microsoft/agentrc

What it solves

AI coding agents often lack the specific context—such as build processes, testing frameworks, and architectural conventions—needed to be truly effective. AgentRC solves this by automatically generating and maintaining the necessary context files (like instructions and tool configurations) so that AI agents can better understand and operate within a specific codebase.

How it works

AgentRC analyzes a codebase to identify missing context and then uses the Copilot SDK to generate tailored instruction files and configurations. It operates through a three-step cycle:

1. Measure: It scores the repository's "AI-readiness" across nine pillars and a five-level maturity model to identify gaps.
2. Generate: It creates specific files such as .github/copilot-instructions.md for conventions, .vscode/mcp.json for tool connections, and .vscode/settings.json for editor settings.
3. Maintain: It evaluates whether the generated instructions actually improve agent responses and can be integrated into CI/CD pipelines to detect when context becomes stale (drift).

Who it’s for

Developers and engineering teams who use AI coding agents (like GitHub Copilot) and want to optimize their agents' performance by providing high-quality, codebase-specific context without manual writing.

Highlights

• Automated Context Generation: Reads actual code to produce instructions without relying on templates.
• AI-Readiness Scoring: Provides a maturity model to quantify how well a repo is prepared for AI agents.
• AI-Agent Integration: Works as a CLI, VS Code extension, and CI/CD quality gate.
• APM Integration: Works with the Agent Package Manager (APM) to distribute shared agent instructions and skills across an organization.
• Broad Compatibility: Supports GitHub, Azure DevOps, monorepos, and multi-root VS Code workspaces.