https://github.blog/ai-and-ml/github-copilot/how-to-build-reliable-ai-workflows-with-agentic-primitives-and-context-engineering/

Type: url

Original: https://github.blog/ai-and-ml/github-copilot/how-to-build-reliable-ai-workflows-with-agentic-primitives-and-context-engineering/

Status: generated

Summary

A guide on GitHub’s blog proposing a three-layer framework for turning ad-hoc AI prompting into a repeatable engineering practice: Markdown as a vehicle for prompt engineering, agent primitives as reusable Markdown files that each carry one capability or rule (.instructions.md, .chatmode.md, .prompt.md, .spec.md, .memory.md, .context.md), and context engineering to keep an agent’s limited attention on what matters. Its second half argues that these files are genuine software written in natural language and therefore need the infrastructure any software ecosystem develops, proposing APM — the author’s own Agent Package Manager — as the npm-equivalent layer for distributing them and running them in CI/CD. It reports no evaluation of the framework.

Generated pages

• How to build reliable AI workflows with agentic primitives and context engineering
• Agent Primitives
• Context engineering
• Outer Loop
• APM (Agent Package Manager)