Intent
What is being built, written as a spec the plan traces back to, not a prompt that disappears with the session.
Concept
What is a Human-Agent-Contract?
Coding agents changed implementation speed, but chat history is not a durable engineering artifact. A Human-Agent-Contract makes the agreement between human intent and agent execution explicit: what is being built, where changes should happen, what validation proves success, what changed, and what remains unresolved.
What belongs in the contract
Five things every agent handoff should leave behind.
Day Shift stores each element as a plain repository file, so the contract is reviewable, versioned, and readable by the next human or agent.
Scope
Which files and behaviors a task is allowed to touch, declared before the agent writes code.
Validation
The executable check that proves the task is done, named up front so success is not a matter of opinion.
Evidence
What actually changed and what actually passed, recorded in an implementation summary beside the code.
Reconciliation
An explicit review checkpoint that rolls task evidence up against the milestone's acceptance criteria.
Deterministic versus model-assisted
Not every layer of the contract needs a frontier model.
Workflow structure reduces required model capability. Day Shift makes the routing explicit.
Deterministic checks
Structure, traceability, and readiness are verified by the CLI itself: no model needed and no drift possible.
Smaller models
Routine planning and summarization layers work with the models you already have access to.
Frontier models
Reserved for ambiguous product intent, architecture decisions, risky changes, and conflicting evidence.
Where Day Shift fits
Day Shift is the repository-native home for the contract.
It works around the coding assistants you already use instead of replacing them, and it is free to evaluate with no signup and no online activation.