Search Prior Art
Structured prior art search → find pubs|patents|products|disclosures predating invention. Used for patentability (can patent?), validity challenge (should have been granted?), FTO (covered by existing rights?).
Use When
- Eval novelty+non-obvious pre-file
- Challenge existing patent validity → find art examiner missed
- Support FTO → find art limiting blocking patent scope
- Document defensive pub → prevent others patenting concept
- Respond to office action questioning novelty|obviousness
In
- Required: Invention desc (what, how, problem)
- Required: Purpose (patentability|invalidity|FTO|defensive)
- Required: Critical date (filing date or invention date)
- Optional: Known related patents|pubs
- Optional: Tech classification codes (IPC, CPC)
- Optional: Key inventors|companies
Do
Step 1: Decompose Invention
Break into constituent technical features.
- Read desc (or claims if vs existing patent)
- Extract essential elements — each independent feature:
- Components?
- Process steps?
- Technical effect?
- Problem + how solved?
- ID novel combination — what's diff from known:
- New element added to known?
- New combo of known?
- Known element new field?
- Gen search terms per element:
- Tech terms, synonyms, abbrev
- Broader+narrower (hierarchy)
- Alt descriptions
- Doc Search Map: elements, terms, relationships
Search Map Example:
+------------------+-----------------------------------+-----------+
| Element | Search Terms | Priority |
+------------------+-----------------------------------+-----------+
| Attention layer | attention mechanism, self- | High |
| | attention, multi-head attention | |
| Sparse routing | mixture of experts, sparse MoE, | High |
| | top-k routing, expert selection | |
| Training method | knowledge distillation, teacher- | Medium |
| | student, progressive training | |
+------------------+-----------------------------------+-----------+
→ Complete decomposition w/ terms per element. Novel combo ID'd → search must find (invalidate) or confirm absent (support novelty).
If err: too abstract → ask more specific. Claims unclear → broadest reasonable interp per element.
Step 2: Search Patent Literature
Patent DBs systematic.
- Construct queries:
- Each element individually first (broad)
- Combine to find closer art (narrow)
- Classification codes filter by tech area
- Multi DBs:
- Google Patents: Full-text, free, large
- USPTO PatFT/AppFT: US patents+apps, official
- Espacenet: EU, excellent classification
- WIPO Patentscope: PCT, global
- Date filters:
- Prior art must predate critical date
- Up to 1yr pre-filing (grace varies by jurisdiction)
- Per relevant result record:
- Doc number, title, filing date, pub date
- Which elements disclosed (map to Search Map)
- Discloses novel combo?
- Classify by relevance:
- X: Discloses invention alone (anticipation)
- Y: Key elements, combinable (obviousness)
- A: Background art
→ Classified patent ref list mapped to elements. X (if found) = showstoppers for novelty. Y = building blocks for obviousness.
If err: no relevant patent art → doesn't mean novel — non-patent (Step 3) may have critical ref. Absence in 1 DB ≠ absence everywhere.
Step 3: Non-Patent Literature
Academic, products, OSS, other.
- Academic:
- Google Scholar, arXiv, IEEE Xplore, ACM Digital Library
- Same Step 1 terms
- Conf papers + workshop proceedings often predate patents
- Products + commercial:
- Product docs, manuals, marketing
- Internet Archive (Wayback) for date-verified web
- Trade pubs + press releases
- OSS + code:
- GitHub, GitLab — search impls of features
- READMEs, docs, commit history for date evidence
- Software releases w/ ver dates
- Standards:
- IEEE, IETF (RFCs), W3C, ISO
- Standards-essential patents must be disclosed; search standard bodies' IP DBs
- Defensive pubs:
- IBM Technical Disclosure Bulletin
- Research Disclosure journal
- IP.com Prior Art DB
- Verify pub date before critical date:
- Web: Wayback for date evidence
- Software: release dates|commit timestamps
- Papers: pub date not submission
→ Non-patent refs complement patent search. Academic + OSS often most powerful — describe details more explicitly than patents.
If err: non-patent sparse → tech primarily corp R&D (patent-heavy). Shift emphasis to patent + combo-based obviousness.
Step 4: Analyze + Map
Eval how art relates to invention.
- Claim chart mapping art → elements:
Claim Element vs. Prior Art Matrix:
+------------------+--------+--------+--------+--------+
| Element | Ref #1 | Ref #2 | Ref #3 | Ref #4 |
+------------------+--------+--------+--------+--------+
| Element A | X | X | | X |
| Element B | | X | X | |
| Element C | X | | X | |
| Novel combo A+B+C| | | | |
+------------------+--------+--------+--------+--------+
X = element disclosed in this reference
- Novelty: Single ref discloses all elements?
- Yes → anticipated (not novel)
- No → may be novel (proceed obviousness)
- Obviousness: Few refs (2-3) combinable to cover all?
- Motivation to combine? (skilled person sees reason?)
- Teach away? (suggest wouldn't work?)
- FTO: Does art narrow blocking patent claims?
- Art overlapping blocking patent's claims limits enforceable scope
- Document analysis w/ specific passage citations
→ Clear claim chart showing element coverage by refs, w/ novelty + obviousness assessment. Each mapping cites specific passages|figures.
If err: chart shows gaps (elements not in any art) → those = potentially novel. Focus follow-up on specific gaps.
Step 5: Document + Deliver
Package for intended use.
- Write Prior Art Search Report:
- Purpose + scope
- Methodology (DBs, queries, date ranges)
- Results summary (count, classification breakdown)
- Top refs w/ detailed analysis (claim charts)
- Assessment: novelty, obviousness, FTO implications
- Limitations + further-search recommendations
- Organize refs:
- Sorted by relevance (X first, Y, A)
- Each w/ full bibliographic + access link
- Key passages highlighted|extracted
- Recommendations by purpose:
- Patentability: File|don't, suggested claim scope by gaps
- Invalidity: Strongest combo, suggested legal arg
- FTO: Risk level, design-around opportunities, licensing
- Defensive: Whether to publish defensive disclosure based on white space
→ Complete organized report directly supporting decision. Refs accessible, analysis traceable.
If err: inconclusive (no strong X|Y, but relevant background) → state clearly: "No anticipatory art; closest addresses A+B not C. Recommend file w/ claims emphasizing C." Inconclusive valid + useful.
Check
Traps
- Keyword tunnel vision: Exact terms only misses synonyms. Use Step 1 hierarchy.
- Patent-only search: Non-patent (papers, products, code) often more explicit. Don't skip Step 3.
- Date carelessness: Must predate critical date. Brilliant ref 1 day after = worthless.
- Ignore foreign: Major inventions may first appear in CN|JP|KR|DE patents. MT makes searchable.
- Confirmation bias: Searching to confirm novelty vs to find invalidating art. Best search tries hardest to find closest.
- Stop too early: First results rarely best. Iterate based on field vocabulary revealed.
→
assess-ip-landscape — broader landscape mapping
screen-trademark — TM-specific (diff DBs + legal frame than patent)
file-trademark — TM filing post-screen
review-research — lit review methodology overlaps
security-audit-codebase — systematic methodology parallels (thoroughness, doc, reproducibility)
1---2name: search-prior-art-23title: Search Prior Art4description: Prior art search for invention|patent claim. Patent + non-patent (academic, products, OSS), defensive pubs, standard-essential patents. Use → eval novelty+non-obvious pre-file, challenge existing patent validity, support FTO analysis, document defensive pub, respond to office action questioning novelty.5author: pjt2226author_url: https://github.com/pjt222/agent-almanac/tree/main/i18n/caveman-ultra/skills/search-prior-art7license: MIT8version: 0.1.19execution_mode: open10jurisdiction: general11practice: ip12language: en13metadata:14 source: https://skillmd.com/skills/thomasmoreai/search-prior-art-215---1617# Search Prior Art1819Structured prior art search → find pubs|patents|products|disclosures predating invention. Used for patentability (can patent?), validity challenge (should have been granted?), FTO (covered by existing rights?).2021## Use When2223- Eval novelty+non-obvious pre-file24- Challenge existing patent validity → find art examiner missed25- Support FTO → find art limiting blocking patent scope26- Document defensive pub → prevent others patenting concept27- Respond to office action questioning novelty|obviousness2829## In3031- **Required**: Invention desc (what, how, problem)32- **Required**: Purpose (patentability|invalidity|FTO|defensive)33- **Required**: Critical date (filing date or invention date)34- **Optional**: Known related patents|pubs35- **Optional**: Tech classification codes (IPC, CPC)36- **Optional**: Key inventors|companies3738## Do3940### Step 1: Decompose Invention4142Break into constituent technical features.43441. Read desc (or claims if vs existing patent)452. Extract **essential elements** — each independent feature:46 - Components?47 - Process steps?48 - Technical effect?49 - Problem + how solved?503. ID **novel combination** — what's diff from known:51 - New element added to known?52 - New combo of known?53 - Known element new field?544. Gen search terms per element:55 - Tech terms, synonyms, abbrev56 - Broader+narrower (hierarchy)57 - Alt descriptions585. Doc **Search Map**: elements, terms, relationships5960```text61Search Map Example:62+------------------+-----------------------------------+-----------+63| Element | Search Terms | Priority |64+------------------+-----------------------------------+-----------+65| Attention layer | attention mechanism, self- | High |66| | attention, multi-head attention | |67| Sparse routing | mixture of experts, sparse MoE, | High |68| | top-k routing, expert selection | |69| Training method | knowledge distillation, teacher- | Medium |70| | student, progressive training | |71+------------------+-----------------------------------+-----------+72```7374→ Complete decomposition w/ terms per element. Novel combo ID'd → search must find (invalidate) or confirm absent (support novelty).7576If err: too abstract → ask more specific. Claims unclear → broadest reasonable interp per element.7778### Step 2: Search Patent Literature7980Patent DBs systematic.81821. Construct queries:83 - Each element individually first (broad)84 - Combine to find closer art (narrow)85 - Classification codes filter by tech area862. Multi DBs:87 - **Google Patents**: Full-text, free, large88 - **USPTO PatFT/AppFT**: US patents+apps, official89 - **Espacenet**: EU, excellent classification90 - **WIPO Patentscope**: PCT, global913. Date filters:92 - Prior art must predate **critical date**93 - Up to 1yr pre-filing (grace varies by jurisdiction)944. Per relevant result record:95 - Doc number, title, filing date, pub date96 - Which elements disclosed (map to Search Map)97 - Discloses novel combo?985. Classify by relevance:99 - **X**: Discloses invention alone (anticipation)100 - **Y**: Key elements, combinable (obviousness)101 - **A**: Background art102103→ Classified patent ref list mapped to elements. X (if found) = showstoppers for novelty. Y = building blocks for obviousness.104105If err: no relevant patent art → doesn't mean novel — non-patent (Step 3) may have critical ref. Absence in 1 DB ≠ absence everywhere.106107### Step 3: Non-Patent Literature108109Academic, products, OSS, other.1101111. **Academic**:112 - Google Scholar, arXiv, IEEE Xplore, ACM Digital Library113 - Same Step 1 terms114 - Conf papers + workshop proceedings often predate patents1152. **Products + commercial**:116 - Product docs, manuals, marketing117 - Internet Archive (Wayback) for date-verified web118 - Trade pubs + press releases1193. **OSS + code**:120 - GitHub, GitLab — search impls of features121 - READMEs, docs, commit history for date evidence122 - Software releases w/ ver dates1234. **Standards**:124 - IEEE, IETF (RFCs), W3C, ISO125 - Standards-essential patents must be disclosed; search standard bodies' IP DBs1265. **Defensive pubs**:127 - IBM Technical Disclosure Bulletin128 - Research Disclosure journal129 - IP.com Prior Art DB1306. Verify **pub date** before critical date:131 - Web: Wayback for date evidence132 - Software: release dates|commit timestamps133 - Papers: pub date not submission134135→ Non-patent refs complement patent search. Academic + OSS often most powerful — describe details more explicitly than patents.136137If err: non-patent sparse → tech primarily corp R&D (patent-heavy). Shift emphasis to patent + combo-based obviousness.138139### Step 4: Analyze + Map140141Eval how art relates to invention.1421431. **Claim chart** mapping art → elements:144145```text146Claim Element vs. Prior Art Matrix:147+------------------+--------+--------+--------+--------+148| Element | Ref #1 | Ref #2 | Ref #3 | Ref #4 |149+------------------+--------+--------+--------+--------+150| Element A | X | X | | X |151| Element B | | X | X | |152| Element C | X | | X | |153| Novel combo A+B+C| | | | |154+------------------+--------+--------+--------+--------+155X = element disclosed in this reference156```1571582. **Novelty**: Single ref discloses all elements?159 - Yes → anticipated (not novel)160 - No → may be novel (proceed obviousness)1613. **Obviousness**: Few refs (2-3) combinable to cover all?162 - Motivation to combine? (skilled person sees reason?)163 - Teach away? (suggest wouldn't work?)1644. **FTO**: Does art narrow blocking patent claims?165 - Art overlapping blocking patent's claims limits enforceable scope1665. Document analysis w/ specific passage citations167168→ Clear claim chart showing element coverage by refs, w/ novelty + obviousness assessment. Each mapping cites specific passages|figures.169170If err: chart shows gaps (elements not in any art) → those = potentially novel. Focus follow-up on specific gaps.171172### Step 5: Document + Deliver173174Package for intended use.1751761. Write **Prior Art Search Report**:177 - Purpose + scope178 - Methodology (DBs, queries, date ranges)179 - Results summary (count, classification breakdown)180 - Top refs w/ detailed analysis (claim charts)181 - Assessment: novelty, obviousness, FTO implications182 - Limitations + further-search recommendations1832. Organize refs:184 - Sorted by relevance (X first, Y, A)185 - Each w/ full bibliographic + access link186 - Key passages highlighted|extracted1873. Recommendations by purpose:188 - **Patentability**: File|don't, suggested claim scope by gaps189 - **Invalidity**: Strongest combo, suggested legal arg190 - **FTO**: Risk level, design-around opportunities, licensing191 - **Defensive**: Whether to publish defensive disclosure based on white space192193→ Complete organized report directly supporting decision. Refs accessible, analysis traceable.194195If err: inconclusive (no strong X|Y, but relevant background) → state clearly: "No anticipatory art; closest addresses A+B not C. Recommend file w/ claims emphasizing C." Inconclusive valid + useful.196197## Check198199- [ ] Invention decomposed into searchable elements200- [ ] Novel combo explicitly ID'd201- [ ] Patent DBs searched (min 2)202- [ ] Non-patent searched (academic + products + OSS)203- [ ] All refs predate critical date (verified)204- [ ] Claim chart maps elements w/ passage citations205- [ ] Novelty + obviousness assessed w/ reasoning206- [ ] Classified (X, Y, A)207- [ ] Report has methodology, limitations, recommendations208- [ ] Reproducible (queries + DBs documented)209210## Traps211212- **Keyword tunnel vision**: Exact terms only misses synonyms. Use Step 1 hierarchy.213- **Patent-only search**: Non-patent (papers, products, code) often more explicit. Don't skip Step 3.214- **Date carelessness**: Must predate critical date. Brilliant ref 1 day after = worthless.215- **Ignore foreign**: Major inventions may first appear in CN|JP|KR|DE patents. MT makes searchable.216- **Confirmation bias**: Searching to confirm novelty vs to find invalidating art. Best search tries hardest to find closest.217- **Stop too early**: First results rarely best. Iterate based on field vocabulary revealed.218219## →220221- `assess-ip-landscape` — broader landscape mapping222- `screen-trademark` — TM-specific (diff DBs + legal frame than patent)223- `file-trademark` — TM filing post-screen224- `review-research` — lit review methodology overlaps225- `security-audit-codebase` — systematic methodology parallels (thoroughness, doc, reproducibility)