Deterministic technical-analysis agent for stocks: signals not advice, every number verifiable Python (no LLM math), golden-fixture tested, with a self-contained Jupyter notebook proof for every run.

Regression-testing CI gate for self-edited Claude Agent Skills (SKILL.md, MEMORY.md) -- golden-transcript record/replay, zero hosted infra.

Nothing hits the belt until it passes the factory. A zero-dependency gate that proves every agent tool runs and hasn't changed — before your agent is allowed to call it.

Schema-driven Documentum overlay generator: one validated Pydantic manifest in, a byte-stable per-environment Helm overlay tree out. CLI/REST/web-wizard parity enforced by test; built for AI agents to operate safely. All customer data fictional.