Zero to Agent in 30 Minutes: Build a Supply Chain for Agent Context with Maxim Salnikov

We still haven’t solved the problem of keeping track of everything we’re feeding our AI agents. Developers now install agent skills, instructions, and other customizations from public repositories by the dozens, and those files end up scattered across user profiles, application folders, and codebases with no record of where they came from or whether they’ve changed since they were first installed.

In this episode of Zero to Agent in 30 Minutes, Microsoft senior solution engineer Maxim Salnikov walked through the Agent Package Manager (APM), a terminal-driven open source product from Microsoft that treats agent context the way modern software already treats its dependencies: versioning it, pinning it, and checking it before it ships. It’s a technical session, but rather than building an AI agent, you’ll discover how to manage all the customizations you’ve installed for your agents and make them portable, secure, and governed by policies you or your company define.

How to build a supply chain for agent context

Maxim demoed the process of setting up and using APM step-by-step. Here’s how it works.

Keep every customization in your codebase. Agent skills, custom instructions, and hooks tend to spread across a developer’s machine, a harness’s install folder, and enterprise-level defaults. Maxim recommends storing them in the codebase instead so that anyone on a new machine or using a new harness ends up with the same setup as everyone else. Yes, your codebase will be bigger, but the trade-off for consistency is worth it. But be sure that skills and other customizations are secure before adding them to your codebase. Consider building your very own enterprise repository of the customizations you’ve identified as safe to use. (See step 5.) Install APM and initialize a project. APM works like npm for AI agent customizations. Running apm init sets up an apm.yaml file targeting one or more harnesses, such as GitHub Copilot, Claude Code, or Cursor, and apm install pulls a skill from a repository into the right location for each one. When you install a skill, APM also creates a log file documenting the entire resolution history. Pin to a version or a commit hash. Once a skill is installed, pin it the way you’d pin a library dependency. Doing so protects against a scenario where a legitimate skill later gets compromised and a bad actor slips malicious content into a new version under the same name. APM also scans files for hidden Unicode characters that may contain malicious instructions. Reproduce the exact setup with a lock file. APM logs every file and hash it installs. Running apm install --frozen rebuilds that exact environment from the log instead of reresolving everything from apm.yaml, so a teammate’s machine ends up with precisely the same setup as yours. Source customizations from a vetted registry, and set a policy for it. Public repositories work for experimentation, but Maxim suggests enterprises build a private, gated registry of reviewed skills. APM policy files enforce which sources are approved, so a customization from an unvetted location gets flagged. Audit for drift, then enforce it in CI. The apm audit command checks installed customizations against policy and catches unauthorized sources or content that has changed since installation. Run that same audit as a gate in a CI/CD pipeline to protect the entire organization against skill drift and bad actors.

The software supply chain already has decades of tooling behind it. That discipline hasn’t caught up with agentic AI, but APM is attempting to close that gap. Explore the project GitHub repo and get started.

Coming up next

On September 9, Menyala’s Sajal Sharma joins Zero to Agent in 30 Minutes to build a shared knowledge base that acts as a common brain across agents. He’ll show how a single repository of research, daily logs, and notes can give Claude Code, Codex, OpenClaw, and Hermes access to the same accumulated information instead of starting from zero with every new session.

Follow along with Zero to Agent in 30 Minutes on Radar, or watch the latest episode on YouTube, Spotify, Apple, or wherever you get your podcasts. If you’re an O’Reilly member, you can watch live. Save your seat.