Why use XML tags instead of standard Markdown?

While Markdown is ideal for human readability, XML provides rigid, machine-scannable anchors that help the LLM identify where one instruction ends and another begins, which is critical for complex context.

Does this skill help reduce token usage?

Yes, it applies a 'minimum viable structure' philosophy, ensuring that tags are used as functional anchors rather than decoration, which optimizes the model's focus and token efficiency.

When should I activate the 'think harder' or 'ultrathink' modes?

Use these modes when architecting system prompts longer than 2,000 tokens or when designing coordination logic for multi-agent systems where attention economics are vital.

What is XML context engineering?

It is a technique for structuring LLM prompts using XML tags to create clear semantic boundaries, making it easier for models to process, isolate, and retrieve specific segments of information.

Can I use this for everyday coding tasks?

This skill is best reserved for prompt engineering and system instructions. For regular code and simple messaging, standard Markdown remains the more efficient and readable choice.

XML Context Architect

Name: XML Context Architect
Author: mahidalhan

bymahidalhan

•

生産性とワークフロー

Transforms unstructured prompt content into optimized XML-tagged structures to maximize LLM comprehension and attention efficiency.

XML Context Architect is a specialized tool designed to engineer high-performance system prompts and complex instructions for Claude and other LLMs. By applying rigorous principles of attention economics, it helps developers structure raw data into semantic units that create clear cognitive boundaries for the model. This skill eliminates 'tag soup' and prevents context rot by focusing on retrieval-centric naming, flat hierarchies, and progressive disclosure, ensuring that agents can accurately isolate and act upon critical information within even the most complex prompt environments.

主な機能

01Retrieval-optimized tag naming for improved model referencing

02Extended reasoning modes (think harder/ultrathink) for complex context design

03Semantic XML tagging to establish clear cognitive boundaries

04Flat-first structural patterns to maintain high signal-to-noise ratios

05Attention economics analysis to minimize token waste and context rot

0620 GitHub stars

ユースケース

01Architecting high-stakes system prompts for production-grade LLM agents

02Optimizing long-context prompts to ensure reliable information retrieval

03Structuring complex multi-part instructions for multi-agent coordination

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add mahidalhan/claude-hacks xml-context-engineering

For use in Claude.ai and ChatGPT

Download Skill