Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Evaluate novelty of specific invention — find relevant prior art across patents, publications, and products. Budget: 80 patent families, 20 claim parses, 50 web searches.
.claude/skills/yogsoth-ai-prior-art-search/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 22% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 6% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 5% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 34% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 26% | 0% |
Targeted prior art search to evaluate the novelty of a specific invention. Searches patents, academic publications, and commercial products for anticipating or obviating references.
Determine whether a specific invention is novel and non-obvious by systematically finding the closest prior art. Supports freedom-to-operate analysis and patentability assessments.
| Metric | Target | |--------|--------| | Patent families analyzed | 80 | | Claim parses | 20 | | Web searches completed | 50 |
| Metric | Target | Current | % | |--------|--------|---------|---| | Patent families analyzed | 80 | 0 | 0% | | Claim parses completed | 20 | 0 | 0% | | Web searches completed | 50 | 0 | 0% | | Relevant prior art refs | — | 0 | — |
HARD-GATE: Cannot exit iteration loop until 80% of patent families (64), claim parses (16), and web searches (40) budget met.
| Tactic | When to Use | |--------|-------------| | patent-family-tracing | Trace citation chains from closest prior art | | classification-navigation | Find related patents via IPC/CPC neighborhood | | claim-decomposition | Parse claims of closest references for element-by-element comparison |
| SOP | Role in This Strategy | |-----|----------------------| | patent-query-formulation | Generate targeted queries from invention disclosure | | claim-parsing | Parse claims of found prior art for comparison | | citation-network-analysis | Identify main citation paths leading to the invention | | legal-status-assessment | Check if prior art references are still active | | quality-scoring | Rank prior art by relevance and quality | | saturation-detection | Check if prior art search has converged | | patent-synthesis | Produce prior art search report |
patent-query-formulation with invention features as inputpatent-family-tracing from closest hitsclassification-navigation in relevant IPC/CPC areasclaim-parsingsaturation-detectionmarkdown# Prior Art Search Report: [Invention Title] ## Invention Summary [Key technical features of the target invention] ## Search Strategy [Queries used, databases searched, classification codes explored] ## Closest Prior Art (ranked by relevance) | Rank | Reference | Type | Date | Relevance | Elements Anticipated | |------|-----------|------|------|-----------|---------------------| ## Element-by-Element Comparison [Feature matrix: invention elements vs. prior art coverage] ## Novelty Assessment [Analysis of which elements are novel vs. anticipated] ## Non-Obviousness Analysis [Combination analysis — would PHOSITA combine references?] ## Conclusion [Overall patentability/FTO assessment]
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| Tactic | When to use | | --- | --- | | classification-navigation | IPC/CPC hierarchy drill-down and lateral expansion for patent discovery | | knowledge-acquisition-claim-decomposition | Independent/dependent claim parsing, element extraction, and feature mapping to technical domains | | patent-family-tracing | Forward/backward patent citation and priority tracing until saturation |
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | citation-network-analysis | Build and analyze patent citation networks — main path analysis, PageRank, cluster detection | | claim-parsing | Patent claim syntax parsing — independent/dependent relationships and element extraction | | knowledge-acquisition-saturation-detection | Determine when additional searching yields diminishing returns. Analyzes the latest expansion batch against existing corpus to judge continue/near-saturation/saturated. Used by snowball and systematic-survey. | | legal-status-assessment | Determine patent legal status — active, expired, pending, lapsed, or revoked | | patent-query-formulation | Construct keyword + IPC/CPC + assignee combination search strategies for patent databases | | patent-synthesis | Produce final structured patent intelligence report from all analysis results | | quality-scoring | Multi-dimensional patent quality assessment — forward citations, family size, claim count, geographic breadth |
<!-- END available-tables (generated) -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 47,957 | 98,666 | +106% | 1 | 1 | 0% | 6,756 | 8,219 | +22% | 0 | 0 | — |
case-02 | fail→fail | 51,799 | 50,929 | -2% | 1 | 1 | 0% | 8,316 | 9,421 | +13% | 0 | 0 | — |
case-03 | fail→fail | 43,240 | 49,165 | +14% | 1 | 1 | 0% | 6,492 | 9,397 | +45% | 0 | 0 | — |
case-04 | fail→pass | 15,295 | 13,252 | -13% | 1 | 1 | 0% | 2,341 | 2,491 | +6% | 0 | 0 | — |
case-05 | fail→pass | 13,218 | 11,672 | -12% | 1 | 1 | 0% | 2,093 | 2,188 | +5% | 0 | 0 | — |
case-06 | pass→pass | 19,521 | 23,242 | +19% | 1 | 1 | 0% | 2,426 | 4,132 | +70% | 0 | 0 | — |
case-07 | pass→pass | 17,135 | 21,866 | +28% | 1 | 1 | 0% | 2,010 | 4,861 | +142% | 0 | 0 | — |
case-08 | pass→pass | 19,845 | 25,780 | +30% | 1 | 1 | 0% | 2,520 | 4,634 | +84% | 0 | 0 | — |
case-09 | fail→pass | 20,945 | 12,210 | -42% | 1 | 1 | 0% | 2,358 | 3,150 | +34% | 0 | 0 | — |
case-10 | fail→fail | 40,602 | 21,287 | -48% | 1 | 1 | 0% | 3,062 | 4,799 | +57% | 0 | 0 | — |
case-11 | pass→pass | 19,686 | 21,456 | +9% | 1 | 1 | 0% | 2,598 | 3,872 | +49% | 0 | 0 | — |
case-12 | pass→pass | 12,008 | 18,544 | +54% | 1 | 1 | 0% | 1,419 | 2,848 | +101% | 0 | 0 | — |
case-13 | pass→pass | 18,159 | 43,801 | +141% | 1 | 1 | 0% | 2,249 | 9,329 | +315% | 0 | 0 | — |
case-14 | pass→pass | 21,821 | 14,246 | -35% | 1 | 1 | 0% | 2,819 | 2,761 | -2% | 0 | 0 | — |
case-15 | fail→fail | 27,171 | 43,575 | +60% | 1 | 1 | 0% | 3,574 | 9,359 | +162% | 0 | 0 | — |
case-16 | fail→fail | 59,719 | 41,172 | -31% | 1 | 1 | 0% | 8,244 | 9,349 | +13% | 0 | 0 | — |
case-17 | fail→fail | 23,981 | 64,733 | +170% | 1 | 1 | 0% | 3,899 | 9,345 | +140% | 0 | 0 | — |
case-18 | fail→pass | 42,785 | 7,749 | -82% | 1 | 1 | 0% | 1,338 | 1,684 | +26% | 0 | 0 | — |
case-19 | fail→pass | 36,716 | 2,981 | -92% | 1 | 1 | 0% | 2,807 | 1,472 | -48% | 0 | 0 | — |
case-20 | fail→pass | 16,487 | 12,211 | -26% | 1 | 1 | 0% | 2,588 | 2,841 | +10% | 0 | 0 | — |
case-21 | pass→pass | 17,060 | 33,980 | +99% | 1 | 1 | 0% | 2,502 | 6,464 | +158% | 0 | 0 | — |
case-22 | pass→pass | 20,139 | 23,007 | +14% | 1 | 1 | 0% | 2,299 | 4,973 | +116% | 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 20 counted toward the lift figure. The other 2 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 +32 percentage points is the difference between those two pass rates over the 20 comparable cases. 1 case got worse with the skill loaded, and it is 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.