optimize-skill

仓库创建 2026年6月26日最近提交 21 天前SkillHot 收录 20 天前

⬈ GitHub 仓库 ⬈ SKILL.md 原文件 ⬇ Raw 原始文件

▸ 精选理由

系统化提升技能性能，适合持续改进流程。

▸ 风险提示

实验可能消耗模型调用配额并需保存测试结果

这个 Skill 做什么

循环基准→修改→再基准，直到技能在同一测试上实际提升。

把一个技能变得真正有用的循环保养：先跑 benchmark 找出失败点，读出原因改代码，再在同一套基准上复测，直到新版确实更好。适合有人说“把这个技能做得更好”或要持续优化时用。核心特点是以可量化的测量为导向，反复验证而不是凭感觉改动。

▸ 展开 SKILL.md 英文原文

Improve a skill until it measurably pays off — the loop that benchmarks a skill, reads where it fails, edits it, and re-benchmarks on the same ruler until a stronger revision wins. Use when asked to "improve / optimize / develop a skill", "make this skill better", "search for a better version of the skill", or to run a skill-improvement loop. Drives `/benchmark-skill` as its measurement subroutine — reach for that one directly when you only need to measure a fixed skill, not change it.

Skill 开发管理优化迭代基准实验通用

Stars

Forks

仓库内 Skill

7 日增星

安装 / 使用

给你的 Agent 一句话（通用）

帮我安装这个 skill：https://raw.githubusercontent.com/Svendsys/agent-skills/master/skills/optimize-skill/SKILL.md

或 curl 直取 SKILL.md

curl -fsSL "https://raw.githubusercontent.com/Svendsys/agent-skills/master/skills/optimize-skill/SKILL.md"

SKILL.MD 节选查看完整文件 ↗

# Optimize a skill

Benchmarking measures a skill; **optimizing is the goal that uses it.** A benchmark tells
you whether one frozen revision pays off and where it fails; this loop turns that signal
into a *better* skill — read the failures, form a hypothesis, edit the skill, measure
again on the same ruler, keep what wins. The instrument is `/benchmark-skill`; everything
here is about spending it well and knowing when you've converged.

The deliverable is a **stronger skill revision**, backed by a benchmark cohort that shows
it beats the one you started from. Keep the two operations distinct: you never change how
a benchmark runs to move a number — you change the skill, and let the unchange

via SKILL·HOT · 数据来自 GitHub 公开信息 · 原文版权归作者所有