Back to rankings

Claude Code: Agent Tool Gets an Effort Parameter for Sub-Agents

Claude CodeView original changelog

Claude Code 2.1.292 added an effort parameter to the Agent tool, so Claude can run a sub-agent at the effort level a user asks for. Delegated work no longer has to inherit the main session's effort setting.

Key Takeaways

  • Per-sub-agent effort is now possible because the Agent tool accepts an effort parameter in Claude Code 2.1.292.
  • Users ask in plain language and Claude sets the level when it launches the sub-agent, with no new command to learn.
  • Cheap tasks such as searches can run at lower effort to save time and usage, while hard problems get more reasoning.
  • Delegated work no longer inherits the main session's effort, removing a compromise in mixed conversations.
  • The change fits the broader trend of finer control over multi-agent workflows in Claude Code.
  • The release notes do not enumerate accepted values, so behavior should match existing effort levels.

Choosing effort per sub-agent

Claude Code 2.1.292 added an effort parameter to the Agent tool. When a user asks Claude to run a sub-agent at a particular effort level, Claude can now pass that level straight to the sub-agent instead of leaving it on whatever the main session uses.

Why it matters

Sub-agents are commonly used to split a larger task into parallel or isolated pieces. Those pieces rarely need the same amount of reasoning. A quick file search or a mechanical refactor can run at a low effort level to save time and usage, while a tricky design review or a hard debugging task can be given a higher one. Until now the effort level of delegated work followed the surrounding session, so users who wanted a cheap lookup and a deep analysis in the same conversation had to compromise.

How to use it

No new command is involved. Users state the effort they want in their request, for example asking for a sub-agent to investigate a failing test at high effort, and Claude sets the parameter when it launches the sub-agent. The release notes do not list the accepted values, so users should expect the same effort levels that Claude Code already exposes elsewhere.