - audit and optimization checklist
• knowledge_base/06_antipatterns.md - named failure modes and fixes