Open-source repository for production-ready AI agent skills
面向实战的 AI Agent Skills 开源仓库
This repository is dedicated to sharing reusable, high-quality agent skills. Current release includes:
senior-legal-contract-reviewer-v1.3(latest)senior-legal-contract-reviewer-v1.2(baseline)- Jurisdiction: PRC law (Mainland China)
- Focus: context-first contract review, Plan B / Plan C clause design, expected loss (EL) estimation
- Load
skills/senior-legal-contract-reviewer-v1.3/SKILL.mdas your agent system instruction. - Provide contract text/snippets plus role, jurisdiction, business objective, red lines, and amount range.
- Parse structured output (risk level, vulnerability, Plan B / Plan C clauses, citations, EL estimation).
中文:以下数据来自仓库内 20 个 PRC 合同风险样本、4 组配置、每组 100 次自测。
English: The following numbers come from the in-repo PRC contract benchmark with 20 cases, 4 groups, and 100 runs per group.
| Group | Mean Avg Score | Mean Pass Rate |
|---|---|---|
| Layman Prompt | 28.09 | 5.15% |
| Lawyer Prompt | 72.47 | 49.25% |
| v1.2 | 88.09 | 75.75% |
| v1.3 | 93.99 | 87.15% |
v1.3 vs v1.2:+5.90mean score,+11.40%pass rate- Main driver: context-first questioning about business objective, red lines, tradeables, and BATNA before clause rewriting
xychart-beta
title "Benchmark Mean Avg Score"
x-axis ["Layman","Lawyer","v1.2","v1.3"]
y-axis "Score" 0 --> 100
bar [28.09, 72.47, 88.09, 93.99]
xychart-beta
title "Benchmark Mean Pass Rate (%)"
x-axis ["Layman","Lawyer","v1.2","v1.3"]
y-axis "Pass Rate" 0 --> 100
bar [5.15, 49.25, 75.75, 87.15]
agent-skills-hub/
├── README.md
├── README.zh-CN.md
├── LICENSE
├── benchmark/
│ └── data/
│ └── test_cases/
├── examples/
│ ├── case_a_input.md
│ └── case_a_output.json
└── skills/
├── senior-legal-contract-reviewer-v1.2/
│ ├── SKILL.md
│ └── README.md
└── senior-legal-contract-reviewer-v1.3/
├── SKILL.md
└── README.md
- Main skill:
SKILL.md - Intro doc:
README.md - Benchmark report:
reports/v1.3-comparison-100-runs.md
- Main skill:
SKILL.md - Intro doc:
README.md - Example input:
examples/case_a_input.md - Example output:
examples/case_a_output.json
- v1.3 100-run comparison:
reports/v1.3-comparison-100-runs.md - Benchmark dataset:
benchmark/data/test_cases/README.md - Official single-round comparison:
reports/v1.2-official-comparison.md - 100-run self-test summary:
reports/v1.2-self-test-100-runs.md - Raw data (JSON):
- This repository provides drafting and risk-review support, not formal legal advice.
- For production use, legal outputs should be reviewed by licensed lawyers.