Container-First Runtime Kit for Claude Code

I built a Container-First Runtime Kit for Claude Code.

The goal was simple: let Claude edit code in the real workspace, but make execution explicit, isolated, and observable.

So I designed a workflow where:

• project execution runs in containers
• normal app work goes through app
• risky or isolated project work goes through task-runner
• one-off experiments go through a scratch container
• logs, test results, and command output come back through clear observable paths

A big part of the project was separating:

• what the repo supports
• what defaults are reasonable
• what a specific request should actually activate

That led to a cleaner architecture with:

• deterministic detection
• draft planning
• resolver-owned activation
• runtime MCP control
• runtime-state-aware reruns
• post-change validation when feasible

I wrote about the full design here:

I Built a Container-First Runtime Kit for Claude Code
https://medium.com/@touqeershah32/i-built-a-container-first-runtime-kit-for-claude-code-68355c6da7ad

Would love feedback from people working on AI coding workflows, Claude tooling, Docker-based dev environments, or agent runtime design.