Judgment Document Generation logo

Judgment Document Generation

Community
THUYRan
judgment-document-generation

Based on a case-fact narrative, coordinate and invoke 8 atomic capabilities to generate a complete criminal judgment that is format-compliant and rigorously reasoned. Covers the full pipeline of fact-finding, legal application, sentencing reasoning, and document output.

Overview

PublisherTHUYRan
RepositoryLegal-Skills-Chinese
Skill namejudgment-document-generation
Stars
876
Forks
97
Bundled files
Instructions only
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by THUYRan on GitHub. Read the source before you install it.

Installation

Install the Judgment Document Generation AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/THUYRan/Legal-Skills-Chinese/tree/main/English/skills/judgment-document-generation
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/THUYRan/Legal-Skills-Chinese.git /tmp/Legal-Skills-Chinese
mkdir -p .claude/skills
cp -r /tmp/Legal-Skills-Chinese/English/skills/judgment-document-generation .claude/skills/judgment-document-generation
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Judgment Document Generation in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Judgment Document Generation on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Judgment Document Generation is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Chinese source (authoritative): ../../skills/judgment-document-generation/SKILL.md

Judgment Document Generation (8 Skills)

I. Capability Overview

This composite capability targets first-instance criminal cases. It takes a case-fact narrative (including prosecution charges, evidence summary, defendant statements, etc.) and outputs a structurally complete, rigorously reasoned, and format-compliant criminal judgment (刑事判决书). Execution strictly follows the mainline: "fact extraction → concept comprehension → dispute identification → statute retrieval → case retrieval → deductive reasoning → format application → terminology standardization," with quality checks and conditional branches at key nodes.


II. Selected Atomic Capabilities

StepCapability IDCapability NameRationale
S14Legal Core Element ExtractionThe foundation of a judgment is accurate fact-finding; all core elements—subject, act, object, time, causation, legal consequences, etc.—must be extracted from the case narrative to provide structured input for every subsequent step
S25Legal Concept ComprehensionPrecise understanding of the statutory constitutive elements of the charged offense, forms of crime (consummated / attempted / discontinued), joint crime, sentencing circumstances, and related concepts (intension and extension) is required to ensure correct legal application
S36Dispute Focus and Legal Relationship IdentificationThe "本院认为" (this Court's opinion) section must address the prosecution–defense dispute foci; the case's core disputed issues and the nature of the legal relationships must be accurately identified
S48Legal Provision RetrievalA judgment must cite specific legal provisions as the basis of adjudication; applicable Criminal Law articles, judicial interpretations, and sentencing norms must be retrieved
S59Case RetrievalSentencing should reference judicial practice patterns in similar cases to keep the outcome within a reasonable range and strengthen uniformity and persuasiveness
S616Deductive ReasoningThe core argumentative logic of a judgment is "major premise (legal norm) → minor premise (case facts) → conclusion (adjudicative result)"; deductive reasoning is the key capability for generating reasons and sentencing conclusions
S749Legal Document Format ApplicationCriminal judgments have strict format norms (caption/header, facts, reasons, operative part / disposition, closing); the document type must be identified and the corresponding format applied
S850Legal Terminology StandardizationA judgment is a formal legal instrument; all wording must be accurate, unambiguous, and consistent with legal drafting style throughout the output process

III. Atomic Capabilities Not Selected

This composite capability uses all 8 available atomic capabilities; there are no exclusions.

Rationale: Judgment-document generation is among the most complex end-to-end tasks in legal AI. It requires full-chain coverage from fact-finding through legal application to document output. The 8 atomic capabilities map to indispensable links in that chain—fact layer (AS4), concept layer (AS5), dispute layer (AS6), norm layer (AS8, AS9), reasoning layer (AS16), and output layer (AS49, AS50). Omitting any one would leave a defect in some dimension of the judgment.


IV. Complete Execution Flow

Flow Overview

Input: Case-fact narrative
  │
  ▼
[S1] AS4 Legal Core Element Extraction
  │
  ▼
[S2] AS5 Legal Concept Comprehension
  │
  ├──Quality Checkpoint QC1──→ If fail, backtrack to S1
  ▼
[S3] AS6 Dispute Focus and Legal Relationship Identification
  │
  ▼
[S4] AS8 Legal Provision Retrieval
  │
  ├──Quality Checkpoint QC2──→ If fail, backtrack to S2/S3
  ▼
[S5] AS9 Case Retrieval
  │
  ▼
[S6] AS16 Deductive Reasoning
  │
  ├──Quality Checkpoint QC3──→ If fail, backtrack to S4/S5
  ▼
[S7] AS49 Legal Document Format Application
  │
  ▼
[S8] AS50 Legal Terminology Standardization
  │
  ├──Quality Checkpoint QC4──→ If fail, backtrack to S7
  ▼
Output: Complete criminal judgment

S1: Legal Core Element Extraction

ItemContent
Capability invokedAS4 — Legal Core Element Extraction
InputFull original case-fact narrative (including prosecution charges, evidence list, defendant's attitude)
ActionsExtract the following irreducible core elements item by item: ① criminal subject (defendant identity, co-offender information); ② criminal act (specific manner and means); ③ object / target of the offense (legal interest infringed, specific property); ④ purpose / motive; ⑤ criminal result (consummated / attempted, loss amount); ⑥ timeline (offense date, apprehension date, restitution date, etc.); ⑦ causation; ⑧ sentencing-related circumstances (voluntary surrender / 自首, truthful confession / 坦白, restitution of stolen property / 退赃, prior convictions, etc.); ⑨ evidence summary
OutputStructured element inventory (JSON or table)
Passed to next stepComplete structured element inventory for S2 to comprehend related legal concepts

Case 1 example:

Extraction result: subject (Zhang 3), act (forging ID card, impersonating to defraud property), purpose (repaying debts), target amount (RMB 100,000), result (discovered; unsuccessful), time (28 April 2017)

Case 2 example:

Extraction result: subjects (An Xu + Yang jointly), act (forced entry with tools and household theft / 入户盗窃), object (gold and silver jewelry), time (offense 9 May 2016; arrested 15 April 2018), mitigating circumstance (returned RMB 3,000)


S2: Legal Concept Comprehension

ItemContent
Capability invokedAS5 — Legal Concept Comprehension
InputStructured element inventory from S1
ActionsBased on extracted act features and results, identify possible offenses and, for each, comprehend constitutive elements and intension/extension of: ① statutory constitutive elements of the charged offense (four-element theory or three-tier theory); ② forms of crime (consummated, attempted, discontinued / 中止, preparatory); ③ joint crime concepts (principal / 主犯, accessory / 从犯, coerced accomplice / 胁从犯); ④ sentencing circumstance concepts (statutory lighter / mitigated / heavier circumstances; discretionary circumstances); ⑤ other special concepts involved (e.g., "household theft / 入户盗窃", "huge amount / 数额巨大")
OutputLegal concept analysis table for the case (each concept with definition, constitutive elements, and applicability conditions)
Passed to next stepConcept analysis table + S1 element inventory for S3 to identify dispute foci

Case 1 example:

Comprehend fraud (诈骗罪) constitutive elements—purpose of illegal appropriation, fabricating facts or concealing the truth, obtaining another's property by deception, relatively large amount; criminal attempt—having begun to commit the crime, failing to complete it for reasons independent of the actor's will

Case 2 example:

Comprehend theft "household theft / 入户盗窃"—criminalized regardless of amount; joint crime; criteria for distinguishing principal and accessory


🔍 Quality Checkpoint QC1: Element–Concept Match Verification

Check ItemVerification LogicOn Failure
Element completenessWhether elements extracted in S1 cover all facts corresponding to the offense's constitutive elements in S2Backtrack to S1; supplement omitted elements
Concept coverageWhether S2 covers legal concepts corresponding to all act features in S1Supplement omitted concepts in S2
Preliminary offense characterization reasonablenessWhether the preliminarily identified offense matches act features (e.g., violent taking of property should not be characterized as theft)Backtrack to S2; re-analyze offense direction

S3: Dispute Focus and Legal Relationship Identification

ItemContent
Capability invokedAS6 — Dispute Focus and Legal Relationship Identification
InputS1 structured element inventory + S2 legal concept analysis table
Actions① Identify the nature of legal relationships among the parties (criminal prosecution relationship, joint-offender relationship, victim–defendant relationship, etc.); ② Identify possible prosecution–defense dispute foci, including but not limited to: offense characterization disputes; form-of-crime disputes (consummated vs. attempted); co-offender status disputes (principal vs. accessory); sentencing circumstance disputes (recognition of voluntary surrender / truthful confession); amount determination disputes; ③ For each dispute focus, mark priority for response in the "本院认为" section
OutputDispute focus list (priority-ranked) + legal relationship map
Passed to next stepDispute focus list to guide precise statute retrieval in S4

Case 1 example:

Dispute foci: ① whether fraud was consummated or attempted; ② characterization of "huge amount" (which sentencing tier under Article 266 applies to RMB 100,000)

Case 2 example:

Dispute foci: ① An Xu's status in the joint crime (principal or accessory); ② effect of returning RMB 3,000 on sentencing


S4: Legal Provision Retrieval

ItemContent
Capability invokedAS8 — Legal Provision Retrieval
InputS2 offense and concept analysis + S3 dispute focus list
ActionsRetrieve applicable legal norms by layer: ① Conviction provisions: Criminal Law Special Part articles for the offense (including sentencing ranges); ② Form-of-crime provisions: General Part articles such as attempt (Art. 23), discontinuation (Art. 24); ③ Joint-crime provisions: e.g., joint crime (Art. 25), principal/accessory (Arts. 26–29); ④ Sentencing circumstance provisions: e.g., voluntary surrender (Art. 67), truthful confession (Art. 67 para. 3), meritorious service (Art. 68); ⑤ Penalty enforcement provisions: fine (Arts. 52, 53), restitution (Art. 64), etc.; ⑥ Judicial interpretations: amount thresholds for the offense, sentencing guidelines, etc.
OutputApplicable provision list (full text, applicability rationale, corresponding dispute focus IDs)
Passed to next stepProvision list to constrain S5 case retrieval scope and to serve as major premises for S6 deductive reasoning

Case 1 example:

Retrieval: Criminal Law Art. 266 (fraud and sentencing ranges), Art. 23 (criminal attempt), Art. 67 para. 3 (truthful confession → lighter punishment)

Case 2 example:

Retrieval: Criminal Law Arts. 264, 25, 67 para. 3, 52, 53, 64


🔍 Quality Checkpoint QC2: Provision–Dispute Coverage Verification

Check ItemVerification LogicOn Failure
Full dispute-focus coverageWhether each dispute focus in S3 has supporting provisionsBacktrack to S4; supplement retrieval
Offense provision accuracyWhether retrieved offense provisions match the offense determined in S2Backtrack to S2 to reconfirm offense, or backtrack to S4 to correct retrieval
Sentencing circumstance provision completenessWhether each sentencing circumstance extracted in S1 has a corresponding provisionBacktrack to S4; supplement sentencing-circumstance provisions
Judicial interpretation currencyWhether retrieved judicial interpretations were in force at the time of the offenseReplace with the version effective at the time of the offense

S5: Case Retrieval

ItemContent
Capability invokedAS9 — Case Retrieval
InputS1 structured elements (offense, manner of act, amount, circumstances) + S3 dispute foci + S4 provision list
Actions① Using offense + manner of act + form of crime as keywords, retrieve judgments in similar cases; ② Focus on: sentencing ranges under similar amounts/circumstances; adjudicative tendencies on dispute foci (e.g., practical grasp of accessory criteria); sentencing habits of courts in the same locality; ③ Extract comparable-case sentencing baselines and adjustment factors
OutputComparable-case retrieval report (3–5 reference cases with offense, circumstances, sentence, and adjudicative gist / 裁判要旨)
Passed to next stepComparable-case sentencing range and adjudicative tendencies to calibrate the sentencing conclusion in S6

Conditional branches:

  • If highly similar cases are found (manner of act, amount, and circumstances substantially consistent): extract sentencing reference values directly
  • If no highly similar cases are found: broaden the search (relax amount range or manner of act) and mark limited reference value

Case 1 example:

Find sentencing patterns in similar attempted-fraud cases to assess reasonableness of a seven-month sentencing range

Case 2 example:

Induce Shanghai-area judicial practice sentencing patterns from similar household-theft cases; find precedents on accessory determination criteria


S6: Deductive Reasoning

ItemContent
Capability invokedAS16 — Deductive Reasoning
InputS1 structured elements (minor premise) + S4 provision list (major premise) + S3 dispute foci (argumentation targets) + S5 comparable-case references (sentencing calibration)
ActionsReason layer by layer along the following chain; each layer uses the syllogistic structure "major premise → minor premise → conclusion": Layer 1: Conviction reasoning — Map case facts item by item against constitutive elements → conclude whether the offense is established; Layer 2: Form-of-crime reasoning (conditionally triggered) — If attempt/discontinuation etc. are disputed, map facts against form requirements → conclude on form; Layer 3: Joint-crime reasoning (conditionally triggered) — If joint crime exists, argue each defendant's status and role → conclude principal/accessory; Layer 4: Sentencing circumstance reasoning — Argue item by item whether statutory/discretionary lighter, mitigated, or heavier circumstances are established; Layer 5: Sentencing conclusion reasoning — Combine conviction + form + co-offender status + sentencing circumstances + comparable-case references → derive specific term of imprisonment and additional penalties; Layer 6: Ancillary matters reasoning — Derive restitution, confiscation, restraining orders, and other ancillary dispositions
OutputComplete reasoning chain (major premise, minor premise, and conclusion for each layer) + final adjudicative conclusions (principal penalty + additional penalties + ancillary dispositions)
Passed to next stepReasoning-chain text (for the "本院认为" section) + adjudicative conclusions (for the operative part / 判决主文)

Case 1 example:

① Fraud constitutive elements satisfied item by item → fraud established; ② discovered and unsuccessful for reasons independent of will → attempt established; ③ RMB 100,000 falls in the "huge amount" sentencing tier; attempt may be punished more lightly or mitigated by reference to consummated offense; ④ plus truthful confession for lighter punishment → derive fixed-term imprisonment of seven months + fine of RMB 30,000

Case 2 example:

① Household-theft constitutive elements satisfied → theft established; ② two persons plotted and jointly committed → joint crime established; ③ defense claim of accessory status lacks sufficient evidence → not accepted; ④ truthful confession + return of funds for lighter punishment → derive fixed-term imprisonment of six months + fine of RMB 2,000 + restitution order


🔍 Quality Checkpoint QC3: Reasoning-Chain Completeness and Consistency Verification

Check ItemVerification LogicOn Failure
Full constitutive-element coverageWhether conviction reasoning argued every constitutive element of the offenseBacktrack to S6 Layer 1; supplement omitted-element argument
Full dispute-focus responseWhether each dispute focus in S3 received a clear response in the reasoning chainBacktrack to S6; supplement argument on omitted dispute foci
Sentencing reasonablenessWhether the derived sentence falls within the statutory range in S4 and is broadly consistent with the S5 comparable-case rangeIf outside statutory range → backtrack to S4 to confirm provision application; if large deviation from comparable cases → backtrack to S5 to broaden retrieval or supplement special-circumstance reasoning in S6
Logical coherenceWhether conclusions across layers contradict (e.g., finding attempt but sentencing as consummated)Backtrack to S6; correct contradictory links

S7: Legal Document Format Application

ItemContent
Capability invokedAS49 — Legal Document Format Application
InputS1 structured elements + S6 reasoning chain and adjudicative conclusions + S4 provision list
Actions① Identify document type as "first-instance criminal judgment"; ② Per Supreme People's Court criminal judgment format norms, assemble prior outputs into: Header / 首部 — court name, case number, prosecuting organ, defendant basic information, defense counsel information, case origin and trial process; Facts / 事实部分 — "经审理查明" (facts found after trial) paragraph (restating facts based on S1 elements) + evidence enumeration; Reasons / 理由部分 — "本院认为" paragraph (expanding argument from S6 reasoning chain, responding to each dispute focus); Operative part / 判决主文 — conviction and sentence + additional penalties + restitution and other ancillary dispositions (from S6 conclusions); Closing / 尾部 — sentence credit calculation, appeal rights notice, panel signatures, date; Appendix — full text of cited legal provisions (from S4 provision list)
OutputFormatted judgment draft (all structural sections)
Passed to next stepFull judgment draft for S8 terminology standardization and polishing

Case 1 example:

Criminal judgment format, stating in order: party information, prosecution charges, fact-finding, this Court's opinion, operative part, sentence calculation, appeal notice, appended legal provisions


S8: Legal Terminology Standardization

ItemContent
Capability invokedAS50 — Legal Terminology Standardization
InputFull judgment draft from S7
ActionsSentence-by-sentence review and standardization: ① Terminological accuracy: Ensure offense names, forms of crime, and sentencing circumstances use statutory terms (e.g., "犯罪未遂" / criminal attempt rather than "未遂犯"; "如实供述自己的罪行" / truthfully confessing one's crimes rather than colloquial "坦白交代"); ② Operative-part norms: Ensure formulations such as "判处有期徒刑X年/月" (sentenced to fixed-term imprisonment of X years/months), "并处罚金人民币X元" (and a fine of RMB X), "责令退赔" (ordered to make restitution) conform to judicial drafting practice; ③ Citation norms: Ensure correct format "依照《中华人民共和国刑法》第X条第X款" (pursuant to Article X, paragraph X of the Criminal Law of the PRC); ④ Stylistic consistency: Third-person narration, legal-document sentence patterns, no colloquial expression; ⑤ Logical connective norms: Proper use of "鉴于" (whereas/in light of), "综上" (in sum), "据此" (accordingly), etc.
OutputFinal judgment after terminology standardization
Passed to final outputComplete criminal judgment

Case 1 example:

Ensure formulations such as "犯罪未遂", "比照既遂犯从轻或者减轻处罚" (punished more lightly or mitigated by reference to a consummated offender), "判处有期徒刑", "并处罚金" conform to criminal-document norms


🔍 Quality Checkpoint QC4: Final-Draft Completeness Verification

Check ItemVerification LogicOn Failure
Structural completenessWhether the judgment includes header, facts, reasons, operative part, and closing in fullBacktrack to S7; supplement missing sections
Operative part–reasons consistencyWhether offense, term, and fine in the operative part fully match conclusions in "本院认为"Backtrack to S7; correct inconsistencies
Provision citation completenessWhether all cited provisions are appended at the endBacktrack to S7; supplement omitted provisions
Terminology zero-defectWhether colloquial expressions, terminological errors, or format irregularities remainBacktrack to S8; re-polish
Sentence calculation accuracyWhether start/end dates of the sentence (including pretrial detention credit / 羁押折抵) are correctBacktrack to S7; correct date calculation

V. Conditional Branches and Backtracking

5.1 Form-of-Crime Branch

In S1 extraction results, check the "criminal result" element:
├── Result is "unsuccessful / failed / attempted" → S2 must comprehend "criminal attempt"
│                                                  → S3 must list "consummated vs. attempted" as a dispute focus
│                                                  → S4 must retrieve Criminal Law Art. 23
│                                                  → S6 must add a "form-of-crime reasoning" layer
├── Result is "discontinued / 中止" → similar path, substitute Art. 24
└── Result is "consummated / successful" → skip form-of-crime–related steps

Case 1 triggers this branch: Zhang 3 "discovered and unsuccessful" → attempt path → S2 comprehends criminal attempt → S4 retrieves Art. 23 → S6 argues attempt established and applies lighter/mitigated punishment

Case 2 does not trigger this branch: An Xu "stole gold/silver jewelry and fled" → consummated → skip form-of-crime steps

5.2 Joint-Crime Branch

In S1 extraction results, check the "criminal subject" element:
├── Multiple subjects → S2 must comprehend "joint crime" and "principal/accessory"
│                     → S3 must list "each defendant's status" as a dispute focus
│                     → S4 must retrieve Criminal Law Arts. 25–29
│                     → S6 must add a "joint-crime reasoning" layer
└── Single subject → skip joint-crime–related steps

Case 2 triggers this branch: An Xu "together with Yang" → joint-crime path → S2 comprehends joint crime → S3 identifies principal/accessory dispute → S6 argues that accessory defense is not accepted

Case 1 does not trigger this branch: Zhang 3 acted alone → skip joint-crime steps

5.3 Defense-Opinion Response Branch

In S1 extraction results, check whether defense counsel opinions exist:
├── Defense opinions present → S3 must list issues raised as dispute foci
│                            → S6 must respond to each opinion (accept / reject with reasons)
│                            → S7 "本院认为" must include commentary on defense opinions
└── No defense opinions → skip defense-opinion response; respond only to prosecution charges

5.4 Sentencing Anomaly Backtracking

In QC3, check sentencing reasonableness:
├── Sentence outside statutory range → backtrack to S4 to confirm sentencing tier
│                                    → backtrack to S6 to recalculate sentence
├── Deviation from comparable cases >50% → backtrack to S5 to broaden case retrieval
│                                        → backtrack to S6 to supplement special-circumstance reasoning
└── Sentence within reasonable range → continue to S7

5.5 Offense-Change Backtracking

In S6 conviction reasoning, constitutive elements not satisfied:
├── A given element clearly not satisfied → backtrack to S2 to re-analyze possible offenses
│                                        → backtrack to S3 to re-identify dispute foci
│                                        → backtrack to S4 to re-retrieve provisions
│                                        → restart reasoning from S6
└── All elements satisfied → continue subsequent reasoning layers

VI. Complete Examples

Example 1: Attempted Fraud Case (Case 1)

Input: Case-fact narrative of the prosecution charging Zhang 3 with forging an ID card, impersonating another to defraud RMB 100,000, being discovered, and failing to obtain the funds.

Execution process:

StepCapabilityKey OperationOutput Summary
S1AS4Extract subject (Zhang 3), act (forge ID, impersonate to defraud), purpose (repay debts), target amount (RMB 100,000), result (discovered; unsuccessful), time (28 April 2017)Structured element inventory
S2AS5Comprehend four constitutive elements of fraud; comprehend definition and elements of criminal attemptConcept analysis table
QC1—Verify: elements cover all fraud constitutive elements ✓; "unsuccessful" maps to criminal attempt ✓Pass
S3AS6Identify dispute foci: ① consummated vs. attempted; ② which sentencing tier for RMB 100,000Dispute focus list
S4AS8Retrieve Criminal Law Arts. 266, 23, 67 para. 3Provision list
QC2—Verify: both dispute foci have provision coverage ✓; sentencing-circumstance provisions complete ✓Pass
S5AS9Retrieve sentencing patterns in similar attempted-fraud cases; confirm reasonableness of seven-month rangeComparable-case report
S6AS16Four-layer reasoning: ① elements satisfied → fraud established; ② discovered for reasons independent of will → attempt; ③ "huge amount" tier + attempt lighter/mitigated; ④ truthful confession lighter → seven months + fine RMB 30,000Reasoning chain + adjudicative conclusions
QC3—Verify: sentence within statutory range ✓; consistent with comparable cases ✓; all dispute foci addressed ✓Pass
S7AS49Assemble per criminal judgment format: header → fact-finding → this Court's opinion → operative part → sentence calculation → appeal notice → appended provisionsJudgment draft
S8AS50Standardize formulations such as "犯罪未遂", "比照既遂犯从轻或者减轻处罚", "判处有期徒刑", "并处罚金"Final judgment
QC4—Verify: structure complete ✓; operative part consistent with reasons ✓; terminology standard ✓; sentence calculation correct (2017.4.28–2017.11.27) ✓Pass

Output: Complete criminal judgment sentencing Zhang 3 for fraud to fixed-term imprisonment of seven months and a fine of RMB 30,000.

Conditional branches triggered: Form-of-crime branch (attempt path) ✓; joint-crime branch not triggered (acted alone).


Example 2: Household Theft Joint-Crime Case (Case 2)

Input: Case-fact narrative of the prosecution charging An Xu, together with Yang, with forced entry and household theft of gold and silver jewelry, and return of RMB 3,000.

Execution process:

StepCapabilityKey OperationOutput Summary
S1AS4Extract subjects (An Xu + Yang), act (forced entry with tools; household theft), object (gold/silver jewelry), timeline (offense 2016.5.9 → arrest 2018.4.15), mitigating circumstance (returned RMB 3,000)Structured element inventory
S2AS5Comprehend special criminalization standard for "household theft"; comprehend joint crime and principal/accessory criteriaConcept analysis table
QC1—Verify: elements cover household-theft constitutive elements ✓; joint-crime concepts comprehended ✓Pass
S3AS6Identify dispute foci: ① An Xu's status in joint crime (principal vs. accessory); ② sentencing effect of returning RMB 3,000Dispute focus list
S4AS8Retrieve Criminal Law Arts. 264, 25, 67 para. 3, 52, 53, 64Provision list
QC2—Verify: co-offender status dispute covered by Arts. 25–27 ✓; return-of-funds circumstance covered by Art. 64 ✓Pass
S5AS9Retrieve Shanghai-area household-theft sentencing patterns; retrieve precedents on accessory determinationComparable-case report
S6AS16Four-layer reasoning: ① household-theft elements satisfied → theft established; ② plotted and jointly committed → joint crime established; ③ accessory claim lacks sufficient evidence → not accepted; ④ truthful confession + return lighter → six months + fine RMB 2,000 + restitutionReasoning chain + adjudicative conclusions
QC3—Verify: sentencing reasonable ✓; accessory defense addressed ✓; logically coherent ✓Pass
S7AS49Assemble per criminal judgment format; operative part includes two items (conviction/sentence + restitution order)Judgment draft
S8AS50Standardize formulations such as "入户盗窃", "共同犯罪", "责令退赔违法所得后发还被害人"Final judgment
QC4—Verify: structure complete ✓; both operative items consistent with reasons ✓; defense-opinion response included ✓Pass

Output: Complete criminal judgment sentencing An Xu for theft to fixed-term imprisonment of six months and a fine of RMB 2,000; ordered to make restitution of illegal gains to be returned to victim Gong; defense accessory opinion not accepted.

Conditional branches triggered: Joint-crime branch ✓; defense-opinion response branch ✓; form-of-crime branch not triggered (consummated).


VII. Exception Handling and Fallback Mechanisms

Exception ScenarioHandling
Case-fact narrative too brief; S1 cannot extract sufficient elementsMark "facts unclear" in output, list missing elements, request supplementation, then re-execute
S2 cannot determine a unique offense (concurrence of offenses / 罪名竞合)In S3, list offense selection as the primary dispute focus; in S6, argue constitutive elements for each candidate offense separately, choose one with reasoned justification
S5 finds no comparable casesMark "no direct comparable-case reference"; S6 sentencing reasoning relies only on statutory sentencing ranges and strengthens reasoned argumentation in the judgment
Logical contradiction appears in S6 reasoning chainTrigger full-chain backtracking; re-examine from S1 whether element extraction is accurate; locate contradiction source step by step
Case involves multiple offenses with concurrent sentencing / 数罪并罚Run the full S2–S6 flow for each offense; at the end of S6 add a "concurrent sentencing reasoning" layer; in S7 assemble using multi-offense judgment format

VIII. Capability Call Dependency Summary

AS4 ──→ AS5 ──→ AS6 ──→ AS8 ──→ AS9 ──→ AS16 ──→ AS49 ──→ AS50
 │        │        │        │        │        │         │         │
 │        │        │        │        │        │         │         └─ Final output
 │        │        │        │        │        │         └─ Format assembly
 │        │        │        │        │        └─ Core reasoning (depends on AS4+AS8+AS6+AS9)
 │        │        │        │        └─ Sentencing calibration
 │        │        │        └─ Major premises supplied
 │        │        └─ Argumentation targets set
 │        └─ Conceptual framework established
 └─ Factual foundation laid

Key dependency notes:

  • AS16 (deductive reasoning) is the core hub of the entire flow; it depends simultaneously on outputs of four upstream capabilities: AS4 (minor premise), AS8 (major premise), AS6 (argumentation targets), and AS9 (sentencing calibration)
  • AS49 (format application) and AS50 (terminology standardization) are sequential output-layer capabilities: AS49 handles structure, AS50 handles wording; together they ensure formal quality of the final document
  • AS5 (concept comprehension), though early in the flow, permeates all subsequent steps and is the foundational capability ensuring accuracy of legal application

Frequently asked questions

What does the Judgment Document Generation AI skill do?

Based on a case-fact narrative, coordinate and invoke 8 atomic capabilities to generate a complete criminal judgment that is format-compliant and rigorously reasoned. Covers the full pipeline of fact-finding, legal application, sentencing reasoning, and document output.

Why use Judgment Document Generation on TypingMind?

Because you install it once and use it with any model. Judgment Document Generation is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Judgment Document Generation in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/THUYRan/Legal-Skills-Chinese/tree/main/English/skills/judgment-document-generation. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Judgment Document Generation?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Judgment Document Generation?

As many as you like. As long as a model supports skills, you can use Judgment Document Generation with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Judgment Document Generation AI skill free?

It is published on GitHub by THUYRan. Check the repository for licensing terms. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇