patent

---
name: patent
description: Use when the user asks for patent prior-art searching or landscape analysis — novelty search, freedom-to-operate, competitive landscape, acquisition diligence, or litigation prior-art.
category: Research
version: 1.0.0
tools: []
---

# Patent — Prior-Art + Landscape Intelligence

Produce prior-art and patent-landscape research signal, delivered as a formatted chat report. This is technical search signal, **not legal advice** — always close by recommending the user consult a patent attorney before any filing or licensing decision. Out of scope: trademark, copyright, trade-secret questions — flag those and redirect to qualified counsel.

## The Differentiator: Commit to a Sub-Use-Case First

Refuse to be a generic "search patents" bucket. Every invocation commits to one of five sub-use-cases via intake before any search runs, because each dictates a different search strategy and report emphasis:

| Sub-use-case | Search strategy | Report emphasis |
|---|---|---|
| Novelty search | Narrow + claims-text focused; filing date irrelevant | Closest art + claim differentiation |
| Freedom-to-operate | Broad + active-patents-only; jurisdiction-filtered | FTO risk flags + claim-by-claim risk |
| Competitive landscape | Broad + filer tally + classification trends | Filer map + investment hotspots |
| Acquisition diligence | Specific assignee + portfolio scope + ownership chain | Portfolio table + ownership verification |
| Litigation prior-art | Specific target patent + adjacent art before its priority date | Ranked knock-out candidates |

## Integrity Rules

- Search sequentially. Cite only patents actually surfaced by search this session; label anything else `[Not from search — reference information]` and exclude it from counts.
- Track queries sent / patents found / patents cited, surfaced in the closing audit section.
- On failure, retry once with reworded terms; after repeated failures across sources, stop and tell the user what's missing.

## Phase 1: Intake (6 questions, one at a time)

1. **Invention description**, 2-3 sentences: what it does, what's new. Reject generic answers ("AI for healthcare," "a better widget") — push once for "what does it do that existing systems don't?" then proceed with a caveat if still vague.
2. **Sub-use-case commitment** — pick one of the five above; refuse to start without a choice. If the user says "all of them," get the primary purpose and treat the rest as follow-ups.
3. **Jurisdictions** (asked only for FTO, landscape, or diligence) — US / EP / CN / JP / KR / PCT / worldwide. Skipped for novelty (priority date is jurisdictionally portable) and litigation (jurisdiction follows the target patent).
4. **Known prior art** — a patent number or paper the user has already seen, if any. Anchoring one known piece of art makes the search dramatically more precise than starting cold; "none" is a fine answer.
5. **Risk tolerance** (asked for novelty and FTO) — strict (one close hit means abandon the path) vs. signal-gathering (wants the lay of the land regardless). Skipped for landscape, which is signal-gathering by definition.
6. **Attorney status** (asked only for novelty and FTO) — confirm the user understands this produces search signal, not legal advice; this triggers the legal-disclaimer footer in the report.

Stop after question 6 (or earlier via dependency skips) and commit — never reopen intake mid-run.

## Phase 2: Search Strategy by Sub-Use-Case

- **Novelty:** narrow queries on invention-specific terminology, broad synonym queries, and a classification-restricted query once a likely classification is identified from initial hits.
- **Freedom-to-operate:** filter to active (not expired/abandoned) patents, priority date before today, and pull active claim text for each hit.
- **Competitive landscape:** broader technology-space queries, then tally the top filers by classification and describe the filing trend over the last decade for the top 5.
- **Acquisition diligence:** search by specific assignee (target + subsidiaries + named inventors), check the assignment/ownership chain, and resolve patent families to avoid double-counting.
- **Litigation prior-art:** take the target patent number, extract its priority date, and search for adjacent art with similar classification and claim language published before that date.

Search sources in priority order: Google Patents (broad coverage, no auth), Espacenet (strong non-US coverage), USPTO (US deep dive), and — if the user has a Lens.org key — Lens.org for citation-graph signal; note if it's unavailable and recommend manual citation review instead.

## Phase 3: Claim Extraction + Relevance Scoring

For each closest-art hit, pull independent claim 1 (the broadest claim, and the primary anticipation/obviousness vehicle) plus any dependent claims that add the inventive step, and score relevance by how much its claim language overlaps with the invention description. Rank by score and assign a verdict appropriate to the sub-use-case: NOVEL / POTENTIALLY NOVEL / NOT NOVEL for novelty searches; CLEAR / FLAGGED / HIGH RISK per jurisdiction for FTO.

**Classification awareness matters:** keyword search alone misses adjacent art. After the initial pass, pull the classification codes (CPC/IPC) from the top 5 hits and run one classification-restricted query — this consistently surfaces art that keyword search misses.

## Phase 4: Citation Signal + Family Resolution

If a citation-graph tool (e.g., Lens.org) is available, use it to identify foundational patents (high cited-by counts, roughly 50+) and recent high-citation activity as a proxy for current activity; if unavailable, skip and recommend manual citation review on Google Patents instead. Group hits by patent family (same invention filed across multiple jurisdictions) to avoid double-counting the same invention as multiple distinct results.

## Date Discipline

Distinguish four dates at every hit — filing date (first submission), priority date (earliest claim of priority, often earlier than filing), publication date (typically ~18 months after priority), and grant date — and surface the one that's legally relevant to the sub-use-case: priority date for novelty, grant date + active status for FTO, publication date for landscape, grant + assignment date for diligence, and the target patent's priority date for litigation (it sets the prior-art cutoff).

## Report Structure

1. **Executive Summary + Verdict** — sub-use-case banner, one-line verdict, 3-4 key findings, legal-disclaimer footer.
2. **Closest Prior Art** — 5-10 patents ranked, each with title/link, assignee, filing/priority dates, independent claim 1 text, relevance score, and rationale.
3. **Patent Landscape** — top-filer table and classification distribution (landscape/diligence; abbreviated otherwise).
4. **Citation Signals** — foundational and recent high-cite patents if available; otherwise "manual review recommended."
5. **Geographic Coverage** — filings by jurisdiction for the top hits (FTO/landscape/diligence only).
6. **FTO Flags** (FTO only) — active patents posing infringement risk, each with jurisdiction, relevant claims, risk level, and a mitigation note.
7. **Strategy + Recommendations**, tailored to the sub-use-case (claim-differentiation suggestions for novelty; design-around hints for FTO; who-to-watch for landscape; red flags for diligence; ranked knock-out candidates for litigation) — always closing with the mandatory attorney-consultation reminder.
8. **Audit Log** — searches run, source, results, counts, any failed steps.

## Error Handling

| Situation | Response |
|---|---|
| User won't commit to a sub-use-case | Refuse to proceed; re-ask with examples. |
| Invention description stays generic | Reject and re-ask "what does it do that existing systems don't?" |
| A source rate-limits or fails | Retry once after a pause; fall back to an alternate source and log it. |
| Citation-graph tool unavailable | Skip that section, note "manual review recommended." |
| Claim text extraction fails | Fall back to the abstract; flag the result as "abstract-only." |
| Family resolution incomplete | Note it; duplicates may remain; suggest manual dedup. |
| All searches return fewer than 3 hits | State explicitly — "either niche art or a genuine gap" — never fabricate results. |
| Repeated failures across sources | Stop, tell the user what's missing, share what was found. |