Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →证据三性审查——对证据清单进行首轮审查，做出明显的 合法性/关联性判断并标记需要律师审查的疑难项目。 当用户说"审查证据清单"、"证据三性审查"、 "检查这些证据的可采性"或有证据清单需要在质证前审核时使用。
.claude/skills/thomasmoreai-privilege-log-review-zhou210712/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-17 | ✗→✓ | ▲ Improved | 53% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 61% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 49% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 40% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 76% | 0% |
~/.claude/plugins/config/claude-for-legal/litigation-legal/CLAUDE.md → 审查规程。证据审查有三种条目：明显具有可采性的、明显不具有可采性的、以及需要判断的。本技能筛查前两种，使律师的时间完全用于第三种。
这是首轮审查。律师审查每个标记。无例外。
中国民事诉讼中，证据审查的核心是三性审查：
法律依据：
[法条原文][法条原文][法条原文]非法证据排除（《民事诉讼法》司法解释第106条）： 对以严重侵害他人合法权益、违反法律禁止性规定或者严重违背公序良俗的方法形成或者获取的证据，不得作为认定案件事实的根据。[法条原文]
~/.claude/plugins/config/claude-for-legal/litigation-legal/CLAUDE.md → 审查规程。
三态规则。本技能绝不沉默地决定主观性判断标准不满足。 在任何不确定的判断上——合法性存疑、关联性边界模糊——技能保留"待审查"标记并添加 ⚠️ 标记给律师。错误地标记为可采是单向门；过度标记由律师在审查中纠正（双向门）。倾向可恢复的错误。
以下为默认情况。技能不基于自己的主观判断排除证据。示例：
[法条原文]。保留；标记律师判断。每个标记记录具体的开放问题和证据的两面，使律师无需从头阅读即可决定。
如果任何一项是临界的——录制内容是否构成严重侵害——则属于不确定，不是❌。
证据清单是否包含必要字段？
| 字段 | 有无？ | |---|---| | 编号 | | | 证据名称 | | | 证据来源 | | | 证明内容 | | | 证据形式（书证/物证/电子数据/证人证言等） | | | 是否为原件/原物 | |
缺失字段 → 标记需在实质性审查前补全。
对于每个条目：
条目[N]：[✅ 可采 | ✅ + ⚠️ 保留 & 标记 | ❌ 建议排除]
[如✅：一行理由]
[如✅ + ⚠️：保留；律师需要回答的具体问题；两面证据]
[如❌：一行理由——但条目保留在清单上直到律师决定移除]跨整个清单：
markdown[工作成果标题] ## 证据三性审查：[案件] — [日期] **适用规则：** 《民事诉讼法》第66条、第65条 `[法条原文]` + 司法解释第104-106条 `[法条原文]` **已审查条目：** [N] **结果：** [N] ✅ 可采 / [N] ✅+⚠️ 保留并标记 / [N] ❌ 建议排除 ### ✅ + ⚠️ 已标记 —— 律师决定 | 条目 | 证据编号 | 问题 | 支持可采 | 反对可采 | 待判断 | |---|---|---|---|---|---| ### ❌ 建议排除（律师确认后移除） | 条目 | 证据编号 | 理由 | |---|---|---| ### ✅ 可采（无需行动） [计数。可按需提供列表。] ### 模式观察 [重复问题、过度提交、描述问题] --- **律师必须在任何行动前审查所有 ⚠️ 和 ❌。**
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-17 | fail→pass | 14,560 | 11,197 | -23% | 1 | 1 | 0% | 2,146 | 3,285 | +53% | 0 | 0 | — |
case-18 | fail→fail | 23,249 | 18,749 | -19% | 1 | 1 | 0% | 3,342 | 4,445 | +33% | 0 | 0 | — |
case-04 | pass→pass | 18,208 | 19,555 | +7% | 1 | 1 | 0% | 3,091 | 4,698 | +52% | 0 | 0 | — |
case-05 | pass→pass | 15,508 | 17,908 | +15% | 1 | 1 | 0% | 2,943 | 5,049 | +72% | 0 | 0 | — |
case-06 | pass→pass | 20,504 | 24,797 | +21% | 1 | 1 | 0% | 3,199 | 5,426 | +70% | 0 | 0 | — |
case-01 | fail→pass | 19,973 | 29,043 | +45% | 1 | 1 | 0% | 3,197 | 5,147 | +61% | 0 | 0 | — |
case-02 | fail→pass | 22,975 | 25,491 | +11% | 1 | 1 | 0% | 3,656 | 5,431 | +49% | 0 | 0 | — |
case-03 | fail→pass | 23,767 | 24,077 | +1% | 1 | 1 | 0% | 3,799 | 5,309 | +40% | 0 | 0 | — |
case-07 | fail→pass | 16,074 | 18,533 | +15% | 1 | 1 | 0% | 2,547 | 4,487 | +76% | 0 | 0 | — |
case-08 | fail→pass | 20,120 | 15,000 | -25% | 1 | 1 | 0% | 3,181 | 3,965 | +25% | 0 | 0 | — |
case-09 | fail→fail | 17,567 | 18,081 | +3% | 1 | 1 | 0% | 2,790 | 4,334 | +55% | 0 | 0 | — |
case-10 | pass→pass | 11,609 | 10,511 | -9% | 1 | 1 | 0% | 1,959 | 3,347 | +71% | 0 | 0 | — |
case-11 | pass→pass | 9,418 | 6,612 | -30% | 1 | 1 | 0% | 1,511 | 2,748 | +82% | 0 | 0 | — |
case-12 | pass→pass | 18,596 | 19,748 | +6% | 1 | 1 | 0% | 2,584 | 3,863 | +49% | 0 | 0 | — |
case-13 | fail→pass | 16,167 | 15,120 | -6% | 1 | 1 | 0% | 2,432 | 3,909 | +61% | 0 | 0 | — |
case-14 | pass→pass | 16,061 | 16,460 | +2% | 1 | 1 | 0% | 2,453 | 4,259 | +74% | 0 | 0 | — |
case-15 | pass→pass | 13,447 | 16,205 | +21% | 1 | 1 | 0% | 2,153 | 4,190 | +95% | 0 | 0 | — |
case-16 | fail→fail | 19,169 | 26,567 | +39% | 1 | 1 | 0% | 3,082 | 5,176 | +68% | 0 | 0 | — |
case-19 | fail→pass | 20,073 | 15,283 | -24% | 1 | 1 | 0% | 2,346 | 3,907 | +67% | 0 | 0 | — |
case-20 | pass→pass | 16,718 | 15,953 | -5% | 1 | 1 | 0% | 2,697 | 3,633 | +35% | 0 | 0 | — |
case-21 | fail→pass | 11,560 | 14,109 | +22% | 1 | 1 | 0% | 1,508 | 3,755 | +149% | 0 | 0 | — |
case-22 | fail→pass | 24,634 | 21,884 | -11% | 1 | 1 | 0% | 3,880 | 5,143 | +33% | 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. The headline lift of +45 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.