Master the terms behind delegating isolated tasks to Claude Code subagents.
0 / 5 completed
1 / 5
At standup, a dev wants a task delegated to an isolated Claude Code agent so it doesn't pollute the main conversation's context. Which feature fits?
A subagent runs in its own isolated context window, performs a delegated task, and reports a summary back to the main conversation. This keeps the primary session's context clean of intermediate exploration. It is distinct from skills, which inject instructions into the current context instead.
2 / 5
During a design review, the team wants a subagent scoped to only read files and never write them. Which configuration achieves this?
Subagents can be configured with a restricted tool set, such as read-only file access, so their capability matches the intended scope of the delegated task. This limits the blast radius if the subagent misbehaves. Scoping tools per subagent is a standard safety practice.
3 / 5
In a code review, a dev notices a subagent returned only a concise summary rather than its full exploration transcript. Why?
Subagents intentionally return a condensed result to the parent conversation instead of their full internal transcript, preserving the parent's context budget. The detailed work happens in the subagent's isolated context and is discarded once summarized. This tradeoff is central to why subagents help manage long-running tasks.
4 / 5
An incident report shows a subagent was launched for a task better suited to a quick, in-context tool call. What was the likely inefficiency?
Spinning up a subagent adds isolation overhead that is wasted on trivial tasks that could be handled with a direct tool call in the current context. Reserving subagents for substantial, self-contained work avoids this inefficiency. Matching the mechanism to the task size is a key design consideration.
5 / 5
During a PR review, a teammate wants multiple subagents working on independent parts of a large refactor in parallel. What does this pattern enable?
Launching several subagents in parallel lets independent parts of a large task (like separate modules of a refactor) progress simultaneously, each in its own isolated context. This can meaningfully cut wall-clock time for decomposable work. Coordinating and merging their results remains the orchestrating agent's responsibility.