skills.rest
YoLaub
YoLaubCOMMUNITY
Explore all skills by this author →
GitHub

skill-optimizer

Optimizes existing SKILL.md files through validated micro-edits and eval-set scoring.

1|Updated Aug 6, 2026
Search Tags:#prompt optimization#skill#agent instructions#iterative refinement#micro-edits#rubric scoring#eval set
One-click install
npx skills add https://github.com/YoLaub/JohnAI --skill skill-optimizer-yolaub
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: skill-optimizer
Source: https://github.com/YoLaub/JohnAI/tree/main/.claude/skills/skill-optimizer
Command: npx skills add https://github.com/YoLaub/JohnAI --skill skill-optimizer-yolaub

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Existing AI agent skills often underperform, drift from their purpose, or degrade when rewritten wholesale. This Skill treats a SKILL.md file like source code, applying small validated edits instead of risky full rewrites, so improvements are measurable and regressions are caught. ## Core Features & Use Cases - Iterative micro-editing loop: Runs a baseline → propose 1-4 targeted edits → evaluate → commit or revert cycle, with snapshots before every change. - Eval set with rubrics: Builds 6-12 test prompts with weighted scoring criteria, then uses scripts to aggregate scores and compare iterations automatically. - Protected zones: Separates INVARIANTS (core objective, never auto-modified) from OPTIMISABLE content (hints, examples, phrasing) to prevent goal drift. - Use Case: Your agent's skill produces inconsistent outputs. This Skill diagnoses the root cause, proposes a targeted reformulation, tests it against the eval set, and commits only if the score improves. ## Quick Start Ask the assistant to optimize your existing skill file by analyzing its weaknesses and improving it through tested micro-edits.

Frequently Asked Questions about skill-optimizer

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I improve an underperforming AI agent skill file?▼

Diagnose the root cause first, then apply 1-4 targeted micro-edits to the optimizable sections rather than rewriting everything. Validate each change against an eval set of test prompts and commit only if the score improves.

How to test prompt changes before deploying them?▼

Build an eval set of 6-12 representative prompts with weighted scoring rubrics covering nominal, edge, and regression cases. Run the baseline, apply your edit, re-run the eval set, and compare aggregate scores to decide commit or revert.

What is the difference between skill-creator and skill-optimizer?▼

skill-creator builds a new skill from scratch, while skill-optimizer improves an existing SKILL.md through small validated edits. Use the optimizer when the skill already exists but produces mediocre or inconsistent results.

Why does my agent skill drift from its original purpose?▼

Drift happens when successive edits lack a protected zone separating fundamental objectives from adjustable content. Mark INVARIANTS sections that must never change automatically, and only optimize hints, examples, and phrasing.

When should I revert a prompt modification?▼

Revert when the eval set score shows a null or negative delta compared to the previous iteration. If the score improves but regressions appear on previously passing tests, discuss the tradeoff with the user before committing.

Related Skills

Explore More Software Engineering Skills→
tech-leads-clubtech-leads-club
5.0k

cursor-skill-creator

Scaffold Cursor skills with SKILL.md frontmatter under .cursor/skills/.

Official
Basic
CHENyiru3CHENyiru3

skill-creator

Automate end-to-end creation, evaluation, and refinement of Claude Code skills.

Community
Advanced
getsentrygetsentry
905

prompt-optimizer

Optimize agent prompts across OpenAI, Claude, and Gemini with eval-driven refinements.

Official
Advanced
JHostalekJHostalek
10

transformer

Rewrite existing SKILL.md instructions for frontier reasoning model calibration.

Community
Advanced