Context

In GTD, the tool, place or person an action needs before it can be done — @computer, @phone, @errands, @home, @agenda-with-Sam — and the list that groups next actions by it, so that in any situation the list shows only what is possible there

Contexts are the first of the four criteria for choosing what to do, because an action you cannot take where you are is noise. Allen’s lists were physical; with a laptop that is everywhere, the useful contexts today are energy and attention (deep, shallow, errands) and people (an agenda per recurring meeting) more than places.

In AI work, everything the model can see on a given turn — the system prompt, the conversation so far, the files it has read, the tool results, the instructions a harness injected — assembled into one input and charged for by the token

The word names the contents; context window names the fixed size they have to fit in. That distinction is the whole discipline: context is curated, not accumulated. What an agent gets right is mostly a function of what made it into this turn, which is why a repository that states its conventions in a file beats one that expects them to be inferred, and why a well-written handoff outperforms a longer conversation.