teaql/teaql-agent-kit

What it solves

TeaQL Agent Kit addresses the unreliability of coding agents that move directly from requirements to implementation. It prevents agents from inventing domain contracts and business logic simultaneously by introducing a deterministic, model-mediated layer between intent and code.

How it works

Instead of a prompt-to-code loop, TeaQL implements a "harness pattern" that forces a specific execution order:

1. Domain Modeling: The agent creates an inspectable intermediate representation (KSML model) of the business intent.
2. Deterministic Evaluation: A feedback oracle checks the model for errors and provides specific repair guidance.
3. Contract Generation: Once validated, the model is turned into a typed API boundary (domain libraries) for multiple languages.
4. Constrained Implementation: The agent implements the application logic against this fixed contract, aided by model-aware assist tools that teach the agent the available APIs.
5. Verification: The result is verified through compilation, tests, and runtime policy checks to create an evidence-backed result.

Who it’s for

Software developers and AI engineers building coding agents or agentic workflows who require high reliability, auditability, and deterministic structure around non-deterministic AI generation.

Highlights

• Multi-Language Support: Generates typed domain libraries for Java, Rust, Go, Swift, Python, C#/.NET, and TypeScript.
• Runtime Governance: Enforces mandatory identity, intent auditing (purpose/comments for reads, audit reasons for writes), and capability sandboxing.
• Evidence-Based Completion: Creates a traceable chain of evidence from the initial requirement through evaluation and testing to the final application.
• Parallel Human Review: Allows human reviewers to inspect the KSML model and provide asynchronous feedback without blocking the agent's progress.