Claude adds per-command effort levels for faster or deeper answers

Claude now lets developers set an effort level inside skills and slash commands, overriding the session default. It’s a small change with big workflow implications, enabling quick tasks to stay snappy while reserving deeper deliberation for high-stakes prompts.

claude cover

TL;DR

  • Per-command “effort level” in skills/slash commands: Sets how long Claude thinks before answering
  • Overrides session default: Task-by-task tuning of latency versus deliberation
  • Workflow split suggested: low effort for quick lookups/boilerplate; max effort for architectural or high-stakes decisions
  • Cost angle raised: Potential to save tokens by reserving deeper reasoning for select tasks
  • Developer questions: low/medium/high definitions, override vs stacking behavior, mapping to extended thinking token budget, dynamic adjustment via skills

Claude’s skills and slash commands just picked up a small but meaningful control surface: an “effort level” setting that can be specified per command. In a weekend update, Lydia Hallie shared that this setting controls how long the model thinks before answering and overrides the session default, making it possible to tune latency and deliberation on a task-by-task basis. The announcement was posted on X as a short feature note.

Per-command “effort” as a practical lever

The key detail is the scoping: effort can be set inside skills/slash commands, rather than forcing a single choice for an entire session. That changes how teams can structure agent workflows, especially when tasks vary widely in complexity and urgency.

Several replies framed it as a straightforward way to split work: low effort for quick lookups or boilerplate, and max effort for architectural or high-stakes decisions—without the overhead of moving work into separate sessions to “reset” the thinking mode.

Others immediately connected the feature to cost control. One developer noted it could “save a ton of tokens,” and another suggested this kind of dial is an “unlock” for balancing mundane recurring tasks against the ones that deserve deeper reasoning.

Questions developers immediately asked

The replies also surfaced the gaps that tend to matter once a toggle becomes part of daily tooling:

  • Definitions: what, concretely, separates low/medium/high, and whether “low” has a specific meaning beyond “faster.”
  • Mechanics: whether the per-command effort setting overrides or stacks with the session default (multiple people asked this explicitly).
  • Mapping: whether effort corresponds directly to an internal “extended thinking token budget,” or if it’s a separate control.
  • Automation: whether skills could dynamically adjust effort based on scope, rather than relying on manual selection.

Not all feedback was purely technical. A few replies questioned whether the growing set of knobs risks adding configuration overhead, while others joked about the pace of weekend shipping.

Docs, changelog, and RSS

One practical thread in the discussion was about discoverability: features like this often travel fastest through social posts. Hallie replied that there’s an official changelog and RSS feed in the docs, linking to: https://t.co/wg0FQ1YTem

Original source: https://x.com/lydiahallie/status/2035426943777263751

Continue the conversation on Slack

Did this article spark your interest? Join our community of experts and enthusiasts to dive deeper, ask questions, and share your ideas.

Join our community