Agents fill the gaps you didn't

Hand an AI agent a feature to build and it will do the parts you specified — and confidently invent the parts you didn't. It has limited context and no way to ask, so every gap in your plan becomes a guess. Usually a plausible one. Often the wrong one.

You can't close that gap with a better task ticket. Two reasons. First, the plan lives in Linear or Asana — somewhere apart from the code — so it can't actually constrain what gets built; it sits beside the work, not inside it. Second, every plan is shaped a little differently, so the agent can't rely on finding what it needs in a predictable place. A freeform plan and a stochastic builder is a bad combination: the agent fills the unspecified with its own assumptions, and you find out at review.

So the question Specline answers is narrow and concrete: how do you give an agent a feature to build such that it knows exactly what's in scope, what's out, when it's done, and who decides — with nothing left to assume?

The answer has two moving parts, and the next two chapters are each of them: put the spec where the code is, and give every spec the same shape. Everything else in this handbook follows from those two.