Stop Burning Your Limit on Thinking: Give Each Claude Code Subagent Its Own Effort Level
Effort controls how hard Claude thinks, and max is prone to overthinking small jobs. How to set effort low on a subagent in its file, how to ask for it per run, how to check it with /tasks, and the one variable that overrides both
Effort levels control how much Claude thinks on each step. The docs say lower effort is faster and cheaper for straightforward tasks, and that max "may show diminishing returns and is prone to overthinking." You are charged for all thinking tokens generated. So a subagent that only renames files or searches the repo does not need to run at max.
The setup (from the docs)
- Put the level in the subagent's file. In .claude/agents/renamer.md add effort: low to the frontmatter. Options are low, medium, high, xhigh and max, depending on the model. It overrides your session's level while that subagent runs.
- Or ask per run. Tell Claude: run the renamer subagent at low effort. Claude can pass an effort parameter for that one run (Claude Code v2.1.292 or later). It overrides the file's setting.
- Check it with /tasks. It names the model on each subagent's row and adds the effort level set for it (Claude Code v2.1.242 or later).
- Keep max for hard problems: security reviews, tricky bugs, big refactors.
A SUBAGENT FILE TO COPY (.claude/agents/renamer.md) --- name: renamer description: Renames files and symbols across the repo. Use for small mechanical edits. effort: low --- You rename files and symbols exactly as asked, update every import, and report what changed. Do not refactor anything else.
Run the renamer subagent at low effort for this job.
Keep reading
Free, for an email.
Unlocks 2 more sections, the document version, and every other guide on the site, for free. Enter your email once to keep reading.
45,000+ follow where these guides come from.No spam. Unsubscribe in one click.