What problem does it solve? Writing and maintaining agent instructions, skill definitions, and slash commands often produces vague, unreliable prompts that degrade agent behavior over time. This Skill provides structured patterns, quality checklists, and an evidence-driven refinement loop to keep instruction surfaces clear and testable. ## Core Features & Use Cases - Prompt Structure Patterns: Apply reusable templates like Role-Context-Task-Format (RCTF), chain-of-thought gates, few-shot rubrics, and Azoth effect labels for write-capable instructions. - Instruction Quality Checklist: Validate prompts against clarity, reliability, boundary, and testability criteria before finalizing them. - L2 Auto-Refinement: Consume machine-readable evaluation evidence from .azoth/memory/l2-refinement-evidence.jsonl to propose A/B instruction variants for human approval. - Use Case: When evaluation data shows an agent repeatedly failing a task, use this Skill to read the L2 evidence records, hypothesize an instruction fix, and propose a refined SKILL.md variant for human review. ## Quick Start Ask the agent to review and improve a specific SKILL.md or slash command file using the prompt-engineer patterns and quality checklist.
