Positioning
This skill is used to organize an already-developed code project into materials that a patent agent can continue to draft and evaluate. The core goal is not to translate code into patent language, but to organize the real implementation, technical problem, technical solution, technical effect, and evidence locations into a traceable invention patent draft.
The default jurisdiction is Chinese invention patents, mainly applicable to solutions that are primarily implemented in code, such as software, algorithms, Agent systems, scheduling optimization, state representation, authentication, memory, context orchestration, and file systems.
Recommended main workflow:
用户材料 + 代码仓库
↓
输入成熟度判断 + archive 归档目录
↓
项目边界与依赖画像
↓
方案-代码证据映射
↓
算法/软件类说明书式技术交底书
↓
权利要求布局卡 + 权利要求-证据矩阵
↓
发明专利初稿 + 初稿自检表
Startup Gate
Before starting to read code, first confirm:
- Whether there is already a candidate patent solution list.
- Whether the candidate list contains only names, or already has technical problems, core implementations, technical effects, and code scope.
- Whether the goal this time is a technical disclosure document, a claim layout card, an invention patent first draft, or candidate solution mining.
- Whether there are PRDs, requirement specifications, meeting minutes, customer disclosure templates, existing application styles, or agent format requirements.
- What the archiving topic is for this session, used to create
archive/YYYY-MM-DD-主题/。
A patent name is not a technical solution. If the user only provides a title or a one-sentence direction, do not directly write a disclosure document or first draft.
Input Maturity Triage
| Level | Input State | Default Action |
|---|---|---|
| T0 Name List | Only patent names, titles, or one-sentence directions | Output a "Patent Name Reverse Clarification Card" and wait for user confirmation |
| T1 Direction List | Has names and business directions, but lacks technical problems, implementation paths, or effects | First fill in candidate explanations, code evidence directions, and questions to be confirmed |
| T2 Technical Solution List | Already has technical problems, core implementations, target effects, and preliminary module scope | Proceed to targeted search and generate code evidence mapping and a disclosure document |
| T3 Evidence-Based Solution List | Already has technical solutions, code paths, key processes, and evidence levels | After reviewing the evidence, continue to generate a disclosure document, layout card, or first draft |
| No List | The user hopes to mine patent points from code | First output a list of candidate patentable solutions, and draft only after manual screening |
T0/T1 without manual confirmation or automatically mined results can only be used as candidate directions and must not be written as technical solutions already confirmed by the customer.
Material Priority
During execution, read user-provided materials first, then read code. Recommended order:
- Candidate patent solution list or title list
- PRD, product solution, requirement specification
- Communication minutes, recording transcripts, customer supplementary explanations
- Technical disclosure templates, existing application documents, agent format requirements
- Code repository README, architecture documentation, directory descriptions
- Source code, configuration, tests, deployment and runtime files
If the user provides a template or sample, first extract the section structure, step numbering, drawing conventions and technical effect wording, then generate the deliverable. If no template is provided, follow references/algorithm-software-disclosure-format.md and the built-in template.
Required specifications
Read the corresponding file only when needed, to avoid loading all references into the context at once.
| Problem to be handled | Read file |
|---|---|
| Algorithm/software disclosure document structure, S1...Sn, rules for placing code evidence afterward, formulas and notation conventions | references/algorithm-software-disclosure-format.md |
| Project technical solution profile, dependency boundaries, distinguishing self-developed and third-party capabilities | references/project-analysis-spec.md |
| S/F/E extraction, A/B/C grading, translation from code to patent expression | references/code-extraction-spec.md |
| Quickly enter claim layout card, first draft and self-check writing | references/patent-drafting-quick-reference.md |
| Need complete drafting basis, claim hierarchy and abstract rules | references/patent-drafting-spec.md |
This file retains only entries, triage, routing and strong rules; detailed rules are henceforth maintained preferentially in references/。
Template routing
| Target output | Template |
|---|---|
| Patent title reverse clarification card | templates/patent-title-clarification-card-template.md |
| Algorithm/software invention patent technical disclosure document | templates/invention-patent-disclosure-template.md |
| Claim layout card | templates/invention-patent-claim-layout-template.md |
| Claim-evidence matrix | templates/invention-patent-claim-evidence-matrix-template.md |
| Invention patent first draft | templates/invention-patent-draft-template.md |
| Invention patent first draft self-check table | templates/invention-patent-draft-self-check-template.md |
templates/ is a deliverable skeleton that can be filled in directly;references/ contains execution rules and judgment criteria. Do not merge the two.
Core strong rules
- First determine input maturity, then read code details.
- The solutions, templates and samples provided by the user take precedence over AI's own inferences.
- Candidate directions that have not been manually confirmed must not directly enter the disclosure document or patent first draft.
- First profile project boundaries and dependencies, to avoid writing default capabilities of third-party libraries, model platforms or frameworks as the applicant's innovation.
- Code evidence mapping and the main text of the technical disclosure document must be layered.
- The main text of the disclosure document defaults to an algorithm/software specification-style structure, not a code inventory report.
- File paths, function names, field names, test files and internal event names go by default into the evidence mapping or appendix; the main text retains only necessary evidence summaries.
- The summary of the invention is preferably organized as technical problem, overall solution, steps S1...Sn, further limitations and technical effects.
- The specific embodiments expand input, model/state, steps, parameters, output and alternative implementations using example scenarios.
- The code architecture must be understood, but the final main text should translate it into technical objects, relationships, states, actions, timing and outputs, without presenting technology choices or module lists.
05-技术交底书、05A-权利要求布局卡、05B-权利要求-证据矩阵、06-发明专利初稿The S1...Sn step numbering in should remain consistent.- When the target is a first draft, by default first generate the layout card and evidence matrix, then generate the complete first draft.
- Grade A evidence may enter the independent claim skeleton; Grade B evidence preferably enters dependent claims or preferred embodiments; Grade C evidence may only enter pending-confirmation items or alternative embodiments.
- The abstract is by default kept within 300 characters and does not use commercial promotional language.
- Formal application documents still require review by a patent agent.
- For solutions involving scoring, ranking, thresholds, vectors, or graph structures, the main text formulas must first establish a symbol table, use subscripts for dimensions, with symbols unique and in identical form across sections; for style conventions see
references/algorithm-software-disclosure-format.md"Formula and Symbol Style Conventions". - Product revisions under the same archive topic must not overwrite the original; save them separately with a timestamp suffix and leave a record in
08-修订记录.md; solution-level revisions must not go back to T0/T1 for re-mining.
Output Levels
| Level | Applicable Scenario | Main Deliverables |
|---|---|---|
| L0 Name Reverse Clarification | The user provides only a patent name, title, or a one-sentence direction | Patent name reverse clarification card + manual confirmation record |
| L1 Code Evidence Mapping | A candidate solution already exists and it is necessary to determine whether it maps to a code implementation | Solution-code evidence mapping table |
| L2 Technical Disclosure Document | Currently the most recommended main deliverable | Algorithm/software-type specification-style technical disclosure document + necessary evidence appendix |
| L2.5 Claim Layout | Need to advance to the application document level but stabilize the claim tree first | Claim layout card + claim-evidence matrix |
| L3 Invention Patent Draft | Need a complete draft before agent drafting | Specification draft + claims draft + abstract + self-check table |
| L4 Patentable Solution Mining | The user has not yet summarized candidate solutions | Candidate solution list + priority recommendations + manual screening record |
Standard Execution Order
Execute trimmed to the target level; it is not necessary to output all files every time.
- Pre-read user materials and confirm the target level and template requirements.
- Create
archive/YYYY-MM-DD-主题/。 - Determine input maturity: clarify first for T0/T1; mine first if there is no list.
- Read the project README, PRD, architecture, and dependency files to form a boundary profile.
- Read code around the confirmed solution, extract evidence, and grade it.
- Translate the engineering implementation into S1...Sn, technical features, technical effects, and alternative embodiments.
- Generate the disclosure document with a specification-style structure in the main text and evidence placed afterward.
- If the application document level is required, generate the layout card and evidence matrix.
- If a complete draft is required, generate the invention patent draft and self-check table.
- Output questions pending confirmation by R&D, the patent agent, or the applicant entity.
Recommended Deliverable Files
archive/YYYY-MM-DD-主题/
├── 00-输入材料摘要.md
├── 01-专利名称反向澄清卡-待确认.md
├── 01A-人工确认记录.md
├── 02-项目边界与依赖画像.md
├── 03-项目技术方案画像.md
├── 04-方案代码证据映射-<方案名>.md
├── 05-技术交底书-<方案名>.md
├── 05A-权利要求布局卡-<方案名>.md
├── 05B-权利要求-证据矩阵-<方案名>.md
├── 06-发明专利初稿-<方案名>.md
├── 06A-发明专利初稿自检表-<方案名>.md
├── 07-研发补充问题清单.md
└── 08-修订记录.md(迭代时生成,非每次必产)
If currently in candidate solution mining mode, first output 03-候选可专利方案清单-待人工筛选.md and 04-优先申请建议.md, and wait for user confirmation before proceeding to targeted retrieval.
Iteration and Version Traceability
For product revisions under the same archive/YYYY-MM-DD-主题/ directory, follow these rules:
- Do not overwrite the original: when revising an existing product, save it as a new file named "main product name + timestamp suffix", for example
05-技术交底书-<方案名>-250622-1430.md. Multiple versions in the same directory constitute the version history; do not create subdirectories. - Maintain
08-修订记录.md: append a record for each revision, including at least the revision intent, affected product files, whether S1...Sn changed, and the timestamp. Create it on the first revision and only append thereafter. - Access control — do not fall back to re-mining: plan-level incremental revisions only update evidence, steps, or expression on a confirmed plan; do not return to T0/T1 to redo candidate mining or name clarification. If the user wants to change plans or re-mine, a new
archive/YYYY-MM-DD-主题/directory should be opened, not mixed with the original plan's files. - Step chain synchronization: if a revision causes S1...Sn to change, the disclosure document, claim layout card, claim-evidence matrix, and invention patent draft must be updated in sync, and register the affected files in
08-修订记录.md.
Default structure of the disclosure document
The technical disclosure document for software, algorithm, and Agent system plans uses the following order by default:
- Basic plan information
- Technical field
- Background technology
- Summary of the invention
- Description of the drawings
- Detailed embodiments
- Technical effects
- Alternative embodiments
- Code evidence summary and appendix index
- Items pending confirmation by R&D / patent agent
If multiple code paths, function names, or field names appear consecutively in the main text, rewrite by moving them into 04-方案代码证据映射-<方案名>.md or the evidence appendix.
Input/output
Input
- Required: code files or project directories, technical description, innovation point description, candidate patent plans — at least two of these categories; if only a patent name is available, handle it as L0.
- Optional: PRD, requirements specification, meeting minutes, technical field, application scenarios, drawing preferences, output level, applicant information, inventor candidates, priority information, client template.
Output
- Patent name reverse clarification card
- List of candidate patentable plans
- Project boundary and dependency profile
- Plan code evidence mapping table
- Algorithm/software-style specification technical disclosure document
- Claim layout card
- Claim-evidence matrix
- Invention patent draft
- Invention patent draft self-check table
- List of items pending confirmation