agentleFS
flowleap-aiflowleap-ai/flowleap-cli/skills/recipe-find-better/SKILL.md

recipe-find-better

Skill1 starsChanged 4 months agoMITView on GitHub

Find Better recipe for a granted patent — read the code-computed Examiner Baseline (every document cited across the family, by office), run three logged expansion tracks (two-hop backward citations, inventor and author networks, classifications with Discriminating Terms), and score the examiner's best art and the art found on the same claim elements as "disclosed n of m". Trigger when the user asks whether earlier or closer prior art exists than the examiners cited, or what the examiners missed on a granted patent. Not for a pre-filing novelty search (recipe-prior-art-search) or for venue, grounds and invalidity charts (recipe-invalidity-analysis).

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    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. Very specific:

    Read documents[], gaps[] and membersWalked[] (field guide: references/baseline-json.md).

    line 29
  2. Distinctive:

    Run all three tracks. Each one is mandatory.

    line 60
  3. Distinctive:

    Keep only documents published before the critical date.

    line 85
  4. Distinctive:

    If no candidate beats the examiner's art, write "no better art found" for that claim.

    line 100
  5. Very specific:

    Write the report and the working record with the structure in references/report-template.md: one row per independent claim, the Baseline …

    line 114

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Recipe: Find Better

Given a granted patent, find prior art earlier or closer than the art the examining offices cited. Show, per independent claim, the examiner's best art beside the best art found, as element counts. "No better art found" with the full Examiner Baseline and the full tracks log is a complete result.

In a chat client with the FlowLeap connector or in the FlowLeap Patent AI app, call the tools named in the flowleap-shared tool table instead of these commands, and never run a flowleap command there. The steps stay the same.

Find Better makes no invalidity statement and gives no model-rated score. The only comparison is a count over quoted element rows.

Step 1: Baseline

flowleap --json patent examiner-baseline <granted-publication>

In a chat client or the FlowLeap Patent AI app, call the examiner_baseline tool; it returns the same JSON.

Read documents[], gaps[] and membersWalked[] (field guide: references/baseline-json.md). Record the critical date: the earliest dates.priority[].date from flowleap --json ops biblio <granted-publication>. State it as YYYY-MM-DD in the report.

The examiner's best art for an independent claim is every document with an X or Y category on a citation that reaches that claim, whatever its citedBy says (only an examiner assigns a category). Rank X before Y, then by how many offices cite it, then by how many independent claims it reaches. When no X or Y citation exists (a US-origin family often has no categories), take the documents with citedBy: "examiner" and no category, then the documents a US office action rejected claims with under 102 or 103 (source: "uspto_enriched"). The report shows that basis beside each document. Never take a document that only the applicant cited, with no category and no rejection. Each entry in gaps[] goes in the report as a gap, never as "nothing cited".

Done when every independent claim has its examiner's best art list, or the words "no X or Y citation reaches this claim", and every gap is recorded.

Step 2: Claims

Decompose each independent claim of the granted publication into elements (method: recipe-claim-analysis, Step 4). Give each element its Discriminating Terms (see flowleap-patent).

Done when each independent claim is an element list with its Discriminating Terms.

Step 3: Tracks

Run all three tracks. Each one is mandatory. Use the commands in references/tracks.md.

  1. Backward citations, two hops from every document of the examiner's best art (the X or Y documents, or, when none exists, the examiner-cited documents from Step 1).
  2. Inventor and author networks: the target's inventors, the inventors of the X/Y patents, and the authors of the X/Y non-patent documents.
  3. Classification co-occurrence: the classifications of the target and of its X/Y documents, combined with the Discriminating Terms. One query is not enough. Do the term-drop pass: run the first query again with ONE term removed at a time, and once with the classification alone when the first query gave 200 hits or fewer. Log each variant with its count, and read the titles of the top hits of each variant before the critical date. Then run one title-phrase query per Discriminating Term, ti="<two-word phrase from the claim>", because old US documents often have no abstract in OPS. When the target carries 2000-series indexing codes, also run the main group with combinations of 2-3 indexing codes it shares with the examiner's documents, with no terms. Read every result set in full (page through it with ops search --start/--end) or narrow it until you can; never judge a set from its first page. Then log the recall check: how many Baseline documents, matched by family, the Track 3 queries returned. Show the statement as a Search Statement block — see flowleap-patent references/search-statement.md.

Keep only documents published before the critical date. Log every query with its count in the working record, empty results too.

Done when each track has its log, Track 3 has its term-drop variants, its title-phrase queries, its indexing-code combinations and its recall check, and each kept candidate has its claims or full text pulled.

Step 4: Compare

For each independent claim, score the top examiner's art and each candidate on the SAME element rows from Step 2. A row counts as disclosed only with a quoted passage and its anchor (claim, paragraph, column and line, or figure). Write "disclosed n of m" for each document.

"Better" means more elements disclosed, or the same count with an earlier date. If no candidate beats the examiner's art, write "no better art found" for that claim. That is a complete result, but only when all three tracks ran at least one query and Track 1 has a hop 2. Otherwise write "Search incomplete for claim N: <reason>" (for example "Track 2 ran no query" or "Track 1 has no hop 2"). Name in the result line only the examiner documents you scored. List the other examiner's best art under "named but not scored". Flag a document with a P or E category, or published after the critical date, as "published after the critical date; not prior art for this claim unless the priority claim fails".

Done when every independent claim has both columns scored on the same rows.

Step 5: Record

Write the report and the working record with the structure in references/report-template.md: one row per independent claim, the Baseline matrix, the gaps, and the tracks log. Mark each Track 1 query hop 1 or hop 2. If hop 1 found no document outside the Baseline, log one hop 2 entry that says so, with count 0.

Done when the report holds every independent claim, every gap and every track log.

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.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

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.