teaql-agent-kit

Why we included this project

Most coding agents run in a prompt-to-code loop: one pass from a requirement to code, tests, and repairs, with the domain contract ending up whatever the model happened to invent along the way. TeaQL Agent Kit breaks that loop by placing an executable domain model between intent and implementation. The agent models the domain first, a deterministic evaluator checks it and returns concrete repair guidance, and only then does the kit generate a typed API contract for the agent to implement against. So the agent works inside a generated, constrained surface instead of inventing the shape of the system as it goes. The same discipline carries into runtime: operations carry identity and intent, writes declare an audit reason, and external capabilities have to be explicitly granted, which keeps application behavior observable once it ships. The kit ships as an installable agent skill with reference toolchains and a live walkthrough, so you can try the pattern in your own environment. It is less a turnkey assistant than a template for putting deterministic structure around non-deterministic AI output, which makes it most valuable to teams building or customizing agentic development pipelines.