agentleFS
raja21068raja21068/AutoResearch/skills/aris/patent-review/SKILL.md

patent-review

Skill2 starsChanged 6 months agoMITView on GitHub

Get an external patent examiner review of a patent application. Use when user says \"专利审查\", \"patent review\", \"审查意见\", \"examiner review\", or wants critical feedback on patent claims and specification.

We split this skill into the rules it gives an agent

12 rules. About 0 are common, 10 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈10

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before calling the external reviewer, compile a comprehensive briefing:

    line 39
  2. Distinctive:

    Read all claims (independent + dependent)

    line 40
  3. Distinctive:

    Read specification sections (at least summary and detailed description)

    line 41
  4. Distinctive:

    Read prior art report for context

    line 42
  5. Distinctive:

    Must be fixed before proceeding

    line 114
  6. Distinctive:

    Should be fixed or argued

    line 118
  7. Distinctive:

    Consider claim amendments or specification additions

    line 119
  8. Distinctive:

    Document in output for later cleanup

    line 123
  9. Very specific:

    Use mcp__codex__codex with the threadId from Round 1:

    line 131
  10. Very specific:

    Always use model_reasoning_effort: "xhigh" for maximum analysis depth.

    line 197
  11. Distinctive:

    Document all changes in the review report for traceability.

    line 199
  12. Distinctive:

    If the patentability score is below 5/10 after Round 2, recommend significant rework before filing.

    line 200

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Patent Examiner Review via Codex MCP (xhigh reasoning)

Get a multi-round patent examiner review of the patent application based on: $ARGUMENTS

Adapted from /research-review. The reviewer persona is a patent examiner, not a paper reviewer.

Constants

  • REVIEWER_MODEL = gpt-5.4 — Model used via Codex MCP
  • REVIEW_ROUNDS = 2 — Number of review rounds
  • EXAMINER_PERSONA = "patent-examiner" — GPT-5.4 persona

Prerequisites

  • Codex MCP Server configured:
    claude mcp add codex -s user -- codex mcp-server

Inputs

  1. patent/CLAIMS.md — all drafted claims
  2. patent/specification/ — all specification sections
  3. patent/figures/numeral_index.md — reference numeral mapping
  4. patent/PRIOR_ART_REPORT.md — known prior art
  5. patent/INVENTION_DISCLOSURE.md — invention structure

Workflow

Step 1: Gather Patent Context

Before calling the external reviewer, compile a comprehensive briefing:

  1. Read all claims (independent + dependent)
  2. Read specification sections (at least summary and detailed description)
  3. Read prior art report for context
  4. Identify: core inventive concept, claim scope, known prior art, target jurisdiction

Step 2: Round 1 — Full Examiner Review

Send to REVIEWER_MODEL via mcp__codex__codex with xhigh reasoning:

mcp__codex__codex:
  config: {"model_reasoning_effort": "xhigh"}
  prompt: |
    You are a senior patent examiner at the [USPTO/CNIPA/EPO].
    Examine this patent application and issue a detailed office action.

    CLAIMS:
    [all claims]

    SPECIFICATION SUMMARY:
    [key sections: title, technical field, background, summary, abstract]

    PRIOR ART KNOWN:
    [prior art references]

    PATENTABILITY STANDARDS TO APPLY:
    [US: 35 USC 101/102/103/112 | CN: Articles 22, 26 | EP: Articles 54, 56, 83, 84]

    Please issue an office action covering:

    1. CLAIM CLARITY (112(b)/Art 84):
       - Are all terms definite?
       - Any indefinite functional language?
       - Antecedent basis issues?

    2. WRITTEN DESCRIPTION (112(a)/Art 83 first para):
       - Does the spec support ALL claim scope?
       - Any claim elements without spec support?

    3. ENABLEMENT (112(a)/Art 83):
       - Can a POSITA practice the invention?
       - Any missing algorithm/structure for functional claims?

    4. NOVELTY (102/Art 54):
       - Would any known reference anticipate any claim?
       - Identify the closest single reference.

    5. NON-OBVIOUSNESS (103/Art 56):
       - Would any combination render claims obvious?
       - What is the motivation to combine?

    6. CLAIM SCOPE:
       - Are independent claims broad enough to be commercially valuable?
       - Do dependent claims provide meaningful fallback positions?
       - Any claims that are too broad (likely rejected) or too narrow (not valuable)?

    7. SPECIFICATION QUALITY:
       - Language issues (subjective terms, relative terms, result-to-be-achieved)
       - Reference numeral consistency
       - Missing embodiments

    Format your response as a formal office action with:
    - GROUNDS OF REJECTION for each issue (cite statute)
    - SUGGESTED AMENDMENTS for each issue
    - OVERALL PATENTABILITY SCORE: 1-10

    Be rigorous and specific. This is a real examination.

Step 3: Implement Fixes (Round 1)

Based on the examiner's office action:

  1. CRITICAL issues (102 rejection, 112 indefiniteness, missing enablement):
    • Must be fixed before proceeding
    • Amend claims or add specification support
  2. MAJOR issues (103 obviousness, weak claim scope, missing support):
    • Should be fixed or argued
    • Consider claim amendments or specification additions
  3. MINOR issues (language quality, numeral consistency, formatting):
    • Fix if time permits
    • Document in output for later cleanup

For each fix:

  • Show the specific change (old claim -> new claim)
  • Explain how the fix addresses the examiner's concern

Step 4: Round 2 — Follow-Up Review

Use mcp__codex__codex with the threadId from Round 1:

mcp__codex__codex:
  threadId: [from Round 1]
  prompt: |
    Here is the revised patent application after addressing your office action.

    CHANGES MADE:
    [list of all changes with rationale]

    REVISED CLAIMS:
    [updated claims]

    REVISED SPECIFICATION EXCERPTS:
    [changed sections]

    Please re-examine:
    1. Are the previous rejections overcome?
    2. Are there new issues introduced by the amendments?
    3. What is the updated patentability score?
    4. Any remaining grounds for rejection?

Step 5: Generate Improvement Report

Write patent/PATENT_REVIEW.md:

## Patent Review Report

### Application Summary
[Title, claims count, jurisdiction]

### Review Round 1
#### Office Action Summary
[Key findings from examiner]

#### Issues Found
| # | Type | Severity | Claim/Section | Issue | Citation | Fix Applied |
|---|------|----------|--------------|-------|----------|-------------|
| 1 | Clarity | CRITICAL | Claim 3 | Indefinite term "rapid" | 112(b) | Defined in spec |
| 2 | Novelty | MAJOR | Claim 1 | Ref X anticipates element C | 102 | Amended claim |

#### Score After Round 1: [X]/10

### Review Round 2
#### Follow-Up Assessment
[Are previous rejections overcome?]

#### Remaining Issues
[Any issues still outstanding]

#### Score After Round 2: [X]/10

### Recommendations
[Final recommendations before proceeding to jurisdiction formatting]
- [ ] All CRITICAL issues resolved
- [ ] All MAJOR issues resolved or argued
- [ ] Specification supports all claim amendments
- [ ] Ready for jurisdiction formatting

Key Rules

  • The reviewer persona must be a patent examiner, not a paper reviewer or academic.
  • Always use model_reasoning_effort: "xhigh" for maximum analysis depth.
  • Address CRITICAL and MAJOR issues before proceeding to the next phase.
  • Document all changes in the review report for traceability.
  • If the patentability score is below 5/10 after Round 2, recommend significant rework before filing.
  • The review is advisory -- actual prosecution may proceed differently.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.