// CRT MODE ACTIVATED — ↑↑↓↓←→←→BA to toggle
← Projects
[05] dev tooling

verify-kit

A project scaffold where `just verify` is the single source of truth for humans and AI agents alike.

demo / screenshot →
// stack
PythonCopierJinja2FastMCPMCPpydantic-aiLiteLLMClaude Haiku 4.5OpenTelemetryJaegerjustFastAPISQLAlchemyPostgresSchemathesisHypothesisReactDockeruv
// Problem

Autonomous coding agents are half-blind: they can’t click buttons, hear audio, or watch a UI render. Without verifiable feedback loops baked in from day one, “done” is just a claim, and a human ends up hand-checking everything the agent couldn’t.

// Approach

A Copier-based scaffold where every shipped feature is verifiable end-to-end through one command, `just verify`, whose green/red means the same thing to a human in a terminal, to CI, and to a coding agent. Pretty output, clickable IDE errors, structured JSON, and semantic exit codes all derive from the same verifier run. A Universal Foundation, plus optional add-ons (a FastAPI backend, an LLM harness) you select at scaffold time.

// Results

Consumed once via `copier copy`; you own the generated files and pull template improvements later via three-way merge. OpenSSF Scorecard and codecov wired in from the start.

// Learnings

The test I held every feature to: can a human verify it, and can an agent verify it? If neither, it isn’t done.