AgentRC

Context engineering for AI coding agents.

Warning

Experimental — Under active development. Expect breaking changes. Open an issue with feedback.

---

AI coding agents work best when they know how to build, test, and lint your code — plus your architecture, conventions, and the external services your team relies on. Most repos ship none of that.

AgentRC reads your codebase and generates the files that close that gap — then evaluates whether they actually help, so the context doesn't go stale as your code evolves.

npx github:microsoft/agentrc

Works as a CLI, as a VS Code extension, and in your CI/CD pipeline to monitor drift. No config needed — runs on any repo with Node.js 22+.

What it does

Measure

Score your repo’s AI-readiness across 9 pillars and a 5-level maturity model. Find out what context is missing — from basic linting to MCP server configs.

npx github:microsoft/agentrc readiness

Generate

Produce tailored instruction files, evals, and dev configs via the Copilot SDK. No templates — AgentRC reads your actual code and generates context specific to your stack.

npx github:microsoft/agentrc instructions

Maintain

Context goes stale as your codebase evolves. Evaluate whether your instructions still improve agent responses, and run the check in CI so drift doesn't slip through.

npx github:microsoft/agentrc eval

AgentRC maintains itself with GitHub Agentic Workflows

This repository dogfoods GitHub Agentic Workflows for recurring maintenance and demos:

Workflow	Purpose	Output
AgentRC Readiness Report	Runs AgentRC against this repository and explains the highest-impact gaps	One short-lived issue
AgentRC Readiness Upgrade	Improves one evidence-backed instruction and verifies the result	One draft PR
Issue Triage	Classifies new issues and asks for missing information	Bounded labels and one comment
CI Failure Doctor	Diagnoses failed CI from logs and repository changes	One issue or one comment

The source workflows are Markdown. gh aw compile generates the reviewable .lock.yml workflows that GitHub Actions executes. CI fails when those generated files drift.

See Agentic Workflows for the architecture and maintenance commands. Use the golden demo runbook for the lone-maintainer narrative, exact stage flow, and fallback plan.

Works at every scale

Workflow	Command
Interactive hub	agentrc
One-time setup	agentrc init
CI quality gate	agentrc readiness --fail-level 3 --json
Batch across an org	agentrc batch
Automated PR for any repo	agentrc pr owner/repo

Works with GitHub and Azure DevOps. Supports monorepos, multi-root VS Code workspaces, and custom policies.

What gets generated

File	Purpose
.github/copilot-instructions.md	Teaches AI agents your repo's conventions
.vscode/mcp.json	Connects AI to your stack's tools and data
.vscode/settings.json	Tunes VS Code for AI-assisted development
agentrc.eval.json	Test cases to measure instruction quality

For multi-agent support (Copilot + Claude + others), generate AGENTS.md with --output AGENTS.md. See Custom instructions in VS Code.

Works with APM

APM (Agent Package Manager) distributes agent instructions, skills, and prompts across repos — like npm for AI agent configs.

AgentRC generates the content. APM distributes it:

• In your project — run agentrc init to generate instructions, then apm install org/standards to pull in shared agent packages from your team
• For your team — create a dedicated APM package with your best instructions and skills, then teammates install it with apm install
• At scale — apm audit scans for security issues; apm-policy.yml enforces org standards across all repos

The .instructions.md format is shared by both tools — no conversion needed when moving instructions into APM packages.

Documentation

Getting Started	Prerequisites and first run
Concepts	Maturity model, readiness pillars, how generation works
Commands	Full CLI reference
Configuration	Areas, workspaces, monorepos
Policies	Custom readiness scoring
At Scale	Batch processing across orgs
CI Integration	GitHub Actions & Azure Pipelines
VS Code Extension	Sidebar views, commands, settings
Agent Plugin	Install as a Copilot agent plugin with built-in skills
Agentic Workflows	Dogfooding, maintenance, and live demo guide
Golden Demo	Lone-maintainer story and stage-ready runbook
Examples	Configs, evals, and policies

Customize AI in VS Code · Custom instructions · CONTRIBUTING.md

Troubleshooting

"Copilot CLI not found" — Install the GitHub Copilot Chat extension in VS Code. The CLI is bundled with it.

"Copilot CLI not logged in" — Run copilot in your terminal, then /login.

"GitHub auth required" — brew install gh && gh auth login, or set GITHUB_TOKEN (or GH_TOKEN).

"Azure DevOps auth required" — Set AZURE_DEVOPS_PAT or AZDO_PAT.

License

MIT

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.