delivery-harness

Why we included this project

Delivery Harness is a small, readable codebase that shows what it takes to put an LLM inside a real operational workflow. It implements two synchronous delivery workflows, abnormal-order analysis and compensation suggestion, in which the model calls locally registered tools for orders, ETA, station capacity, and compensation rules. A deterministic rule engine, not the model, decides the final compensation amount. The evaluation scaffolding, human-review checkpoints, request trace IDs, and per-step timing give teams a concrete template for structuring their own agent harness. Since it is an educational MVP with in-memory state and synthetic data, treat it as a study aid and starting point rather than something to deploy directly.