Skip to content

Context & Baselines

This section contains the behavioral baselines, instruction templates, and high-fidelity context documents that govern how AI assistants operate in development environments.

What's Here

  • LLM Baseline Behaviors (LLM-BaselineBehaviors.md) — The foundational behavioral model for AI assistants. Covers communication style, action-oriented behavior, tool usage, code quality, and problem-solving.
  • Baseline Behaviors v2.0 (BaselineBehaviors-v2.0.md) — The XML-structured evolution of the baseline. Optimized for machine readability, KV-caching, and attention anchoring.
  • Copilot Instructions (copilot-instructions.md) — Repository-specific directives for GitHub Copilot, defining instruction priority order.
  • High-Fidelity Context (high-fidelity-context.md) — Explains the strategic rationale for migrating AI directives from Markdown to structured XML modules.
  • High-Fidelity Migration Guide (high-fidelity-context-migration.md) — Step-by-step guide for converting existing AGENTS.md files to the XML system.

Instruction Priority Hierarchy

Follow this canonical hierarchy when multiple instruction sources exist:

  1. Explicit user directives — Direct commands in the current conversation (highest priority)
  2. Repository-specific directives — Guidelines in AGENTS.md or AGENTS.xml files
  3. Tool-specific guidelines — Instructions from external tools or systems
  4. Baseline behaviors — Foundational model (lowest priority)

This hierarchy ensures clear decision-making when instructions conflict. For detailed configuration layer documentation, see Agent Configuration.

Evolution Path

The baseline originally shipped as Markdown (LLM-BaselineBehaviors.md). Version 2.0 introduces an XML-structured format for better machine parsing. Both are maintained; new projects should prefer the XML version.