microsoft/agentrc

What it solves

AI coding agents often lack the specific context—such as architecture, coding conventions, and tool configurations—needed to be truly effective. AgentRC solves this by automatically generating and maintaining the necessary instruction files and configurations that bridge the gap between a generic AI agent and a codebase's specific needs.

How it works

AgentRC analyzes a codebase to identify gaps in AI-readiness. It uses a maturity model and nine pillars of readiness to score the repository. It then uses the Copilot SDK to generate tailored instruction files (like .github/copilot-instructions.md), MCP server configurations, and VS Code settings based on the actual code rather than templates. Finally, it provides evaluation tools to ensure these instructions remain effective as the code evolves, allowing for integration into CI/CD pipelines to prevent context drift.

Who it’s for

Developers and engineering teams who use AI coding agents (such as GitHub Copilot) and want to improve the agent's accuracy and performance by providing it with high-quality, automated context engineering.

Highlights

• Automated Context Generation: Creates specific instructions and configs based on actual code analysis, not templates.
• Readiness Scoring: Evaluates a repository's AI-readiness across a 5-level maturity model.
• CI/CD Integration: Monitors for context drift in pipelines to ensure instructions don't go stale.
• Multi-Agent Support: Can generate AGENTS.md for use across different agents like Claude and Copilot.
• APM Integration: Works with the Agent Package Manager to distribute shared agent instructions and skills across an organization.