Claude Code: effort Frontmatter for Skills and Slash Commands

Claude Code

Claude Code v2.1.80 extends the effort level system to skills and slash commands via a new effort frontmatter field in SKILL.md. Skill authors can now pin a specific effort level β€” low, medium, high, or max β€” directly in a skill's YAML frontmatter, overriding the global session setting for the duration of that skill's execution. This enables cost-sensitive workflows such as quick linting or commit message generation to run at lower effort levels without requiring manual /effort adjustments between invocations, and completes the effort override surface across all three skill types in Claude Code.


effort Frontmatter for Skills and Slash Commands

Claude Code v2.1.80 introduces the effort frontmatter key for skills and slash commands, giving skill authors direct control over the model's thinking budget for each workflow they ship.

Prior to this release, the effort level was a global session setting. Developers could adjust it with /effort or by editing settings.json, but individual skills had no way to declare their own preference. A resource-intensive architectural review skill and a quick single-file formatter would both run at the same global effort level unless the user manually switched between invocations.

How It Works

Adding effort: low (or medium, high, or max) to a SKILL.md YAML frontmatter block causes Claude Code to use that effort level for the duration of the skill's execution, then restore the previous session-level setting when the skill completes. The global setting is never permanently changed.

Example SKILL.md frontmatter:

---
name: quick-lint
description: Run a fast lint pass on the current file
effort: low
---

This pairs naturally with the existing model: frontmatter field. A skill can now fully specify its execution profile β€” both the model variant and the effort level β€” without requiring any manual setup from the user at invocation time.

Practical Use Cases

  • Low effort: Commit message generation, single-file formatting, changelog summarization, and other tasks where extended thinking adds token cost without meaningfully improving output quality.
  • High or max effort: Architectural review workflows, security audit agents, and complex multi-file refactoring tasks where deeper reasoning produces substantially better results.
  • Medium effort: General-purpose development tasks that benefit from some deliberation but do not require maximum computation.

Completing the Effort Control Surface

The effort frontmatter key for skills and slash commands is the last piece of a three-part effort override system built across several recent releases. The /effort slash command (introduced in v2.1.76) gave users a session-level control. The effort frontmatter for plugin-shipped agents (introduced in v2.1.78) gave agent definitions a per-invocation override. With v2.1.80, the same override is now available in skills and slash commands β€” completing consistent effort control across every skill surface in Claude Code.

Skill library authors can now ship cost-calibrated workflows out of the box, removing the need for end users to understand or manage effort levels themselves.