In September a dev deploy of one of my side projects took 29 minutes, and a deploy of a commit that changed nothing updated all 11 of its CloudFormation stacks and rebuilt both of its Docker images. The pipeline had been built the way I build most things now, one feature at a time with Claude writing most of the code, and each piece of it had been correct when it was added.
I started muda to find out where those minutes went. It began as an audit script and a written standard inside that project, and it is now a standalone tool at muda.tools: a Go binary that reads a repository's GitHub Actions history, plus a set of agent skills that walk a coding agent through measuring, diagnosing, fixing and verifying. This post covers the project the standard came from: what the measurements showed, what changed, and why a pipeline built this way ends up like this. A second post covers the first pipeline muda ran on that it had never seen.