diligence-issue-extraction

/diligence-issue-extraction

Load ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md + ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/deals/[代码]/deal-context.md。
Use the following workflow.
Check ai-tool-handoff — if the category is high-volume and tools are configured, hand off first.
Read documents, apply materiality filtering, extract by category.
Present findings in internal memo format. Hand off agreed items to the closing checklist.

Matter context

Matter context. Check the practice-level CLAUDE.md for ## 事项工作区. If Enabled is ✗(the default for corporate legal users), skip the rest of this section — the skill uses practice-level context and matter mechanics are invisible. If 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

The data room has 2,000 documents. About 30 of them are material to the deal. This skill reviews documents by the due diligence categories and materiality thresholds in ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md , extracts issues, and writes them in internal memo format.

Load context

~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md → due diligence structure (categories, materiality thresholds)
~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md → issue memo format (how findings are stated)
~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/deals/[代码]/deal-context.md → deal-specific thresholds, data room location

If deal-context.md does not exist, ask which deal this is.

Workflow

Step 1: Inventory the data room

If data room MCP (Data Room/Feishu/Nutstore) is connected, pull the index. Map data room folders to due diligence request list categories. Flag gaps — request list categories with no corresponding data room content.

## 数据室盘存：[交易代码]

| 需求类别 | 数据室文件夹 | 文件数 | 状态 |
|---|---|---|---|
| 公司及组织 | /01-公司文件 | 45 | 已审查 |
| 重大合同 | /02-合同 | 312 | 进行中 |
| 知识产权 | /03-知识产权 | 89 | 未开始 |
| [等] | | | |

**缺口：** [没有数据室内容的需求类别——需发送后续请求]

Step 2: Apply materiality filtering

By the thresholds in ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md / deal context. If the threshold says contracts over ¥X, do not review all of them.

For contracts specifically: rank by stated amount (if in file name/metadata) or by counterparty materiality. Review from the top down until the threshold is reached or the category is exhausted.

Step 3: Extract issues

For each document read, check against the standard due diligence concerns under that category:

Material contracts — standard extraction items:

Change of control provisions (are they triggered by this deal? Is counterparty consent required?)
Contract assignment restrictions (can the contract be transferred to the buyer?)
Exclusivity/non-compete (does it restrict the buyer's business?)
Most favored nation (pricing constraints)
Termination rights (does the counterparty have the right to terminate because of the deal?)
Unusual indemnities or liability assumptions

Corporate — standard extraction items:

Accuracy of equity structure, unexercised options/warrants
Board consent requirements needed for the deal
Shareholder agreement restrictions (drag-along, tag-along, rights of first refusal)
Subsidiary structure and related-party arrangements

Intellectual property — standard extraction items:

Chain of title (are founder/employee assignments in place?)
Open source code in the product (public copyright risk)
Key IP is licensed or owned
Pending or threatened IP litigation

Labor — standard extraction items:

Change-of-control severance triggers (parachute costs)
Key employee retention risk
Pending labor disputes
Classification risk (contractors who look like employees)

Litigation — standard extraction items:

Pending matters and reserves
Threatened claims
Regulatory investigations
Pattern litigation (consumer class actions etc.)

Step 4: State each finding

Source attribution. When a finding cites a statute, regulation, case, or regulatory action — e.g. a change-of-control provision analyzed under applicable law, an IP ownership gap cited under a specific legal doctrine, a pending litigation matter with a case number — label the cited source: for citations retrieved from legal research tools label [yuandian检索] or [北大法宝]/[威科先行]; for web search citations label [联网检索 — 需复核]; for citations recalled from training knowledge label [模型知识 — 需验证]; for citations from data rooms, deal team memos, or outside counsel feedback label [用户提供]. Document source citations (data room path, page number, file name) retain their original index. Citations labeled 需核实 carry higher fabrication risk and should be checked first. Never delete or collapse labels.

When disagreeing with a cited statutory provision, quote the provision's text or decline to describe it. If the user (or deal team notes, or seller disclosure documents) cites a statutory provision in support of their claim and you believe it is incorrect, and the text of that provision cannot be obtained through the connected research tools or data room, do not fabricate a description of that provision. Say: "This provision does not match my expectation for a [bulk sale notice/successor liability/other matter] requirement — the actual text needs to be pulled to determine what it actually covers.[法条未调取 — 需核实]" Then (a) retrieve and cite the original text through the configured research tools, (b) ask the user to paste the text, or (c) flag it for outside counsel judgment. A confident but incorrect description of a real statutory provision is worse than "unclear" — a deal team memo citing a fictitious subsection is harder to correct than a blank. Applies to all skills involving statutory characterization, not just issue extraction.

No silent supplementation. If a search of the configured legal research tools returns few or no results for the legal basis needed for a finding (e.g. rules on change-of-control consent requirements, doctrine on IP assignment, tests for employment relationship classification), report what was found and stop. Do not fill the gap from web search or model knowledge without asking. Say: "The search returned [N] results, from [tool]. Coverage on [rule/doctrine] appears insufficient. Options: (1) broaden the search query, (2) try another research tool, (3) web search — results will be labeled [联网检索 — 需复核] and verified against primary sources before reliance, or (4) flag as unverified and stop. Which would you like?" Whether to accept lower-credibility sources is for the lawyer to decide.

According to the ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md finding template. If the seed memo uses the following format:

问题 #N：[标题]
类别：[需求清单类别]
严重程度：[按内部方案定级]
文件：[数据室路径 + 文件名]
发现：[文件说了什么和为什么重要]
建议：[价格调整 / 赔偿 / 需取得同意 / 陈述与保证 / 退出]

...then follow it exactly. If the seed memo uses bullet points, write bullet points.

Severity calibration(If the internal scheme is red/yellow/green):

🔴 Red: Affects deal value or structure. Change of control requiring key customer consent. Undisclosed material litigation. IP ownership gaps.
🟡 Yellow: Needs attention, can be resolved. Consent required but likely obtainable. Open source code requiring remediation. Worker classification risk.
🟢 Green: Documented for the record. Consistent with the representations and warranties. No action required other than the representations and warranties.

Step 5: Summarize by Category

Group findings by requirements list category. Within each category, sort by severity.

[工作成果页眉 — 按插件配置 ## 输出规范 — 因角色而异；参见 `## 使用者`]

> 本产出来源于具有特权、保密或两者兼有的数据室材料。它继承来源的特权和保密状态——向特权保护圈之外分发可能放弃特权。存放于事项的特权文件中并慎重作出分发决定。

# 尽调问题：[交易代码] — [类别]

**已审文件：**[N] of [M] 本类别
**覆盖：**[全部 | >¥X 阈值 | 前N]
**发现：**[N]🔴 [N]🟡 [N]🟢

---

### 结论

[🔴 N 阻断 · 🟠 N 高 · 🟡 N 中] — [交易团队需要知道的一件事]

---

[每项发现以内部格式]

---

## 缺口

- [无回应文件的需求清单项目]
- [有引用但不在数据室内的文件]

Handoff

To ai-tool-handoff: If, as per ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md , AI tools were used, hand off bulk contract review. This skill handles nuanced documents (side letters, amendment agreements, anything an AI tool struggles with).
To deal-team-summary: The summarized findings feed into the deal team briefing.
To material-contract-schedule: The contract-level extraction feeds into the disclosure schedule.
To closing-checklist: Any finding that implies a standalone action item to be completed before closing becomes a checklist item. Handoff is not limited to third-party consents—it also covers:
- Shareholder votes/other closing actions—Voting power clean-up, required shareholder consents, required board resolutions, dissenters' appraisal rights notice period, conversion mechanics, or any other corporate approval required for the closing of the transaction. Specify the action, approval threshold, statutory or charter source, and timing constraints.
- Regulatory filings and approvals—Antitrust review (merger control filing), foreign investment security review, industry-specific approvals.
- Counterparty consents—Consents triggered by change of control, anti-assignment, most-favored-nation treatment.
- Releases, terminations, or payoffs—Labor terminations tied to change of control, payoff letters, release of security interests.
- Escrow/holdback mechanisms—If the extraction identifies indemnity escrows, representation and warranty insurance deliverables, or holdback amounts tied to specific issues. Every finding tagged with a pre-closing action should reach the closing checklist, not just those marked "consent." If a finding is in a gray area (may require a closing action, may be a post-closing covenant), hand it off with a tag—the closing checklist can remove it if the equity purchase agreement provides otherwise. Under-handing off is a one-way door; over-handing off is corrected in review.

Successor liability. Flag: pending or threatened tort/product liability claims, environmental matters and cleanup obligations, bulk sale/fraudulent transfer risk (did the seller retain sufficient assets to pay remaining creditors?), the seller's post-closing liquidation plans (if the seller liquidates, plaintiffs pursue the buyer), and whether the equity purchase agreement has an assumed/excluded liability schedule that actually covers known risks. Even in an asset acquisition, the "de facto merger," "mere continuation," and "product line" doctrines can transfer liability—this is the analysis that surprises buyer clients who think they bought the assets clean.

Batching

For large categories (300 contracts), process in batches. After each batch, update the running issues list and immediately flag any 🔴 red—do not wait for the entire category to be completed before presenting deal-affecting issues.

Close with the next-action decision tree

Follow the next-action decision tree in CLAUDE.md ## 输出规范 . Tailor the options to what this skill just produced—the five default branches (draft X, escalate, gather facts, monitor and wait, other) are a starting point, not locked in. The decision tree itself is the output; the lawyer chooses.

If more than about 10 issues are extracted, or the user asks at any point: provide a dashboard (see CLAUDE.md ## 输出规范 → 数据密集产出的仪表盘选项). Tailor it to this output: counts by severity (🔴 / 🟠 / 🟡 / 🟢), counts by internal category, and a sortable issue grid with materiality, category, and data room source.

What this skill does not do

Does not make materiality judgments on borderline cases. It applies thresholds; humans decide the edge cases.
Does not negotiate representations and warranties. It produces findings that inform.
Does not replace bulk AI review. For high-volume clause extraction, hand off to an AI tool per ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md . This skill is for the judgment layer.