flowleap-ai Recipe: Office Action Response
Turn an office action (OA) into a structured, evidence-backed draft response.
Step 1: Intake — Read the Office Action
For a US application, fetch the OA straight from the USPTO file wrapper — no local PDF needed. List the office actions, then pull the text (OCR'd server-side):
flowleap --json uspto documents <application-number> --code CTNF # non-final rejections
flowleap --json uspto documents <application-number> --code CTFR # final rejections
flowleap uspto document-text <application-number> <documentIdentifier> > office-action.md
If the OA was supplied as a local file instead (or is not a US application):
flowleap ocr ./office-action.pdf > office-action.md
From the extracted text, record:
- Application number and examiner
- Response deadline
- Each rejection: statutory ground (e.g. 35 U.S.C. 102/103, EPC Art. 54/56), the claims it covers, and the references it relies on
- Any objections (formalities, claim clarity) separate from rejections
Step 2: Pull Every Cited Reference
For each cited patent reference:
flowleap --json summary <cited-patent> # biblio + legal status + family
flowleap --json ops claims <cited-patent>
flowleap --json ops abstract <cited-patent>
Verify the reference's effective date actually predates your priority date — date-disqualified references are a complete answer to a rejection.
Step 3: Examiner Citation Context
flowleap --json citation search <application-number> --examiner-cited-only
flowleap --json citation novelty <application-number> # X-category refs
flowleap --json citation stats <application-number>
This shows how the examiner has used each reference (X = novelty-destroying, Y = obviousness combination, A = background) across the prosecution.
Step 4: Decompose the Rejected Claims
The as-rejected claim text is usually AMENDED and exists only in the file wrapper — not in the published application. Pull the claims document (code CLM) dated nearest before the OA (the OA's "Responsive to communication(s) filed on " line names the filing):
flowleap --json uspto documents <application-number> --code CLM
flowleap uspto document-text <application-number> <documentIdentifier> > claims-as-rejected.md
Decompose each rejected independent claim yourself: split on the
comprising transition and the semicolon/wherein boundaries, keep the claim
language verbatim per element (the method is in recipe-claim-analysis,
Step 4).
Build a mapping table per rejection: claim element → where the examiner says the reference discloses it → what the reference actually says (quote it). The argument lives where the mapping breaks.
Step 5: Ground the Arguments in Law
# US obviousness framework
flowleap --json legal search "obviousness motivation to combine rationale" --jurisdiction uspto --comprehensive
# EPO inventive step framework
flowleap --json legal search "inventive step problem solution approach" --jurisdiction epo --comprehensive
# Claim amendments support
flowleap --json legal search "written description support for amendments" --jurisdiction uspto
Cite section + source_url from each result next to the argument it backs.
Step 6: Draft the Response Shell
Per rejection:
- Restate the rejection (ground, claims, references)
- Element-mapping table with the disputed elements highlighted
- Argument: missing element / improper combination / date disqualification, with reference quotes and legal citations
- Amendment option: narrowing language sourced from the description, with a support citation
Output
- OA summary (grounds, claims, references, deadline)
- Per-reference data package with date verification
- Element-mapping tables per rejection
- Argument shells with MPEP/EPO guideline citations
- Amendment options with written-description support