← Back to Skills Marketplace
Skill Quality Scorer
by
Chris1Wang3
· GitHub ↗
· v1.1.7
· MIT-0
48
Downloads
1
Stars
0
Active Installs
3
Versions
Install in OpenClaw
/install skill-quality-scorer
Description
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...
README (SKILL.md)
\r \r
Skill Quality Scorer · 技能质量评分器\r
\r
EN TRACE+ (T-R-F-S-I-E) × 30 sub-items · static script first · formula score · JSON + Markdown. \r 中文 TRACE+ 六维 30 子项 · 先 static_audit · 公式算分 · JSON + Markdown。\r \r When / 何时用： 迭代没方向 · 对标同类 · 发布前自检 · 批量评
skills/\r Not / 不用： 从零写 Skill · 替代 skill-eval 行为实验（E 维默认static_proxy）\r \r
Score portfolio-doctor with TRACE+ — where vs skill-reviewer?\r
给 portfolio-doctor 做 TRACE+ 全维评分。\r
```\r
\r
## 评测模式\r
\r
| 模式 | 输出 |\r
|------|------|\r
| 单个 | JSON + Markdown（[audit-playbook](references/audit-playbook.md)） |\r
| A vs B | 两份 JSON + 分差表 + 推荐 |\r
| 批量 | 汇总表 + 各 skill 简评 |\r
\r
对比/批量：**同一 rubric v2**，不得换公式或跳过子项。\r
\r
## 工作流\r
\r
```text\r
1) 定位 skill 目录（对比/批量则逐个重复 2–6）\r
2) python scripts/static_audit.py "\x3Cskill-dir>" → auto_scores（不可改分）\r
3) Read scoring-engine-deterministic.md → 30 子项 evidence\r
4) Read 目标 SKILL.md + 链接的 references/scripts\r
5) composite = round((T+R+F+S+I+E)×100/60, 1) → 评级 + Verdict\r
6) 按 audit-playbook 输出（含 F 维触发测试各 ≥3 条）\r
```\r
\r
评级 / Verdict / 30 子项定义 / JSON schema → [scoring-engine-deterministic.md](references/scoring-engine-deterministic.md)\r
\r
## 硬约束\r
\r
1. 先脚本后 rubric；`auto_scores` 只补 evidence 不改分\r
2. 30 子项逐项 evidence；禁止旧公式 `(T+R+A+C+E)×2`\r
3. E 维默认 `static_proxy`；有 skill-eval 时切换 `behavioral_eval`\r
4. **Rigid**：子项、公式、Verdict 不可改 · **Flexible**：evidence 表述、Top 修复排序\r
\r
## 参考文件\r
\r
| 文件 | 内容 |\r
|------|------|\r
| [references/scoring-engine-deterministic.md](references/scoring-engine-deterministic.md) | 30 子项 rubric · 公式 · JSON schema |\r
| [references/audit-playbook.md](references/audit-playbook.md) | 报告模板 · 对比/批量 · 触发测试 |\r
| [examples/sample-score-v2.json](examples/sample-score-v2.json) | JSON 样例 |\r
| [scripts/static_audit.py](scripts/static_audit.py) | 静态审计（唯一运行时脚本） |\r
Usage Guidance
Install only if you want an agent to inspect local skill directories for quality scoring. Run it on directories you intentionally choose, avoid pointing it at unrelated private workspaces, and treat its scoring as an advisory review aid rather than a substitute for behavioral testing or human security review.
Capability Tags
Capability Assessment
Purpose & Capability
The stated purpose is to score Agent Skills using a TRACE+ rubric, and the artifacts match that purpose: SKILL.md defines the workflow, references provide the 30-item rubric and report templates, and the Python script performs local static checks used as inputs.
Instruction Scope
The skill instructs the agent to read a target skill directory, linked references, and scripts, and to run `python scripts/static_audit.py <skill-dir>`. That file access is broad if the user points it at a large or private skill tree, but it is disclosed and directly tied to the scoring task.
Install Mechanism
No package installation, network bootstrap, allowlist edits, service restarts, or external dependency setup are present. The only executable component is a bundled Python script using standard-library parsing and local file reads.
Credentials
Reviewing skills can expose local skill contents, including any credentials accidentally stored in those files. This is proportionate for a skill auditor, but users should choose the target directory deliberately.
Persistence & Privilege
The artifacts show no background worker, persistence mechanism, privilege escalation, filesystem mutation, network calls, or credential/session-store use.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install skill-quality-scorer - After installation, invoke the skill by name or use
/skill-quality-scorer - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.1.7
- No user-facing or functional changes in this version.
- SKILL.md remains unchanged; no updates to features, workflow, or documentation.
- Version bump only; existing behavior and scoring logic are retained.
v1.1.6
- Removed redundant documentation files: README.md, README-zh.md, and skill-card.md.
- Simplified and clarified SKILL.md, focusing on workflow, evaluation modes, and constraints.
- Added concise bilingual (EN/中文) summary and clearer usage guidance.
- Centralized details (e.g., rubric definition, formula, schema) into reference files.
- No logic or code changes; documentation update only.
v1.1.5
- Added detailed documentation on the 6-dimension, 30-subitem TRACE+ quality scoring system, including formulas, workflows, and rigid evaluation rules.
- Clarified supported evaluation modes: single, comparison (A vs B), and batch, with required output formats.
- Specified static_audit.py as the mandatory first step; manual scoring must follow automated results and standardized rubric.
- Defined strict pass/fail/conditional verdicts and grade thresholds based on composite scores and critical flags.
- Enumerated fixed constraints and provided references for all rule sources, scoring schemas, and template formats.
Metadata
Frequently Asked Questions
What is 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... It is an AI Agent Skill for Claude Code / OpenClaw, with 48 downloads so far.
How do I install Skill Quality Scorer?
Run "/install skill-quality-scorer" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Skill Quality Scorer free?
Yes, Skill Quality Scorer is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Skill Quality Scorer support?
Skill Quality Scorer is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Skill Quality Scorer?
It is built and maintained by Chris1Wang3 (@chris1wang3); the current version is v1.1.7.
More Skills