---
title: Agent Playground: Portable Agent Environments
date: 2026-02-17
tags: ai, agents, devtools
url: https://kirkdesigns.co.uk/writing/agent-playground
---

# Agent Playground: Portable Agent Environments

## The Context

Agent workflows are messy. Copilot, Claude Code and Gemini each have their own way of storing skills, rules, MCP configs, commands, subagents and hooks. Pick one and you're locked in. Support several and you're maintaining the same config three times.

Agent Playground is our open source reference template for portable agent environments. [rulesync](https://github.com/dyoshikawa/rulesync) holds a single provider-agnostic source of truth, lockfiles pin remote skills to specific commit hashes, and [devenv](https://devenv.sh/) makes the toolchain identical locally, in CI and in prebuilt cloud agent images.

## Why This Matters

Pulling unpinned remote skills into an agent is a supply chain risk. We upstreamed lockfile support to rulesync so skill updates go through the same review as the rest of our code, and cloud runs abort if sources drift from what was vetted.

The bigger point: agent tooling moves too fast for proprietary lock-in. Your agent system should be bigger than your tool choice.