← ClaudeAtlas

prompt-craftlisted

Write and refine production-grade prompts and tool definitions for Claude. Use when the user needs to craft a system prompt, improve an existing prompt, design tool schemas (ACI), or apply techniques like few-shot, XML structuring, role prompting, or chain-of-thought. Covers Claude 4.6 specifics. Do NOT use for skill authoring (use skill-author) or agent architecture (use agent-patterns).
VictorGjn/agent-skills · ★ 1 · AI & Automation · score 70
Install: claude install-skill VictorGjn/agent-skills
# Prompt Craft Write production-grade prompts for Claude's latest models. Based on Anthropic's official prompting best practices (April 2026, Claude 4.6). ## Golden Rule > Show your prompt to a colleague with minimal context. If they'd be confused, Claude will be too. ## Prompt Structure Template ```xml <system> [Role definition — one sentence] [Key constraints and rules] [Output format specification] </system> <context> [Background the model needs — documents, data, history] </context> <examples> <example> <input>[Sample input]</input> <output>[Desired output]</output> </example> </examples> <instructions> [The actual task — clear, sequential steps] </instructions> ``` **Document placement**: Long documents go at the TOP, queries at the BOTTOM. This improves quality by up to 30%. ## Core Techniques ### 1. Be Explicit, Not Implicit Claude 4.6 is literal. It follows precise instructions rather than inferring intent. | Instead of | Write | |---|---| | "Make it good" | "Write in active voice, under 200 words, with one concrete example per point" | | "Don't use markdown" | "Write in flowing prose paragraphs with no bullet points" | | "Be thorough" | "Cover all three scenarios: success, partial failure, and complete failure" | Say what **to do**, not what **not** to do. Provide the **why** behind instructions — Claude generalizes from motivation. ### 2. XML Tags for Structure Wrap distinct content types in descriptive tags. This eliminates amb