| name | prior-art-researching |
| description | Conducts prior art search for patents with Moderate/Significant similarities.
Triggered when:
- The user asks to:
* "search for prior art"
* "perform prior art research"
* "find prior art references"
* "conduct prior art search"
- The user mentions:
* "prior art" with "database" or "similarities"
* "Moderate/Significant" with "prior art"
- `patents.db` exists with `similarities` table containing Moderate/Significant entries
|
Prior Art Researching
Purpose
Search for prior art references (both patent and non-patent literature) for patents with Moderate/Significant similarity levels and store results in the database for further analysis.
Prerequisites
patents.db must exist with similarities table containing Moderate/Significant entries (from claim-analyzing skill)
Constitution
Core Principles
Element-by-Element Analysis (The Golden Rule):
- Every analysis MUST test the target invention against prior art element by element
- Break down inventions into Elements A, B, C
- Find references disclosing A AND B AND C for anticipation (Novelty)
- Do not rely on "general similarity"
Comprehensive Literature Coverage:
- Use BOTH patent and non-patent literature sources
- Check academic papers, conference proceedings, and technical publications
- Document search results from both sources
Evidence-Based Reporting:
- Every assertion MUST be backed by specific citations
- Never say "This feature is known"
- Say "This feature is disclosed in [Patent ID], Column X, Line Y"
Prior Art Cutoff Date:
- Prior art must be published BEFORE the target's priority date
- Use publication dates, not priority dates, for cutoff determination
Skill Orchestration
Execute Prior Art Search
Process:
-
Get Patents to Search:
- Use
investigation-fetching skill
- Request: "Get list of patents with Moderate/Significant similarities without prior art"
-
For each patent, execute Steps 2a–2e in order:
2a. Get Patent Data:
- Invoke
Skill: google-patent-cli:patent-fetch with patent ID
- Invoke
Skill: investigation-fetching with request "Get elements for patent "
- Extract: title, abstract, claims, priority date, elements
2b. Execute Multi-Layer Search:
Execute three search layers per element:
| Layer | Purpose | Keywords | Limit |
|---|
| 1 | General terminology | High-level terms from element description | 10–20 |
| 2 | Specific nomenclature | Model names, algorithms, parameter names | 30–50 |
| 3 | Functional/role-based | "configured to", "means for" | 10–20 |
- Include
publication_before: "<priority date>" in all searches
2c. Screen and Analyze Results:
- Identify Grade A candidates (highly relevant), verify publication dates
- For patent references: invoke
Skill: google-patent-cli:patent-fetch with patent ID to get full details
- For NPL: invoke
Skill: arxiv-cli:arxiv-fetch for full text
- Do NOT delegate to subagents (Agent tool) — invoke Skills directly from this session
- Create claim charts with paragraph-level citations
2d. Record Results:
- Invoke
Skill: investigation-recording with prior art data for each reference:
- patent_id, claim_number, element_label, reference_id, reference_type, title, relevance_level (Significant/Moderate/Limited), analysis_notes, publication_date, claim_chart
- CRITICAL: Record at ELEMENT LEVEL (each reference linked to claim_number and element_label)
-
Verify Results: Confirm all prior arts recorded to database. Provide summary with:
- Patent ID and title
- Number of prior art references found
- Relevance levels for each reference
- Key findings summary
- Overall similarity assessment
State Management
Initial State
- Patents in
similarities table with Moderate/Significant levels without corresponding prior_arts entries exist
Final State
- No patents in
similarities table with Moderate/Significant levels without corresponding prior_arts entries (all searched)