Matter Context
Matter context. Check the ## 事项工作区in the practice-level CLAUDE.md. If Enabled is ✗(the default for enterprise legal users), skip the rest of this section—the skill uses practice-level context and the matter mechanism is not visible. If it is enabled and there is no active matter, ask: "Which matter is this? Run /corporate-legal:matter-workspace switch <事项简称> or say 实务级." Load the active matter's matter.md to get matter-specific context and override rules. Output is written to the matter folder ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/matters/<事项简称>/. Never read files from other matters unless 跨事项上下文 is 开.
Purpose
AI-assisted review tools are good at one thing: reading 500 contracts and finding every change-of-control clause. They are not good at judgment—deciding whether a particular change-of-control clause is actually triggered by this deal structure.
This skill hands off bulk extraction to the appropriate tool, then runs a QA layer on the returned results.
Before handing off: First try tabular-review（/corporate-legal:tabular-review). For anything the user's environment can handle—a few hundred documents, a defined column schema—the native spreadsheet review setup is faster, has no per-document billing cost, and keeps the work product local. Hand off to an AI tool only when the corpus is truly too large, the team already has a license and workflow, or the matter requires a tool with a verified provenance chain.
Load Context
~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md → AI-Assisted Review:
- Tool used (Luminance / Kira / none)
- Used for what (which clause types)
- Trust tier (use directly / spot-check / full review)
- Handoff process (who loads, who QAs)
If ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md shows no AI tool → this skill is a no-op. Everything is handled directly through diligence-issue-extraction.
When to Hand Off
Hand off when all of the following are met:
- The category has more than 50 documents (fewer than 50, direct reading is faster)
- The extraction target is a clause type the tool is good at (change of control, contract assignment, exclusivity, most-favored-nation, termination, automatic renewal)
- The documents are relatively uniform (all customer contracts with similar text—not a mix of contracts, letters, and board minutes)
Do not hand off:
- Custom or heavily negotiated documents
- Side letters and amendment agreements (highly context-dependent; tools miss interactions with the main agreement)
- Any situation where the question is "what does this mean for the deal" rather than "does this clause exist"
Handoff
Step 1: Prepare the batch
- Identify the batch documents from the data room index
- According to
~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.mdspecify the extraction target (which clause types) - Note the materiality threshold to filter tool output
Step 2: Load (or instruct the loader)
According to ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md —who loads. If it is you, generate loading instructions. If it is someone else, generate a request:
## [工具] 加载请求 — [交易代码] — [类别]
**文档：** [N]份 来自数据室文件夹 [路径]
**加载至：** [工具工作区/事项]
**提取目标：**
- 控制权变更 / 合同转让
- 独家性
- [等——按 `~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md`]
**过滤输出：** 仅标记提取目标存在的情况——无需为每份文档报告"未发现控制权变更条款"。
**返回截止：** [日期]
Step 3: QA the output
When the tool returns results, apply according to the trust tier:
"Use directly": Enter due diligence findings directly. (Only when ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md is set this way—this situation is rare.)
"Spot-check X%": Randomly select X% of the tagged documents. For each, read the actual clause and compare it with the tool's extraction. If the error rate is low, accept the batch. If errors are found, expand the sample.
"Full manual review of tagged items": The tool narrows the scope (500 documents → 80 containing change-of-control clauses). Manually read all 80. The tool saved the time of reading 420 clean documents.
Step 4: Judgment Layer
The tool found the clauses. Now apply judgment:
For each tagged change-of-control clause: is it actually triggered by this transaction?
- Equity transfer vs. asset transfer vs. merger—different triggering conditions
- How is "change of control" defined in the contract—majority equity? Board control? Other?
- Are there exception clauses for this type of transaction?
This is the part the tool cannot do. The output goes into due diligence findings in an internal format.
Output
The following QA summary is derived from data room documents that are privileged, confidential, or both. It inherits the privileged and confidential status of the source documents—distribution outside the privileged circle may waive privilege. Store in the matter's privileged file.
## AI 工具交接摘要 — [类别]
**工具：** [Luminance / Kira]
**已处理文档：** [N]
**提取目标：** [条款类型]
### QA
**信任层级：** [按 `~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md`]
**样本量：** [N]份文档抽查
**错误率：** [X]% — [接受 / 扩大样本 / 触发全面复核]
### 结果
| 条款类型 | 工具标记文档 | 经判断层后 | 重大 |
|---|---|---|---|
| 控制权变更 | [N] | [N 实际被交易结构触发] | [N 超过阈值] |
| 合同转让 | [N] | [N] | [N] |
**→ 已将 [N] 项发现加入尽调问题清单**
**→ 已将 [N] 项同意事项加入交割检查表**
Conclude with the next-step action decision tree
Conclude with the next-step action decision tree in CLAUDE.md ## 输出规范 . Customize the options based on what this skill just produced—the five default branches (draft X, escalate, supplement facts, monitor and wait, other) are a starting point, not locked in. The decision tree itself is the output; the lawyer chooses.
What This Skill Does Not Do
- Does not run Luminance or Kira—it manages handoff and QA. Extraction is run by a human (or the tool's own interface).
- Does not fully replace the tool's output with its own judgment—if
~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.mdsays spot-check 10%, then check 10%, not 100%. - Does not decide the trust tier—this is set in CLAUDE.md, determined during cold start based on the team's experience with the tool.