| name | patent-continuous-learning |
| description | Automatically extract reusable patterns from patent drafting sessions, including keyword strategies, writing techniques, and search methods, to build an accumulative patent knowledge base. |
| origin | professional-patent-agents |
| version | 1.0.0 |
Patent Continuous Learning Skill
Automatically extract reusable patterns from patent drafting sessions to form a patent knowledge base.
Trigger Conditions
- After patent drafting is complete (patent-auditor review passed)
- When search strategy is particularly effective
- When user corrects writing style
- When new writing techniques or patterns are discovered
- When user provides access to patent database APIs
- When new patent search skills are found on ClawHub
Core Concept: Patent Instinct
A Patent Instinct is an atomic learning unit that records a specific patent-related experience:
---
id: prefer-quantified-effect
trigger: "When writing technical effects"
confidence: 0.8
domain: "patent-writing"
source: "session-observation"
scope: global
---
When writing the "Advantages Over Prior Art" section of a patent
Use quantified data to describe technical effects, such as:
- Efficiency improved by XX%
- Latency reduced by XXms
- Success rate improved by XX%
- 2026-03-19: User corrected "high efficiency" to "efficiency improved by 30%"
- 2026-03-18: Audit recommendation to add quantified data
Patent Instinct Types
| Type | Description | Scope |
|---|
keyword-strategy | Effective search keyword combinations | project |
writing-pattern | Writing techniques and sentence patterns | global |
tech-description | Technical description patterns | project |
claim-structure | Claim structure patterns | global |
search-tactic | Search platform usage tips | global |
error-avoidance | Common error avoidance | global |
api-recommendation | Patent database API recommendations | global |
skill-discovery | ClawHub skill discovery patterns | global |
Confidence Evolution
| Score | Meaning | Behavior |
|---|
| 0.3 | Tentative | Suggest but don't enforce |
| 0.5 | Medium | Apply when relevant |
| 0.7 | Strong | Auto-apply |
| 0.9 | Certain | Core behavior |
Confidence Increase:
- Pattern observed repeatedly
- User confirms effectiveness
- Audit passed
Confidence Decrease:
- User explicitly corrects
- Causes problems
Learning Flow
Patent drafting session
|
| Observe key events
v
+------------------------------------------+
| observations/ |
| - Successful search strategies |
| - User correction records |
| - Audit feedback |
| - Newly discovered patterns |
+------------------------------------------+
|
| Extract instincts
v
+------------------------------------------+
| instincts/ |
| - keyword-strategy/ (project scope) |
| - writing-pattern/ (global scope) |
| - tech-description/ (project scope) |
+------------------------------------------+
|
| /evolve clustering
v
+------------------------------------------+
| evolved/ |
| - skills/patent drafting enhanced skill |
| - templates/reusable templates |
+------------------------------------------+
Commands
| Command | Description |
|---|
/patent-learn | Extract patent instincts from current session |
/patent-instincts | Display learned patent instincts |
/patent-evolve | Cluster related instincts into skills |
Directory Structure
patent/
├── learning/
│ ├── observations.jsonl # Observation records
│ ├── instincts/
│ │ ├── global/ # Global instincts
│ │ │ ├── prefer-quantified-effect.yaml
│ │ │ └── avoid-complete-code.yaml
│ │ └── projects/
│ │ └── project-name/ # Project scope
│ │ ├── keyword-strategy.yaml
│ │ └── tech-description.yaml
│ └── evolved/
│ ├── skills/
│ └── templates/
Example: Auto-learned Instincts
Patent Database API Recommendation
---
id: recommend-patent-database-api
trigger: "When starting patent prior art search"
confidence: 0.9
domain: "api-recommendation"
scope: global
---
When user requests patent prior art search and default channels may not be sufficient.
1. Ask user about available patent database APIs
2. Recommend appropriate APIs based on search needs:
- Global search: Google Patents, Lens.org
- US patents: USPTO, PatentsView
- European patents: EPO Espacenet
- Chinese patents: CNIPA
Search Keyword Strategy
---
id: keyword-device-pairing
trigger: "When searching device pairing patents"
confidence: 0.85
domain: "keyword-strategy"
scope: project
project: example-project
---
- Primary keywords: device, terminal, pairing, connection
- Combination methods: `device pairing`, `terminal quick connection`
- Platform preference: Google Patents (English), AMiner (Academic)
- 2026-03-18: Found 5 highly relevant references using this combination
- Confidence increased from 0.5 to 0.85
Writing Technique
---
id: avoid-complete-code
trigger: "When writing patent embodiments"
confidence: 0.95
domain: "writing-pattern"
scope: global
---
Patent documents should not contain complete executable code. Use instead:
- Algorithm pseudocode
- Flowcharts
- Functional module descriptions
- 2026-03-17: Audit found complete code, recommended removal
- 2026-03-18: User confirmed this rule
- Verified across multiple patents
Integration into Patent Workflow
Auto-trigger learning in all three scenarios:
Scenario 1: User Idea → Drafting
After patent-auditor review passes
|
| Check for new patterns learned
v
patent-continuous-learning extracts instincts
Scenario 2: User Draft → Optimization
User correction or audit recommendation
|
| Record effective improvements
v
patent-continuous-learning updates instincts
Scenario 3: Agency Feedback
Targeted optimization successful
|
| Record effective differentiation descriptions
v
patent-continuous-learning updates instincts
