instruction-writer
Write and optimize instructions for Claude Code (skills, slash commands, memory files, CLAUDE.md). **ALWAYS use before creating or modifying any CLAUDE.md file, SKILL.md file, command file, or instruction documentation.** Triggers when writing skill descriptions, optimizing prompts for clarity/token efficiency, or when user mentions "write a skill", "create a command", "optimize instructions", or "improve this prompt".
When & Why to Use This Skill
This Claude skill provides expert knowledge and automated guidance for writing, optimizing, and maintaining effective instructions for Claude Code. It focuses on enhancing clarity, ensuring token efficiency, and managing configuration files like CLAUDE.md and SKILL.md to improve agent reliability and performance while reducing operational costs.
Use Cases
- Scenario 1: Developing new custom skills by generating precise SKILL.md files that follow agentic best practices and imperative formatting.
- Scenario 2: Optimizing complex system prompts to minimize token consumption and stay within the attention budget without losing instruction specificity.
- Scenario 3: Auditing and refactoring CLAUDE.md files to eliminate redundant rules and resolve conflicting instructions across multiple project files.
- Scenario 4: Designing intuitive slash commands and memory files that provide clear, actionable context for the agent to execute tasks accurately.
| name | instruction-writer |
|---|---|
| description | Knowledge for writing effective Claude Code instructions. Auto-triggers for CLAUDE.md, SKILL.md, command files. |
Instruction Writer Skill
Load this skill to enter instruction-writing mode. Provides knowledge and principles for writing effective Claude Code instructions.
When Loaded
Auto-triggers when:
- Creating or modifying CLAUDE.md, SKILL.md, or command files
- Optimizing prompts for clarity or token efficiency
- User mentions "write instructions", "create a skill/command", "optimize this prompt"
Core Principles
Apply all principles from @best-practices.md. Key focus areas:
- Clarity & specificity over vagueness
- Token efficiency - every token depletes attention budget
- Canonical examples over exhaustive edge cases
- Actionable steps in imperative mood
What to Check
When reviewing instructions, look for:
- Ambiguity - what could a fresh agent misinterpret?
- Cross-file conflicts - do related files have contradicting rules?
- Redundancy - is this duplicated elsewhere?
- Missing context - does this assume knowledge not provided?
Supporting Files
- @best-practices.md: Comprehensive prompt engineering guidelines for Claude 4.x