Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →连接智合AI法律大模型平台进行法律研究。本技能应在用户需要进行法律问题研究、查找法律法规、检索类似案例、或获取法律研究报告时使用。需要智合AI平台会员账号。
.claude/skills/thomasmoreai-zhihe-legal-research/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 29% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 160% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 17% | 0% |
| case-21 | ✗→✓ | ▲ Improved | 89% | 0% |
连接智合AI法律大模型平台，提供专业的法律调研分析服务。
法律研究是异步长任务（3-10分钟）。采用用户主动查询机制，兼容所有 AI Agent 平台（Claude Code、OpenClaw 等）：
用户提交 → 获得任务 ID → 用户稍后主动查询 → 获取结果并归档核心原则：
bash./scripts/auth.sh check
is_vip: true → 已登录，继续code: 401 → 需要登录（执行步骤 2）bash# 发送验证码（如果已保存手机号，可省略手机号参数） ./scripts/auth.sh send-code [手机号] # 验证登录（自动保存 Token 和手机号） ./scripts/auth.sh verify <手机号> <6位验证码>
自动重登流程：如果 config 中已保存 LEGAL_RESEARCH_PHONE，token 失效时可直接执行 ./scripts/auth.sh send-code（无需传入手机号），系统自动使用保存的手机号发送验证码。用户只需提供验证码即可完成重登。
bash# 提交问题 ./scripts/research.sh submit "<用户的法律问题>"
提交成功后，记录 task_id 并告知用户：
> ✅ 您的法律问题已提交，后台正在进行调研分析。 > ⏱️ 预计需要 3-4 分钟完成。 > 📋 任务 ID：{task_id} > > 👉 请在约 4 分钟后回复"查看结果"或"研究结果好了吗"来获取分析报告。
触发条件： 用户主动询问研究结果（如"查看结果"、"结果出来了吗"等）。
如果用户提供了 task_id，直接查询；如果未提供，先查历史获取最近任务：
bash# 如果没有 task_id，先查历史 ./scripts/research.sh history 1 3
然后查询状态：
bash# 查询状态 ./scripts/research.sh status <task_id>
根据状态处理：
| 状态 | 处理方式 | |------|----------| | completed | 获取结果 → 获取报告 → 归档 → 展示给用户 | | running | 告知用户继续等待 1-2 分钟后再查询 | | pending | 告知用户仍在排队中，稍后再查 | | failed | 告知用户失败原因，建议重新提交 | | timeout | 告知用户超时，建议简化问题重试 |
当状态为 completed 时，依次执行：
bash# 1. 获取文字结果 ./scripts/research.sh result <task_id> # 2. 获取报告下载链接 ./scripts/research.sh report <task_id> # 3. 自动归档（下载报告到 archive/ 目录，含 Markdown 转换） ./scripts/research.sh archive <task_id>
归档完成后，展示研究结果和报告链接给用户。
仅在 Claude Code 环境下可用。 提供更好的异步体验，无需用户手动查询。
提交任务后，使用 Bash 工具的 run_in_background: true 启动后台监控：
bashcommand: "./scripts/monitor.sh monitor <task_id> 600 30" run_in_background: true timeout: 600000
关键注意事项：
archive/ 目录OpenClaw 环境不支持此增强模式，请使用标准流程（步骤 1-4）。
| 命令 | 用途 | |------|------| | ./scripts/auth.sh check | 检查登录状态 | | ./scripts/auth.sh send-code [phone] | 发送验证码（省略手机号时使用已保存的号码） | | ./scripts/auth.sh verify <phone> <code> | 验证登录（自动保存 Token 和手机号） | | ./scripts/auth.sh logout | 清除凭证 |
| 命令 | 用途 | |------|------| | ./scripts/research.sh submit "<query>" | 提交问题 | | ./scripts/research.sh status <task_id> | 查询状态 | | ./scripts/research.sh result <task_id> | 获取结果 | | ./scripts/research.sh report <task_id> | 获取报告链接 | | ./scripts/research.sh archive <task_id> | 归档研究结果 | | ./scripts/research.sh history [page] [size] | 查看历史任务 |
| 命令 | 用途 | |------|------| | ./scripts/monitor.sh monitor <task_id> [timeout] [interval] | 阻塞监控 | | ./scripts/monitor.sh status | 查看监控状态 | | ./scripts/monitor.sh results | 获取已完成待通知的结果 | | ./scripts/monitor.sh clear <task_id> | 标记为已通知 |
| 状态 | 说明 | 处理 | |------|------|------| | pending | 排队中 | 告知用户等待 | | running | 处理中 | 告知用户继续等待 | | completed | 已完成 | 获取结果 → 归档 → 通知用户 | | failed | 失败 | 通知用户失败原因 | | timeout | 超时 | 通知用户超时 |
所有配置文件自包含在 skill 内部：assets/
| 文件 | 用途 | |------|------| | .env | Token 和手机号配置（已加入 .gitignore） | | pending.json | 待处理任务 | | completed.json | 已完成待通知 | | notified.json | 已通知历史 |
注意：assets/.env 已加入 .gitignore，敏感信息不会被提交到 git。
配置示例见 assets/.env.example
任务完成时自动归档：归档时会自动下载报告并保存到 archive/ 目录。
归档命名格式：YYMMDD 主题_法律研究报告
示例：
archive/
├── 260326 美术作品著作权侵权纠纷_法律研究报告/
│ ├── result.md # 研究结果摘要（Markdown）
│ ├── report.docx # 详细报告（自动下载）
│ ├── report.md # 报告 Markdown 版本（需安装 pandoc）
│ └── media/ # 报告中的图片（如有）
└── 260310 劳动合同解除赔偿_法律研究报告/
└── result.mdbash# 归档研究结果（自动下载报告并转换为 Markdown） ./scripts/research.sh archive <task_id> # 列出所有归档 ./scripts/research.sh list-archive
归档时会自动尝试将 docx 报告转换为 Markdown：
media/ 子目录bash# macOS 安装 pandoc brew install pandoc
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | pass→pass | 28,815 | 20,998 | -27% | 1 | 1 | 0% | 3,710 | 5,414 | +46% | 0 | 0 | — |
case-11 | pass→pass | 10,561 | 3,757 | -64% | 1 | 1 | 0% | 1,719 | 2,642 | +54% | 0 | 0 | — |
case-17 | pass→pass | 4,991 | 5,093 | +2% | 1 | 1 | 0% | 797 | 2,893 | +263% | 0 | 0 | — |
case-01 | fail→fail | 17,006 | 4,435 | -74% | 1 | 1 | 0% | 2,779 | 2,423 | -13% | 0 | 0 | — |
case-02 | fail→fail | 18,374 | 6,009 | -67% | 1 | 1 | 0% | 2,719 | 2,424 | -11% | 0 | 0 | — |
case-03 | fail→fail | 17,394 | 4,660 | -73% | 1 | 1 | 0% | 2,707 | 2,345 | -13% | 0 | 0 | — |
case-05 | pass→fail | 10,274 | 3,973 | -61% | 1 | 1 | 0% | 1,964 | 2,308 | +18% | 0 | 0 | — |
case-06 | pass→pass | 14,576 | 4,975 | -66% | 1 | 1 | 0% | 2,789 | 2,779 | -0% | 0 | 0 | — |
case-07 | fail→fail | 6,292 | 3,691 | -41% | 1 | 1 | 0% | 977 | 2,256 | +131% | 0 | 0 | — |
case-08 | fail→pass | 13,473 | 3,298 | -76% | 1 | 1 | 0% | 2,228 | 2,562 | +15% | 0 | 0 | — |
case-09 | fail→fail | 9,093 | 1,707 | -81% | 1 | 1 | 0% | 1,857 | 2,377 | +28% | 0 | 0 | — |
case-10 | fail→pass | 11,219 | 5,563 | -50% | 1 | 1 | 0% | 1,970 | 2,550 | +29% | 0 | 0 | — |
case-12 | pass→pass | 9,884 | 5,146 | -48% | 1 | 1 | 0% | 1,548 | 2,924 | +89% | 0 | 0 | — |
case-13 | fail→fail | 8,286 | 3,750 | -55% | 1 | 1 | 0% | 1,466 | 2,737 | +87% | 0 | 0 | — |
case-14 | fail→pass | 6,879 | 9,079 | +32% | 1 | 1 | 0% | 1,067 | 2,775 | +160% | 0 | 0 | — |
case-15 | pass→pass | 11,284 | 3,826 | -66% | 1 | 1 | 0% | 2,142 | 2,891 | +35% | 0 | 0 | — |
case-16 | pass→pass | 14,312 | 8,870 | -38% | 1 | 1 | 0% | 2,482 | 3,288 | +32% | 0 | 0 | — |
case-18 | fail→fail | 4,892 | 4,271 | -13% | 1 | 1 | 0% | 790 | 2,258 | +186% | 0 | 0 | — |
case-19 | fail→pass | 11,089 | 1,756 | -84% | 1 | 1 | 0% | 2,054 | 2,401 | +17% | 0 | 0 | — |
case-20 | fail→fail | 3,467 | 4,635 | +34% | 1 | 1 | 0% | 420 | 2,310 | +450% | 0 | 0 | — |
case-21 | fail→pass | 11,856 | 6,082 | -49% | 1 | 1 | 0% | 1,628 | 3,082 | +89% | 0 | 0 | — |
case-22 | fail→pass | 10,529 | 3,665 | -65% | 1 | 1 | 0% | 1,606 | 2,754 | +71% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 15 counted toward the lift figure. The other 7 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +23 percentage points is the difference between those two pass rates over the 15 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.