Agent session

An agent session is one continuous run of a coding agent, from the moment you start it to the moment you end it. Within a session the agent remembers what it has seen and done.

Between sessions it does not. Each Claude Code session begins with a fresh context window, so the next run reads your code again but not which task was next or what it already tried. That reset is the root of most agent friction, and the fix is to keep task state outside the session, in a file or on a board the agent reads first thing. For the full approach, read keeping Claude Code on track across sessions.