Skip to content

README: LLM Baseline Behavioral Model (XML Version)

This repository contains the authoritative behavioral baseline for AI development assistants. Transitioning from the legacy Markdown format to this XML-structured model improves machine readability, context efficiency, and consistency across all projects.

Why XML?

While Markdown is excellent for human readability, XML provides a superior "grammar" for Large Language Models (LLMs) when used in system prompts:

  • Attention Anchoring: Explicit tags like <action_logic> act as semantic anchors, helping the model maintain focus on specific rules during long sessions.
  • Context Efficiency: The structure reduces token overhead by removing repetitive formatting while maintaining full fidelity of the original directives.
  • KV-Caching Optimization: This file is designed as a static prefix. By keeping it identical across all projects, you maximize "Key-Value Cache" hits, resulting in faster response times and lower latency.

Core Sections

  • <core_identity>: Defines the persona, communication tone, and expansion triggers for technical explanations.
  • <action_logic>: Governs how the agent decides between acting immediately and asking for clarification.
  • <background_protocols>: Specialized instructions for autonomous or non-interactive CLI operations, including "Assume & Document" logic.
  • <workspace_protocols>: Directives for repository research, tool efficiency, and Git safety.
  • <management_guardrails>: Safety limits, including time-boxing, loop detection, and stateful recovery via .state.json.

Usage Guidelines

  1. Keep it Static: Do not modify the XML structure or add project-specific data to this file. All project-specific rules should go into a local AGENTS.md file.
  2. The Cache Breakpoint: Ensure your loader injects this file at the very beginning of the system prompt. The `` tag marks where variable session data should begin.
  3. Background Mode: When running agents via CLI with background flags, the agent will automatically pivot to the protocols defined in <background_protocols>.

How to Contribute

If you need to propose a change to the baseline behaviors:

  1. Ensure the change is applicable to all projects.
  2. Maintain the XML tag hierarchy to preserve structural clarity.
  3. Test the new directive to ensure it doesn't cause "instruction drift" or conflict with existing <action_logic>.

This model is designed to ensure consistent, high-quality AI assistance across all tools and platforms.