AXIOMARIUM
My open lab for building, testing and debugging an AI coding environment like software: the agents, skills, hooks, policies and evals I use, plus axm, a local CLI that inspects, validates and tests them.
AI coding configuration drifts. Instructions contradict each other, skills overlap, old rules stay forever, and broken behaviour gets fixed by adding more prose. Axiomarium treats that as an engineering problem: failures should become tests, not paragraphs. Its vault holds agents, skills, hooks, policies, workflows, schemas and evals as plain Markdown, YAML and JSON Schema, and every asset carries a manifest saying what it is, which harnesses it supports, what it may touch, how it’s tested and how far it can be trusted.
axm is the engineering around the vault. axm explain shows what Claude Code and Codex actually load for a file, what they silently drop and which skills and hooks reach the agent; axm doctor checks the instruction files, skills and hooks of any repo; axm triggers finds skills whose descriptions overlap, and axm triggers test runs a skill’s trigger prompts on both harnesses to see whether the agent picks it. Since v0.5, axm eval run runs an asset’s eval cases in a sealed home, and axm conflicts --judge has a model find instructions that contradict each other.
It’s in early development and built for my own setup first. In a terminal the output is in colour, with a kaomoji for the outcome; piped, in CI or read by an agent, it’s plain text.