Skip to content
Get packs

Skill Optimizer

Tunes an existing agent skill's SKILL.md against a benchmark, keeping only validated improvements.

Source Download skill
Repository
garrytan/gbrain
Stars
31,447
Licence
MIT
Files
1 file

What it does

Once installed, the agent can take a skill you already have and try to improve its instructions automatically. It generates a starter test set (benchmark) from the skill, lets you strengthen the tests, then runs repeated rounds of rewriting the skill's body text and keeps a change only if it scores better on a held-out validation set. Frontmatter routing settings such as triggers are not changed, and bundled skills are protected from in-place rewrites unless you pass explicit flags and a separate held-out set.

Great for

  • Developers maintaining custom agent skills in the gbrain ecosystem who want measurable quality gains
  • Teams re-tuning a skill after switching the target model
  • Users who already have a benchmark file and want to iterate on a skill's wording

Skip it if

  • Anyone without the gbrain CLI installed, since the skill calls gbrain skillopt commands that the SKILL.md alone cannot run; it is not useful without that tooling
  • Skills you do not own or want to rewrite automatically: bundled skills need --allow-mutate-bundled and a held-out set, and auto-rewrites can drift from intent if judges are weak
  • Quick one-off edits to a skill, because the pipeline makes many LLM calls and can be costly (it has a cost preflight and --max-cost-usd cap, but still spends money)

Try asking

  • “Optimize this skill: skills/meeting-notes”
  • “Make the skill better, it keeps missing the action items section”
  • “Bootstrap a benchmark for my new invoice-summary skill, then run skillopt with the 1:1:1 split”
  • “Run skillopt for the research-brief skill in dry-run mode to preview cost first”
Step-by-step guide: how to use Skill OptimizerInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 1 original files, the MIT licence and an attribution note.

    skill-optimizer.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip skill-optimizer.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip skill-optimizer.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Always review and strengthen the auto-generated judges before deleting the sentinel line; the generated checks are described as weak drafts.
  • Run with --dry-run first to preview cost, and set --max-cost-usd to cap spending.
  • Use --split 1:1:1 for a small starter benchmark of about 15 tasks, since the default split needs around 50 tasks.

What's inside

  • SKILL.md

Redistributed unmodified from garrytan/gbrain under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.