claude-prompt-engineering-guide

This guide provides official Anthropic best practices for writing professional prompts optimized for Claude Opus 4.6, Sonnet 4.6, and Haiku 4.5 models. It covers Skills, MCP integration, Hooks, Agent Teams, and Subagents with 220+ verified sources and real-world examples. The guide is automatically updated weekly via AI research to reflect the latest Claude capabilities, including Claude Code v2.1.83 features like voice mode, 1M context windows, and remote execution. Operations teams use this resource to automate workflows, build multi-agent systems, and optimize prompt engineering across production environments. It includes 100+ curated skills, adaptive thinking strategies, and configurations for extended reasoning with ultrathink mode.

[{"step":"Identify the automation workflow and data sources","action":"Determine which MCP features and data sources (e.g., Salesforce, Zendesk) your prompt will integrate with. Ensure the MCP endpoints are properly configured in your Claude environment.","tip":"Use the `claude mcp list` command to verify available MCP servers before writing your prompt."},{"step":"Define the task and output structure","action":"Clearly outline the task your prompt will automate (e.g., customer onboarding, report generation) and the expected output format (e.g., JSON, table). Include placeholders for dynamic data like customer IDs or dates.","tip":"For complex workflows, break the task into smaller sub-tasks and assign each to a specific section in your prompt (e.g., Context, Instructions)."},{"step":"Integrate MCP and Skills","action":"Reference the MCP features and Skills (e.g., claude-prompt-engineering-guide) directly in your prompt. Use the [SKILL_NAME] and [MCP_FEATURES] placeholders to ensure the AI understands the required integrations.","tip":"Consult the official Anthropic documentation for MCP and Skills to ensure you're using the correct syntax for integration."},{"step":"Test and refine the prompt","action":"Run the prompt in a controlled environment (e.g., Claude's playground) and validate the output against your expectations. Adjust the prompt structure or constraints based on the results.","tip":"Use the `claude analyze` command to evaluate the prompt's efficiency and identify areas for improvement."},{"step":"Implement in your automation workflow","action":"Integrate the finalized prompt into your automation tool (e.g., Zapier, custom script) and set up triggers (e.g., new customer sign-up, support ticket creation). Monitor the output for accuracy and consistency.","tip":"Log the AI's responses to track performance and identify edge cases that may require prompt adjustments."}]