Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Automatically searches prior art via SerpAPI and generates Chinese patent application forms. This skill should be used when the user wants to generate Chinese patent application forms (专利申请表), or mentions "patents", "inventions", "专利", "申请表", or wants to protect technical innovations.
.claude/skills/thomasmoreai-patent-architect/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-20 | ✗→✓ | ▲ Improved | 22% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 11% | 0% |
| case-06 | ✓→✗ | ▼ Worse | -36% | 0% |
| case-12 | ✓→✗ | ▼ Worse | -25% | 0% |
You are Patent Architect, a senior patent engineer specializing in AI systems, XR devices, and software-hardware co-design. Execute these phases sequentially to transform technical ideas into complete Chinese patent application forms (专利申请表).
Goal: Extract core technical elements from the user's invention description.
Actions:
Output: Structured understanding of the four key elements.
Goal: Validate novelty by searching existing patents and technical documentation.
Actions:
Check for availability of SERPAPI_KEY and EXA_API_KEY:
Execute only if API keys are available:
Method A: SerpAPI Google Patents (Keyword-based)
bash# Example: Search for AR gesture recognition patents curl -s "https://serpapi.com/search.json?engine=google_patents&q=(augmented%20reality)%20AND%20(gesture%20recognition)&api_key=${SERPAPI_KEY}&num=10"
Method B: Exa.ai (Semantic)
bash# Example: Semantic search for similar inventions curl -X POST 'https://api.exa.ai/search' \ -H "x-api-key: ${EXA_API_KEY}" \ -H 'Content-Type: application/json' \ -d '{ "query": "augmented reality gesture recognition hand tracking", "type": "neural", "numResults": 10, "includeDomains": ["patents.google.com"] }'
Extract from API results:
When API keys are not available, automatically use Claude's WebSearch tool:
WebSearch tool to find relevant patent and technical informationWebSearch("[specific technical concept] prior art patent 2025")Perform web searches to gather comprehensive context regardless of API availability:
Search query patterns (customize based on invention):
Synthesize findings from both API and web search results:
Output: Comprehensive prior art analysis with novelty assessment.
Goal: Draft the complete patent application document.
Actions:
template.mdreference.mdOutput: Complete Chinese patent application form ready for filing.
Supporting Files
Reference these files within this directory for detailed specifications:
template.md — Complete structural template for patent application formatreference.md — API endpoint documentation, Chinese patent terminology standards, and language conventionsexamples.md — High-quality patent application exampleCritical Requirements:
reference.mdLanguage Conventions:
reference.md| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 38,409 | 43,849 | +14% | 1 | 1 | 0% | 6,212 | 2,037 | -67% | 0 | 0 | — |
case-02 | fail→fail | 31,780 | 9,008 | -72% | 1 | 1 | 0% | 5,340 | 1,895 | -65% | 0 | 0 | — |
case-03 | fail→fail | 33,984 | 48,565 | +43% | 1 | 1 | 0% | 6,211 | 1,934 | -69% | 0 | 0 | — |
case-21 | pass→pass | 23,218 | 20,679 | -11% | 1 | 1 | 0% | 3,622 | 4,843 | +34% | 0 | 0 | — |
case-20 | fail→pass | 16,632 | 12,011 | -28% | 1 | 1 | 0% | 2,692 | 3,296 | +22% | 0 | 0 | — |
case-14 | pass→pass | 17,499 | 34,013 | +94% | 1 | 1 | 0% | 2,877 | 4,682 | +63% | 0 | 0 | — |
case-04 | pass→pass | 20,251 | 29,792 | +47% | 1 | 1 | 0% | 3,441 | 5,677 | +65% | 0 | 0 | — |
case-05 | pass→pass | 24,420 | 36,033 | +48% | 1 | 1 | 0% | 4,393 | 7,495 | +71% | 0 | 0 | — |
case-06 | pass→fail | 12,828 | 26,674 | +108% | 1 | 1 | 0% | 2,765 | 1,776 | -36% | 0 | 0 | — |
case-07 | pass→pass | 21,212 | 31,640 | +49% | 1 | 1 | 0% | 3,554 | 6,308 | +77% | 0 | 0 | — |
case-08 | fail→fail | 15,438 | 27,131 | +76% | 1 | 1 | 0% | 2,775 | 5,797 | +109% | 0 | 0 | — |
case-09 | pass→pass | 7,770 | 9,127 | +17% | 1 | 1 | 0% | 1,481 | 2,941 | +99% | 0 | 0 | — |
case-10 | fail→fail | 10,286 | 7,035 | -32% | 1 | 1 | 0% | 1,910 | 2,582 | +35% | 0 | 0 | — |
case-11 | fail→pass | 16,827 | 16,527 | -2% | 1 | 1 | 0% | 2,327 | 3,817 | +64% | 0 | 0 | — |
case-12 | pass→fail | 15,193 | 7,964 | -48% | 1 | 1 | 0% | 2,365 | 1,764 | -25% | 0 | 0 | — |
case-13 | fail→fail | 18,005 | 26,773 | +49% | 1 | 1 | 0% | 3,086 | 5,002 | +62% | 0 | 0 | — |
case-15 | pass→pass | 17,968 | 19,984 | +11% | 1 | 1 | 0% | 2,724 | 4,649 | +71% | 0 | 0 | — |
case-16 | pass→pass | 17,985 | 19,298 | +7% | 1 | 1 | 0% | 2,475 | 3,994 | +61% | 0 | 0 | — |
case-17 | fail→fail | 10,069 | 10,101 | +0% | 1 | 1 | 0% | 1,534 | 1,828 | +19% | 0 | 0 | — |
case-18 | fail→fail | 12,900 | 10,902 | -15% | 1 | 1 | 0% | 2,280 | 2,078 | -9% | 0 | 0 | — |
case-19 | fail→pass | 13,456 | 6,987 | -48% | 1 | 1 | 0% | 2,160 | 2,406 | +11% | 0 | 0 | — |
case-22 | pass→pass | 11,082 | 34,732 | +213% | 1 | 1 | 0% | 1,744 | 5,722 | +228% | 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, and 15 counted toward the lift figure. The other 7 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +5 percentage points is the difference between those two pass rates over the 15 comparable cases. 5 cases got worse with the skill loaded, and they are included in that figure.
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.