# Patent Prior Art > Runs a systematic prior-art search - feature decomposition into claim elements, CPC/IPC classification plus keyword vocabulary, patent databases AND non-patent literature, forward and backward citation walks - and maps results into claim charts with a novelty assessment and a documented search log. Use when someone asks "is my invention novel", "has anyone already patented this", "find prior art against this claim", or is deciding whether a filing or an invalidity argument is worth pursuing. This is research, not legal advice; filing and freedom-to-operate decisions require a patent attorney. Do NOT use for a general academic literature survey unrelated to patentability - use literature-review instead. - Skill: `skillmedev/patent-prior-art` (Agent Skill) - Install (CLI): `npx skillmds@latest add skillmedev/patent-prior-art` - Raw SKILL.md: https://api.skillmd.com/api/skills/skillmedev/patent-prior-art/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Research & Search - Author: SkillMedev (https://skillmd.com/u/skillmedev) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/skillmedev/patent-prior-art --- # Patent Prior Art Assess whether an invention is novel and non-obvious by finding everything that already discloses it. The costly failure this skill prevents is the false-clear search: keywords-only, English-only, patents-only searching that misses the conference paper or Japanese utility model that later kills the application - after the filing fees, or worse, during litigation. This is a search-and-map task, not a legal opinion. ## Inputs to collect 1. **The invention disclosure**: what it does, how it works, and what the inventor believes is new. 2. **The priority date** (or intended filing date) - everything published on or after it is irrelevant, and date discipline is absolute. 3. **The field and its jargon**: how practitioners, not just patent drafters, name these concepts. 4. **Known competitors and prior products** in the space - likely assignees to search. 5. **Search depth**: quick patentability screen (top databases, hours) vs thorough pre-filing search (all sources below, days). Default to thorough if the user will spend money on the outcome. ## Operating procedure ### Step 1: Deconstruct the invention into claim elements Break the invention into its essential technical features - the problem solved, the mechanism, and what is allegedly new. Separate the novel core from conventional surrounding components; searching the conventional parts wastes effort, and missing the core invalidates the search. Write each element as one row of the eventual claim chart. ### Step 2: Build the search vocabulary - For each element, list synonyms, formal technical terms, and layperson terms - patent drafters deliberately use unusual wording, so keyword search alone systematically misses. - Map elements to classification codes: **CPC and IPC**. Classification search catches documents that use entirely different words for the same concept; a search without a classification pass is incomplete by construction. - List likely assignees and named inventors active in the field for targeted searches. ### Step 3: Search systematically - patents AND non-patent literature Scope rule: any public disclosure anywhere, in any language, before the priority date counts as prior art. A patents-only search is not a prior-art search. - **Patent databases**: Google Patents, Espacenet, USPTO (Patent Public Search), WIPO PATENTSCOPE. Espacenet and PATENTSCOPE cover foreign filings - do not stop at US results. - **Non-patent literature (NPL)**: academic papers (Google Scholar, IEEE, arXiv), conference proceedings, standards documents, product manuals, datasheets, theses, and archived web pages (Wayback Machine). For software and fast-moving fields, NPL is where the killing reference usually lives. - Run three query types and combine results: keyword queries, classification queries, and **citation walks** - for each strong hit, follow its backward citations (what it built on) and forward citations (what built on it). - Log every query, database, and date in the search log as you go - an undocumented search cannot be defended or extended later. ### Step 4: Screen and map into claim charts For each candidate reference, build a claim chart: invention elements in rows, one column per reference, each cell marked disclosed / partially disclosed / not disclosed, with a pinpoint cite (column and line, paragraph, or figure). A single reference disclosing **all** elements defeats novelty; a plausible **combination** of references may defeat non-obviousness. ### Step 5: Assess novelty and obviousness - **Novelty**: is there one prior document disclosing every essential element? Name it. - **Non-obviousness / inventive step**: would combining known references be obvious to a person skilled in the art? Note the motivation to combine (same problem, same field, explicit suggestion) and any unexpected results cutting the other way. - Record the publication date of every reference and verify it precedes the priority date - a disclosure one day after the priority date is not prior art, and an assumed date is not a verified date. ### Step 6: Report - including the negative result Deliver the feature breakdown, the complete search log, the top references with claim charts, and a novelty assessment with confidence. List the strongest single reference and the strongest combination **separately** - they answer different legal questions. "No blocking art found in the scope searched" is a valid, documented outcome; state the scope so its limits are visible. ## Worked artifact: search log template Copy and fill one row per query as the search runs. ``` PRIOR-ART SEARCH LOG Invention: [FILL: one-line description] Priority date: [FILL: YYYY-MM-DD] - only disclosures BEFORE this date count Searcher: [FILL] Search depth: [FILL: screen / thorough] CLAIM ELEMENTS E1: [FILL: essential feature 1] E2: [FILL: essential feature 2] E3: [FILL: add rows as needed] CLASSIFICATION CODES CONSULTED CPC: [FILL: e.g. G06F 16/903] IPC: [FILL] QUERY LOG | # | Date run | Database | Query / class / citation walk | Hits reviewed | Kept | |---|----------|----------|-------------------------------|---------------|------| | 1 | [FILL] | [FILL] | [FILL: exact query string] | [FILL] | [FILL: ref IDs] | REFERENCES KEPT | Ref | Type (patent/NPL) | Pub. date | Before priority? | Elements disclosed | Pinpoint cites | |-----|-------------------|-----------|------------------|--------------------|----------------| | R1 | [FILL] | [FILL] | [FILL: yes/no] | [FILL: E1, E3] | [FILL] | STRONGEST SINGLE REFERENCE: [FILL: ref + which elements it covers/misses] STRONGEST COMBINATION: [FILL: refs + motivation to combine] SCOPE LIMITS: [FILL: languages, databases, or NPL sources NOT searched] ``` ## Deliverable Produce a prior-art report containing: the claim-element breakdown, the filled search log (every query, database, and date), claim charts for the top references with pinpoint citations, the strongest single reference and strongest combination stated separately, a novelty/obviousness assessment with confidence, and the explicit scope limits of the search. ## Do NOT - Do not search patents only - NPL, product documentation, and archived web pages are equally fatal prior art, and in software usually more so. - Do not search English only - foreign-language disclosures count in full; use Espacenet and PATENTSCOPE machine translation rather than skipping them. - Do not skip the classification pass - keyword-only searches miss documents that describe the same mechanism in different words. - Do not cite a reference without verifying its publication date against the priority date. - Do not bury a negative result - "no blocking art found" with a documented scope is a deliverable, not a failure. - Do not drift into legal conclusions ("this is patentable", "you don't infringe") - the report maps evidence; an attorney draws conclusions. ## Quality bar - Every claim element appears as a row in at least one claim chart. - Every reference kept has a verified publication date earlier than the priority date. - The search log lets a second searcher reproduce or extend the search without guessing what was covered. - The report states the strongest single reference, the strongest combination, and the scope limits - all three, always. ## Escalation This is research, not legal advice. Patentability opinions, freedom-to-operate analysis, filing strategy, and invalidity positions require a registered patent attorney or agent - recommend one whenever the user's next step involves spending money at a patent office or asserting/defending against a patent. For surveying a research field without a patentability question, route to literature-review.