Why a Subagent's Investigation Doesn't Bloat Your Session
You ask Claude Code to "use a subagent to investigate how our authentication system handles token refresh" instead of investigating directly yourself.
Why does this keep your main session's context small, even if the investigation reads dozens of files?
The correct answer is "It runs in its own context window and returns a summary."
A subagent (the Agent tool, formerly Task) is a separate
execution with its own context window; all those file reads happen
there, and only a condensed summary comes back into your main
conversation. Files aren't deleted from disk as some side effect, tool
calls inside a subagent are billed the same as anywhere else, and
nothing limits the main session to seeing only "the first three" tool
calls — the main session simply never sees the subagent's tool calls
at all, only its final report.
Share this question