QRefAI
AI Coding

The Agentic OS Field Guide

Building a reusable coding-agent harness for Claude Code and GitHub Copilot (2026)

A question-driven knowledge article for engineers and platform teams in regulated, hybrid-cloud, polyrepo organizations. Read the preface first — it explains what changed in 2026 and how to get the most out of this.

  1. AI Coding

    Part 1 — Foundations

    What is an agentic harness, and why does it take months to build?

    The four-layer harness model, the seven primitives shared by Claude Code and Copilot, why context engineering became the central discipline, and how progressive disclosure makes a large harness viable.

    4 min · Updated June 2026

    Read this part
  2. AI Coding

    Part 2 — Claude Code, in depth

    How do I actually configure and control Claude Code in a real org?

    CLAUDE.md structure, subagents, skills vs instruction files, plugins and marketplaces, hooks, managed-settings.json, sandboxing, hybrid cloud deployment, and the two-pipeline audit approach.

    12 min · Updated June 2026

    Read this part
  3. AI Coding

    Part 3 — GitHub Copilot, in depth

    How does Copilot's agent system work, and how do I govern it?

    Copilot's instruction file hierarchy, custom agents, prompt files and chat modes, MCP governance, automated code review, GitHub Agentic Workflows, and Enterprise AI Controls.

    6 min · Updated June 2026

    Read this part
  4. AI Coding

    Part 4 — The hard cross-cutting problems

    We have hundreds of repos, regulated data, and three clouds — how do we not break everything?

    Distributing harness assets across a polyrepo, end-to-end governance and compliance, the real security risks and how to close them, and running all of this in a hybrid cloud.

    5 min · Updated June 2026

    Read this part
  5. AI Coding

    Part 5 — Encoding domain knowledge

    How do we encode what our org actually knows — and stop the agent from grepping everything?

    The three-layer approach to capturing organizational knowledge, the AGENTS.md cross-vendor standard, why AST/graph search replaces grep, and the eight-point 2026 consensus.

    5 min · Updated June 2026

    Read this part
  6. AI Coding

    Part 6 — The common harness: architecture and build plan

    What does the harness repo look like, and what do I build first?

    The full repo structure, the complete Claude Code-to-Copilot mapping table, the phase-by-phase asset checklist, the ten highest-leverage things to ship this quarter, and a four-phase rollout with explicit gates.

    8 min · Updated June 2026

    Read this part