a6-context-engineer_skill

Overview

This skill teaches context engineering and prompt engineering for AI agents and LLMs. It helps you design system prompts, craft few-shot examples, implement chain-of-thought, and manage context windows across Anthropic, OpenAI, and Google best practices. The tool includes a prompt optimizer that audits drafts and a context builder that generates optimal context windows for any task. Everything runs with Python stdlib only, designed for agent workflows and local use.

How this skill works

The prompt optimizer analyzes a draft prompt against documented best practices and returns actionable suggestions, improved prompt text, and rationale. The context builder assembles system prompt, user prompt, role assignment, and few-shot examples into a prioritized context window tailored to the task and target model. Both scripts output structured, copy-ready sections: analysis, suggestions, and a final optimized prompt or context bundle.

When to use it

• You need to improve an underperforming prompt for accuracy or consistency.
• Building system prompts for sub-agents, scheduled jobs, or multi-step agents.
• Creating few-shot examples to raise task precision and reduce hallucination.
• Designing outputs with strict schemas (JSON, XML, or tagged markdown).
• Managing long context windows and prioritizing critical information placement.

Best practices

• Be explicit and role-specific in the system prompt; assign clear responsibilities.
• Include 3–5 diverse few-shot examples that cover edge cases and normal cases.
• Place most critical information at the top and bottom of the context window.
• Use chain-of-thought only when it demonstrably improves reasoning for the task.
• Specify exact output schema and validation rules to avoid format drift.

Example use cases

• Optimize a customer-support agent prompt to reduce incorrect policy citations.
• Build a context window for an automated code reviewer with role-based constraints.
• Create few-shot examples and a system prompt for a financial analysis assistant.
• Debug and repair hallucination in a summarization pipeline by restructuring context.
• Compose agent instructions and subtask chains for a multi-step data-extraction agent.