patent-reviewer

name

patent_reviewer

description

Simulated USPTO examiner panel for quantum-computing patents. Ingests a Google Patents URL / publication number / saved file, examines every claim individually under 35 U.S.C. §§ 101/102/103/112 + a quantum-operability check, runs claim-compliance and full-application review under USPTO/EPO/PCT standards, and emits an Office Action plus a deterministic machine-actionable `_office_action.json` (disposition, rejected/allowed claims, rejections-by-statute). The patent analogue of quantum_reviewer.

patent_reviewer — USPTO examiner panel

The quantum-patent analogue of quantum_reviewer. Where the paper reviewer runs a journal referee panel, this runs a USPTO examining unit and produces an Office Action keyed to the patent statutes.

It also carries the patent-attorney workflow:

• Check claims for compliance. Run draft or published claims through USPTO 35 U.S.C. § 112(b) definiteness / antecedent-basis / structure review, EPO Art. 84 EPC clarity / support / two-part-form review, or PCT clarity/support review via --filing-standard.
• Review the full application. Specification adequacy, enablement, written description/support, formalities, abstract/title/drawings, and required sections are checked under USPTO, EPO, PCT, or multi standards.

Why patents need their own reviewer

Patent review is not peer review. The object is the claim set, examined claim-by-claim under 35 U.S.C.:

• § 101 — patent-eligible subject matter (Alice/Mayo; quantum algorithm claims draw abstract-idea scrutiny)
• § 102 — anticipation / novelty (a single prior-art reference)
• § 103 — obviousness (KSR combinations of references)
• § 112 — enablement, written description, definiteness (broad functional / fidelity language is the usual quantum failure mode)

A journal reviewer asks "is this good science?"; an examiner asks "are these claims patentable?". Different question, different machinery.

The panel (6 voices)

Voice	Role
1	Primary Examiner — § 101 eligibility + overall disposition
2	§ 102 Examiner — anticipation; names a single anticipatory reference
3	§ 103 Examiner — obviousness; builds primary+secondary combinations
4	§ 112 Examiner — enablement / written description / definiteness
5	Quantum Technical Specialist — operability; no-cloning / Holevo / fault-tolerance overreach
6	Supervisory Patent Examiner — synthesis + the disposition

Modes

Mode	Output	Use
full (default)	office_action.md + _office_action.json	the full 6-voice Office Action
quick	quick_examination.md	fast single-voice patentability triage

Filing standards

Pass --filing-standard to select the application-review lens:

Standard	What gets checked
uspto	35 U.S.C. § 112(a)/(b), MPEP 608 formalities, title, abstract, drawings, required sections, antecedent basis, definiteness, dependency clarity, and § 112(f) risk
epo	EPC Art. 84 clarity/support, essential features, two-part form where appropriate, claim category consistency, reference signs, description support, drawings, abstract, and EPO formalities
pct	PCT request/specification/claims/abstract/drawings completeness, clarity/support, dependency form, unity-relevant structure, sequence listings where applicable, and international-search readability
multi	Runs USPTO, EPO, and PCT review together, separating each issue by standard

Deterministic artifact — _office_action.json

Zero extra LLM cost. Parses the panel into a USPTO Office Action shape (the patent analogue of ARC's quality_gate.json):

{
  "disposition": "non-final-rejection",
  "passes": false,
  "n_claims_examined": 20,
  "n_claims_rejected": 17,
  "rejected_claims": [1, 2, 3, ...],
  "allowed_claims": [14, 19],
  "rejections_by_statute": {"102": [1, 8], "103": [2, 3, 4, ...], "112": [11]},
  "votes": { "Primary Examiner": {"disposition": "...", "confidence": 8.0}, ... }
}

Claim numbers are parsed mechanically from each examiner's per-claim rejection table; the disposition comes from the SPE synthesis. An allowance disposition with any rejected claim is auto-downgraded — the gate cannot certify a patent that still has open rejections. passes is true only for allowance.

Usage

# Direct, against a live Google Patents URL:
bash skills/patent_reviewer/run.sh --mode full \
  --patent "https://patents.google.com/patent/US10614371B2/en" \
  --outdir runs/patent_demo --llm claude

# By publication number:
bash skills/patent_reviewer/run.sh --mode full --patent US10614371B2 --outdir out

# Claims + full-application compliance under USPTO/EPO/PCT:
bash skills/patent_reviewer/run.sh --mode full \
  --patent application.html \
  --filing-standard multi \
  --outdir out

# Quick triage:
bash skills/patent_reviewer/run.sh --mode quick --patent US10614371B2 --outdir out

Granted patents (B1/B2 kind codes) are handled too — the panel reframes itself as a post-grant validity / IPR-style review rather than an examining-office Office Action (see patent_io.Patent.status_line).

Files

• skill.py — driver: ingest patent → render examiner prompt → call LLM → panel-coverage header + _office_action.json
• prompts/full.md — the 6-voice Office Action panel prompt
• prompts/quick.md — single-voice triage prompt
• ../common/patent_io.py — SSOT patent ingestion (URL / number / file → text)

In the chain

Wired as the patent-audit pipeline in chain/pipelines.py (bash chain/run.sh --pipeline patent-audit --patent URL): prior-art research → this examiner panel → logical-fallacy scan → CQE summary.

For named §102/§103 grounding, run skills/patent_prior_art/run.sh first and pass its prior_art_refs.json to this skill with --prior-art. Without an explicit prior-art set the prompt forbids the panel from inventing named references from memory.