Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Apply China contract drafting review with San Guan Si Bu Fa. Trigger on contract review, drafting, clauses, or deal structure.
.claude/skills/thomasmoreai-contract-drafting-and-review-guidance-china-law/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 34 |
| gemini-3.1-pro-preview | 100% | 6 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 90% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 150% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 184% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 85% | 0% |
| case-18 | ✓→✓ | = Same ✓ | 205% | 0% |
This skill applies the Chinese contract drafting and review method known as "三观四步法" from 合同起草审查指南 (T/CAB 0121-2021). It helps lawyers, in-house counsel, business operators, and legal students structure contract review, draft contract frameworks, identify risks, and produce review comments under a China-law lens.
This skill is a workflow and checklist tool. It does not replace current legal research or advice from a licensed lawyer.
Every substantive claim based on the guide's methodology must include a citation to the original text.
Quote files (load the relevant one):
contract-drafting-quotes.md: definitions of macro, meso, micro analysis; drafting principles; transaction structure; parties; three-points-one-line method.contract-review-quotes.md: the complete 三观四步法 statement, review principles, transaction-structure framing, and comprehensive review.Citation format - always use this exact structure:
> "Source text here." > > - 合同起草审查指南, topic anchor: https://github.com/simbajigege/book2skills/blob/main/skills/contract-drafting-and-review-guidance-china-law/quotes/FILENAME.md#ANCHOR
Anchor mapping:
contract-drafting-quotes.md: #transaction-structure-importance, #meso-contract-structure, #micro-clause-expression, #four-steps-process, #promote-transaction-principle, #risk-control-balance, #legal-business-combination, #contract-types, #parties-identification, #three-points-one-line-method, #review-checklist, #standard-final-goalcontract-review-quotes.md: #three-perspectives-four-steps, #promote-transaction, #macro-transaction-structure, #three-points-one-line, #standard-purpose, #comprehensive-reviewRules:
Contract drafting and review should promote viable transactions while controlling legal and business risk. The agent must analyze the transaction at three levels: whether the deal structure works, whether the contract structure fits the deal, and whether the individual clauses are complete, clear, and enforceable.
| Workflow | User question pattern | Inputs | Steps | Output | Independent trigger? | Distinct references? | Triage score | Should be subskill? | Reason | |---|---|---|---|---|---|---|---:|---|---| | Contract review | Please review this contract | Contract text, user role, concerns, governing law | Needs -> macro -> meso -> micro -> review comments -> risk summary | Structured review memo | Yes | Same guide | 3 | No | Primary workflow; uses all dimensions in sequence. | | Contract drafting | Draft a contract framework for this deal | Deal facts, parties, subject, price, timeline, risk allocation | Needs -> transaction structure -> form selection -> clause framework -> open questions | Drafting outline and clause checklist | Yes | Same guide | 3 | No | Shares the same analysis order as review, only output differs. | | Deal-structure diagnosis | Can this transaction be done this way? | Transaction goal, parties, assets/services, approvals, constraints | Macro analysis -> legal/business obstacles -> alternatives | Structure recommendation | Yes | Same guide | 3 | No | A macro slice that remains tightly coupled to later drafting. | | Clause rewrite | Is this clause clear or risky? | Clause text, role, commercial intent | Identify right/obligation/liability points -> rewrite -> residual risk | Revised clause and rationale | Yes | Same guide | 3 | No | Micro slice but still uses the same three-points-one-line method. | | Final review before signing | What should I check before signing? | Final draft, signing process, approvals | Macro/meso/micro spot check -> signing formalities -> submission note | Pre-signing checklist | Yes | Same guide | 2 | No | Review-gate step inside the main workflow. |
Use a single-file skill because the guide is one integrated standard rather than a multi-domain book. All user tasks route through the same 三观四步法 sequence and differ mainly in output format. Separate subskills would duplicate the same macro, meso, micro, and review rules.
The Rule: A contract cannot be reviewed or drafted properly until the transaction goal, user role, risk tolerance, and desired deliverable are clear.
Start every contract review or drafting task by listing the minimum assumptions and missing facts. Do not invent deal facts. If the user provides enough context, proceed directly and mark unresolved assumptions.
The Rule: First decide whether the transaction can be done and what structure should express the parties' main rights and obligations.
For review, produce a macro conclusion: can proceed, can proceed after adjustment, high-risk, or do not proceed without counsel review. For drafting, propose a structure before drafting clauses. Cite contract-drafting-quotes.md#transaction-structure-importance or contract-review-quotes.md#macro-transaction-structure when explaining macro analysis.
The Rule: Select the contract form and clause order that fit the transaction stage, operational workflow, and legal effect the parties need.
When reviewing a contract's form, explain whether the document type matches the transaction and list structural fixes before line-level clause edits. Cite contract-drafting-quotes.md#meso-contract-structure when explaining meso analysis.
The Rule: Clause review must connect prerequisites, rights and obligations, and liability consequences so the contract can be performed and enforced.
For clause review, map the clause into prerequisite, right/obligation, and liability consequence. If any point is missing, propose revised language or a drafting instruction. Cite contract-drafting-quotes.md#three-points-one-line-method or contract-review-quotes.md#three-points-one-line.
The Rule: A contract review is incomplete until macro, meso, and micro findings are checked together and communicated in a usable form.
End every review with a prioritized risk summary and next-action list. If the user needs a negotiation posture, classify each issue as must change, should change, fallback, or business call. Cite contract-drafting-quotes.md#review-checklist or contract-review-quotes.md#comprehensive-review.
markdown## 合同审查意见 **合同类型:** [type] **我方角色:** [role] **适用法律假设:** [China law / unknown / user-specified] **资料完整性:** [complete / missing items] ### 一、宏观 - 交易结构 | 项目 | 结论 | 风险 | 建议 | |---|---|---|---| ### 二、中观 - 合同结构 | 项目 | 结论 | 风险 | 建议 | |---|---|---|---| ### 三、微观 - 条款审查 | 条款 | 问题 | 修改建议 | 优先级 | |---|---|---|---| ### 四、主要风险排序 1. [High-risk issue] 2. [Medium-risk issue] ### 五、下一步 [Redline / negotiate / ask for documents / counsel review / signing checklist]
markdown## 合同起草提纲 **交易目标:** [goal] **当事方:** [parties] **标的:** [subject] ### 一、交易结构设计 [Macro structure and required approvals.] ### 二、合同形式 [Document type and structure.] ### 三、条款框架(三点一线) | 模块 | 应写内容 | 待确认事项 | |---|---|---| ### 四、风险与谈判点 [Must-have terms and fallback positions.]
markdown## 条款审查与改写 ### 条款功能 [What this clause is supposed to do.] ### 三点一线检查 | 点 | 是否完整 | 问题 | |---|---|---| ### 修改建议 [Revised clause or drafting instruction.] ### 剩余风险 [What still needs business/legal confirmation.]
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-18 | pass→pass | 13,568 | 15,893 | +17% | 1 | 1 | 0% | 2,282 | 6,960 | +205% | 0 | 0 | — |
case-09 | pass→pass | 13,663 | 14,465 | +6% | 1 | 1 | 0% | 2,199 | 6,188 | +181% | 0 | 0 | — |
case-10 | pass→pass | 7,937 | 21,752 | +174% | 1 | 1 | 0% | 1,199 | 7,898 | +559% | 0 | 0 | — |
case-11 | pass→pass | 12,352 | 14,055 | +14% | 1 | 1 | 0% | 2,071 | 6,576 | +218% | 0 | 0 | — |
case-01 | fail→pass | 29,775 | 31,937 | +7% | 1 | 1 | 0% | 4,915 | 9,346 | +90% | 0 | 0 | — |
case-02 | fail→pass | 20,848 | 28,102 | +35% | 1 | 1 | 0% | 3,431 | 8,584 | +150% | 0 | 0 | — |
case-03 | fail→pass | 15,276 | 19,411 | +27% | 1 | 1 | 0% | 2,627 | 7,467 | +184% | 0 | 0 | — |
case-17 | pass→pass | 14,688 | 16,450 | +12% | 1 | 1 | 0% | 2,106 | 6,944 | +230% | 0 | 0 | — |
case-04 | fail→fail | 18,585 | 24,123 | +30% | 1 | 1 | 0% | 3,738 | 8,187 | +119% | 0 | 0 | — |
case-05 | fail→pass | 21,372 | 14,705 | -31% | 1 | 1 | 0% | 3,445 | 6,377 | +85% | 0 | 0 | — |
case-06 | fail→fail | 20,148 | 23,544 | +17% | 1 | 1 | 0% | 3,197 | 7,646 | +139% | 0 | 0 | — |
case-07 | pass→pass | 12,196 | 12,090 | -1% | 1 | 1 | 0% | 1,948 | 6,073 | +212% | 0 | 0 | — |
case-08 | pass→pass | 19,139 | 19,499 | +2% | 1 | 1 | 0% | 2,766 | 7,342 | +165% | 0 | 0 | — |
case-12 | pass→pass | 12,453 | 11,801 | -5% | 1 | 1 | 0% | 2,095 | 6,172 | +195% | 0 | 0 | — |
case-13 | pass→pass | 12,315 | 10,799 | -12% | 1 | 1 | 0% | 2,067 | 5,796 | +180% | 0 | 0 | — |
case-14 | pass→pass | 12,312 | 13,204 | +7% | 1 | 1 | 0% | 2,057 | 6,160 | +199% | 0 | 0 | — |
case-15 | pass→pass | 20,680 | 18,739 | -9% | 1 | 1 | 0% | 2,544 | 7,099 | +179% | 0 | 0 | — |
case-16 | pass→pass | 13,385 | 14,779 | +10% | 1 | 1 | 0% | 2,252 | 6,640 | +195% | 0 | 0 | — |
case-19 | pass→pass | 15,794 | 18,417 | +17% | 1 | 1 | 0% | 2,583 | 7,343 | +184% | 0 | 0 | — |
case-20 | pass→pass | 12,810 | 15,225 | +19% | 1 | 1 | 0% | 2,144 | 6,570 | +206% | 0 | 0 | — |
case-21 | pass→pass | 7,644 | 8,383 | +10% | 1 | 1 | 0% | 1,279 | 5,425 | +324% | 0 | 0 | — |
case-22 | pass→pass | 19,992 | 22,473 | +12% | 1 | 1 | 0% | 2,922 | 7,979 | +173% | 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 +18 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.