Skill Quality Scorer
Deterministic TRACE+ quality scorer for Agent Skills (SKILL.md): six dimensions T-R-F-S-I-E, 30 sub-items, merges TRACE, good-skill authoring reverse-rubric...
Chris Wang
@chris1wang3
What This Skill Does
Deterministic quality scorer for Agent Skills that evaluates SKILL.md files across six dimensions (T-R-F-S-I-E) with 30 sub-items, runs a static audit script first, then outputs a composite score in JSON and Markdown formats.
Replaces subjective manual review of skill documentation by providing a consistent, formula-based scoring rubric with automated evidence collection.
When to Use It
- Score a single SKILL.md file to check compliance with quality standards
- Compare two skills side-by-side to identify which has higher quality
- Batch audit an entire skills directory to generate a summary report
- Identify specific improvement areas for a skill with a low SkillHub score
- Validate a skill before publishing to ensure it meets the TRACE+ rubric
Install
$ openclaw skills install @chris1wang3/skill-quality-scorerSkill Quality Scorer · 技能质量评分器
EN TRACE+ (T-R-F-S-I-E) × 30 sub-items · static script first · formula score · JSON + Markdown.
中文 TRACE+ 六维 30 子项 · 先 static_audit · 公式算分 · JSON + Markdown。
When / 何时用: 迭代没方向 · 对标同类 · 发布前自检 · 批量评 skills/
Not / 不用: 从零写 Skill · 替代 skill-eval 行为实验(E 维默认 static_proxy)
Score portfolio-doctor with TRACE+ — where vs skill-reviewer?
给 portfolio-doctor 做 TRACE+ 全维评分。
可直接触发的说法
以下口语输入都应触发本技能:
- 给这个 skill 打分。
- 看看这个技能哪里不符合规范。
- 用 TRACE+ 评一下这个 SKILL.md。
- 这两个技能哪个质量更高?
- 批量扫一下这个 skills 目录。
- SkillHub 分数低,帮我找提分项。
- 和 skill-reviewer 差在哪?
最小可用输入
最低可启动:一个目标 skill 目录,或一个 SKILL.md 文件路径。
推荐输入:目标路径、评分模式(单个/A vs B/批量)、是否有行为评测结果、期望输出格式(JSON / Markdown / 两者)。
可兼容输入:
- 只给目录:自动定位目录下
SKILL.md。 - 只给父目录:按批量模式逐个扫描子目录。
- 只给两条路径:按 A vs B 对比模式。
- 没有 behavior eval:E 维使用
static_proxy,E3 不强行给满。
评测模式
| 模式 | 输出 |
|---|---|
| 单个 | JSON + Markdown(audit-playbook) |
| A vs B | 两份 JSON + 分差表 + 推荐 |
| 批量 | 汇总表 + 各 skill 简评 |
评分前先确认输出格式:JSON、Markdown 或两者。用户未指定时,默认输出 JSON + Markdown。
对比/批量:同一 rubric v2,不得换公式或跳过子项。
工作流
1) 定位 skill 目录(对比/批量则逐个重复 2–6)
2) python scripts/static_audit.py "<skill-dir>" → auto_scores(不可改分)
3) Read scoring-engine-deterministic.md → 30 子项 evidence
4) Read 目标 SKILL.md + 链接的 references/scripts
5) composite = round((T+R+F+S+I+E)×100/60, 1) → 评级 + Verdict
6) 按 audit-playbook 输出(含 F 维触发测试各 ≥3 条)
评级 / Verdict / 30 子项定义 / JSON schema → scoring-engine-deterministic.md
硬约束
- 先脚本后 rubric;
auto_scores只补 evidence 不改分 - 30 子项逐项 evidence;禁止旧公式
(T+R+A+C+E)×2 - E 维默认
static_proxy;有 skill-eval 时切换behavioral_eval - Rigid:子项、公式、Verdict 不可改 · Flexible:evidence 表述、Top 修复排序
验收与失败路径
- 路径不存在:先要求用户提供正确路径,不得凭名称臆测
- 缺少 SKILL.md:判定目标不是标准 skill 包,输出结构问题而非强行评分
- 批量模式:每个 skill 必须使用同一 rubric v2 和同一公式
- 无行为评测:明确标注
effectiveness_mode: static_proxy - 完成标准:JSON 算术校验通过,Markdown 报告含 Top 修复项和触发测试
参考文件
| 文件 | 内容 |
|---|---|
| references/scoring-engine-deterministic.md | 30 子项 rubric · 公式 · JSON schema |
| references/audit-playbook.md | 技能质量审计方法手册:报告模板 · 对比/批量 · 触发测试 |
| examples/sample-score-v2.json | JSON 样例 |
| scripts/static_audit.py | 静态审计(唯一运行时脚本) |
Top skills in this category
self-improving agent
@pskoettCaptures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Claude ('No, that's wrong...', 'Actually...'), (3) User requests a capability that doesn't exist, (4) An external API or tool fails, (5) Claude rea
Skill Vetter
@spclaudehomeSecurity-first skill vetting for AI agents. Use before installing any skill from ClawdHub, GitHub, or other sources. Checks for red flags, permission scope, and suspicious patterns.
Self-Improving + Proactive Agent
@ivangdavilaSelf-reflection + Self-criticism + Self-learning + Self-organizing memory. Agent evaluates its own work, catches mistakes, and improves permanently. Use when...
Proactive Agent
@halthelobsterTransform AI agents from task-followers into proactive partners that anticipate needs and continuously improve. Now with WAL Protocol, Working Buffer, Autonomous Crons, and battle-tested patterns. Part of the Hal Stack 🦞
Agent Browser
@matrixyHeadless browser automation CLI optimized for AI agents with accessibility tree snapshots and ref-based element selection