agentleFS
flowleap-aiflowleap-ai/flowleap-cli/skills/flowleap-patstat-graph/SKILL.md

flowleap-patstat-graph

Skill1 starsChanged 4 months agoMITView on GitHub

Graph Analytics over the PATSTAT snapshot — a named node and the relationships around it. Worldwide DOCDB citation networks (who cites a patent, examiner vs applicant origin), citation/family paths between two patents, family coverage, an applicant's co-applicant network with top CPC and jurisdictions, the CPC codes that cover a technology keyword (`graph cpc`), and one CPC area's landscape (`graph technology`: top applicants, filing trend, grant rate, new entrants), every edge carrying a confidence tag (EXTRACTED/INFERRED/AMBIGUOUS) and a PATSTAT row provenance ref. Trigger when an agent needs a traversal answer about a specific patent or applicant — 'who cites EP3477840', 'how are patent X and patent Y connected', 'why does this patent matter', 'who does this company file with', where a family has coverage — or asks 'which CPC codes cover X', 'who dominates <technology area>', or wants the technology landscape for a CPC area; as opposed to other corpus aggregate counts (flowleap-patstat), free-text keyword analytics (flowleap analytics), or document retrieval (flowleap-patent/flowleap-ops).

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Every section header prints even when empty, showing (none recorded) — an absent section means no data, never a parsing slip.

    line 106
  2. Very specific:

    Which number is which: Every node the backend answers with carries a citable publication — the first grant where one exists, else the …

    line 124
  3. Very specific:

    Every edge carries confidence:

    line 160

≈ “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

FlowLeap Patstat Graph (Graph Analytics)

Auth and global flags: see flowleap-shared.

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.

Eight native commands under flowleap patstat graph. Each one runs one PATSTAT tool on the Tools facade, the same tool flowleap mcp serves under the same name. The graph tools need sign-in only: no plan and no patent-data key.

Routing: which engine answers this?

The routing table between Topic, Portfolio and Graph Analytics lives in flowleap-patstat ("Which engine?"). The served statement is step 1 of every served workflow (flowleap patstat docs --workflow <portfolio-analysis|guarded-sql|graph>); --workflow graph is the served procedure for this skill. A connection (who cites what, what links two patents, who co-files with whom) is here. One known document's text, claims or legal status is not: use flowleap-patent, flowleap-ops or flowleap-uspto.

A CPC/IPC-class landscape is Portfolio Analytics. flowleap patstat graph technology <cpc> is its fast path: find the code with graph cpc first. Use patstat query when the composite does not answer.

The eight commands

flowleap patstat graph resolve EP3477840
flowleap patstat graph cpc "solid electrolyte"
flowleap patstat graph patent EP3477840
flowleap patstat graph applicant 98765
flowleap patstat graph technology H01M10/0562
flowleap patstat graph neighborhood pat:56123456 --depth 2 --edge-types cites,cited_by
flowleap patstat graph path EP3477840 US5960411 --max-hops 3
flowleap patstat graph explain EP3477840 --token-budget 4000
CommandToolGateAnswers
resolve <query>patstat_resolvesign-in, 30/min sharedNumber → its pat:<appln_id> anchor; free text → ranked applicant entities with psn_id, largest portfolio first
cpc <keyword>patstat_cpcsign-in, 30/min sharedTechnology keyword → ranked CPC symbols with scheme title and application count
patent <number>patstat_patentsign-in, 30/min sharedThe whole patent picture in one call: anchor, backward/forward citations, family, applicants/inventors/CPC, priorities
applicant <psn_id>patstat_applicantsign-in, 30/min sharedOne harmonized entity: filings by year, top CPC, jurisdictions, co-applicants
technology <cpc>patstat_technologysign-in, 30/min sharedPortfolio-shaped composite, served by the graph engine. One CPC area: top applicants, filing trend, grant rate by office, new entrants, seminal families, top inventors, geography
neighborhood <node>patstat_neighborhoodsign-in, 30/min sharedBounded 1–2 hop expansion, examiner citations ranked first
path <a> <b>patstat_pathsign-in, 30/min sharedShortest citation/family path between two patents
explain <node>patstat_explainsign-in, 30/min sharedNode card + top connections, the remainder grouped with TRUE counts

Node ids are pat:<appln_id>, person:<psn_id>, family:<docdb_family_id>, cpc:<symbol>. Through tools run the inputs are snake_case: q, number, psn_id, cpc, depth, edge_types (an array), max_hops, token_budget.

Start with resolve, or with cpc

A human input becomes a node id through an entry ramp, and the graph verbs refuse rather than guess:

  • A number or a company name goes through resolve. applicant takes a strict numeric psn_id, which only resolve <name> produces.
  • A technology word goes through cpc, and one symbol from its list goes to technology. Take CPC codes from cpc, never from memory. When nothing matches, cpc prints one line and exits 0: try a shorter or different keyword.

resolve on a company name is a pick-one list, not an answer: present the ranked candidates and let the user choose. It exits 0.

An ambiguous publication number has one answer per verb family, and none exits 0, so a script can never mistake a pick-one prompt for a resolved anchor:

  • resolve answers 200 with kind: "ambiguous" and its candidates, and exits 1.
  • patent answers 422 patstat_patent_ambiguous, with the candidates at error.details.candidates.
  • neighborhood, path and explain refuse with 400 patstat_invalid_request, whose message names the candidates in prose.

In each case, pass the pat: id the user meant.

Reading output

Human mode, neighborhood / path / explain — the command prints the backend's line-per-fact text verbatim:

EP3477840 --cites [EXTRACTED 1.0]--> DE4302443 at=tls212:530028653

Quote those lines with their confidence tag and at= ref rather than re-deriving facts. The Data Edition is in the header, and TRUNCATED notices carry verb-specific narrowing hints — never present a truncated listing as complete.

Human mode, patent / applicant — section tables, in a fixed order. graph patent: Anchor → Applicants → Inventors → CPC Classifications → Backward Citations (Patents → Non-Patent Literature → Unresolved) → Forward Citations → DOCDB Family → Priority Claims → data-quality flags → edition and attribution footer. graph applicant: entity card → Filings by Year → Top CPC → Jurisdictions → Co-Applicants → footer.

Every section header prints even when empty, showing (none recorded) — an absent section means no data, never a parsing slip. Truncation reads literally Showing {shown} of {total} {label}. — repeat the TRUE total when you quote the list. Filings by Year is uncapped; do not imply a cap there.

Human mode, cpc / technology — cpc prints one line per candidate (symbol, scheme title, application count). technology prints the area card, then Top Applicants → Filing Trend → Grant Rate by Office → New Entrants → Seminal Families → Top Inventors → the three geography rankings → notes and data-quality flags → footer. Its notes say the most recent filing years are incomplete: never read that tail as a decline.

--json, every command — the tool data verbatim, with no CLI envelope and no success flag. neighborhood, path and explain answer { text, data, data_edition, attribution }; the other verbs answer their own fields with data_edition and attribution beside them. A typed error carries an error object with code, message and details.

Which number is which. Every node the backend answers with carries a citable publication — the first grant where one exists, else the earliest publication (may be null for an application with no publication at all). application / prior_application are deprecated: they carry DOCDB's own application-number format, which for a US application is a 6-digit serial plus the 2-digit filing year (US10374408 (A) decodes to USPTO application 12/103,744 — it is not itself a lookupable number). EP happens to be the one office where DOCDB's format equals the real application number, which is why this stayed hidden until a US-bearing family was reported (flowleap-backend #419). Prefer publication / prior_publication; read docdb_application / prior_docdb_application only when you need the raw DOCDB string and know to label it as such. In --json all four keys ride on every node — this is additive, not a breaking change. Human-mode text already applies this rule: it prints the citable publication, falling back to a DOCDB appln …-labeled string only when no publication exists.

Budgets and bounds

  • --token-budget (default 2000) trims the text serialization only; --json data stays complete. Out-of-range values are clamped silently into 100–20000 — never an error, so a huge budget simply gives you the maximum.
  • --depth (1 or 2) and --max-hops (1–4) are refused with a relayed patstat_invalid_request message stating the valid range. Read the message rather than guessing.
  • --edge-types takes a comma-separated subset of cites, cited_by, in_family, has_applicant, has_inventor, classified_as, claims_priority. Narrow with it instead of accepting a capped listing.

path reporting found: false is a 200 and exit 0 — the search ran and there is no path within the hop limit. That is an answer, not a failure, and it carries its own caveat: absence is not proof of unrelatedness. Unrelated technology areas commonly have no path; raise --max-hops only if the connection actually matters.

Confidence discipline

Every edge carries confidence:

  • EXTRACTED (1.0) — a direct PATSTAT row. State as fact.
  • INFERRED (0.75–0.85) — a derived join (harmonized-name grouping, extended family). Hedge it: "grouped under the harmonized entity…", never bare fact.
  • AMBIGUOUS (≤0.3) — unresolved citations kept as ghost doc: nodes. Flag, don't omit — and never build conclusions on them.

Provenance at=<table>:<key> points at the PATSTAT row asserting the relationship; carry it when the user needs to verify a claim.

Errors

CodeMeaningWhat to do
patstat_invalid_request (400)Bad node, ambiguous input, or out-of-range boundRead the relayed message — it states the fix. Resolve first if ambiguous.
patstat_patent_ambiguous (422)graph patent only: a number matching several applications, with the candidates at error.details.candidatesRender the candidates and let the user pick.
patstat_patent_not_found / patstat_entity_not_found (404)Nothing matches in the loaded editionCheck the number, or the input may postdate the snapshot.
patstat_unavailable (503)No PATSTAT dataset configured on this deploymentReport plainly; do not retry-loop.

Exit codes follow the CLI-wide table. A 3 means the credential is missing or rejected — a human must run flowleap auth login; see flowleap-auth for the verification states.

Snapshot honesty

PATSTAT is a named snapshot, not live data. Every result names its Data Edition: carry it with any number you quote, compare numbers only within the same edition, and keep the attribution line with any table you hand on. For current legal status (in force, lapsed, opposed) the snapshot is the wrong source: use the live document tools (flowleap ops legal, flowleap-uspto).

Two boundaries worth stating to users:

  • graph applicant vs patstat portfolio draw entity boundaries differently. applicant is one harmonized psn_id; portfolio groups by name-prefix aliases. They may disagree about where one company ends and another begins — say which one a number came from.
  • This engine vs flowleap-citation are different citation universes. Here: the worldwide DOCDB citation network from the PATSTAT snapshot, with examiner-vs-applicant origin. There: USPTO office-action enriched citations — US only, with X/Y/A relevance categories. Neither is a superset; pick by whether the question is about worldwide structure or US examiner reasoning.

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.