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

flowleap-patstat

Skill1 starsChanged 4 months agoMITView on GitHub

Portfolio Analytics AND guarded SQL over the PATSTAT snapshot — structured-criteria aggregates by applicant/CPC/office/year/family/grant status, priority and continuation chains, decoded legal events, concept-to-CPC text discovery, INPADOC extended-family coverage, every number carrying a Data Edition citation. Trigger when an agent needs a named applicant's filing portfolio, structured-criteria corpus counts, what descended from a filing, which codes a legal concept spells out to at an office, which CPC codes a concept actually uses, where an extended family reaches, any other PATSTAT aggregate, or any number that must carry a PATSTAT edition citation.

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 patstat command runs one PATSTAT tool on the Tools facade.

    line 12
  2. Very specific:

    Read the index first, then --view for each view you will query.

    line 120
  3. Very specific:

    Every other patstat_sql_* error: rewrite the SQL once from the message and details, then send it with --retry-of <error code>.

    line 132

≈ “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 (Portfolio 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.

Every patstat command runs one PATSTAT tool on the Tools facade. An agent that also has flowleap mcp sees the same tool under the same name, so the command and the tool are one capability. PATSTAT tools need no patent-data key; each one publishes its own gate and rate limit in the registry (flowleap --json tools describe <tool>).

CommandToolGate
patstat portfolio <applicant> [--from-year Y] [--to-year Y] [--offices top|all]patstat_portfolioplan
patstat query "<SQL>" --question "<question>" [--retry-of <code>]patstat_queryplan, 10/min
patstat docs with one of --section S, --workflow W, --endpoint E, --compactpatstat_docssign-in
patstat graph <verb> …patstat_resolve, patstat_cpc, patstat_patent, patstat_applicant, patstat_technology, patstat_neighborhood, patstat_path, patstat_explainsign-in; see flowleap-patstat-graph

--json prints the tool data verbatim. There is no top-level success; every result carries data_edition and attribution. Through tools run the inputs are snake_case (from_year, to_year, retry_of). patstat portfolio sends offices: "all" for every office in by_year_office; the tool itself defaults to top (the 8 largest offices plus one OTHER row per year), and by_year_office_scope.truncated says when the matrix is cut. The tool caps applicant.other_matches at 10; applicant.other_matches_total is the full count:

flowleap --json tools run patstat_portfolio applicant="<applicant name>"

Which engine?

FlowLeap has three analytics engines, split by criteria shape, not by metric:

The question's essential criterionEngineGo to
A trend or landscape over free-text keywords, and the keywords are the whole criterionTopic Analyticsflowleap analytics
Structured criteria (a named applicant, CPC/IPC class, office, year, family, grant status) giving a table of countsPortfolio Analyticsthis skill
A named node and the relationships around it (who cites it, the path between two patents, co-applicants)Graph Analyticsflowleap-patstat-graph

A concept that must become identifiers, CPC codes or a candidate set for a structured aggregate comes here too, through text discovery (see the recipes below): the text is the way in, never the answer. One known document is none of the three engines: use flowleap-patent, flowleap-uspto or flowleap-ops.

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 served statement of the rule is step 1 of every served workflow (flowleap patstat docs --workflow <portfolio-analysis|guarded-sql|graph>).

Keyless, but not a stand-in. PATSTAT stays live when EPO OPS or USPTO ODP answers provider_keys_required. You may offer it to keep work moving, framed for what it is: aggregate counts from a twice-yearly snapshot, not documents and not current. It never answers "what prior art exists for this claim", and a PATSTAT table never closes a missing-key gap in a prior-art, FTO or invalidity deliverable. See flowleap-keys.

patstat portfolio and graph applicant draw entity boundaries differently: portfolio groups by name-prefix aliases, graph applicant takes one harmonized psn_id. They may disagree about where one company ends and another begins, so always say which one produced a number.

Data Edition and attribution

PATSTAT is published in snapshot editions, about twice a year. Quote the data_edition with every number from this skill, and compare two numbers only within the same edition. Keep the attribution line with any table you hand on.

Portfolio

flowleap --json patstat portfolio "Siemens AG" --from-year 2015 --to-year 2023

The result opens with a quotable summary line: relay it verbatim before any narrative. The aggregate tables by year, office and grant status follow. The served procedure, with the caveats to surface, is flowleap patstat docs --workflow portfolio-analysis.

Ambiguous applicant. An applicant name that matches several distinct entities answers 422 patstat_applicant_ambiguous; in --json the candidates are at error.details.candidates. This is an interaction step: show every candidate to the user and let the user pick. Re-run with the exact candidate name and pin that string. A caller that repeats the query (for example a recipe-custom-dashboard script) hard-codes the resolved name as a constant, so the user picks once.

Guarded SQL — aggregates beyond the typed commands

Grant rates, citation-impact rankings, inventor analytics, jurisdiction coverage and other aggregates that no typed command answers take one SQL SELECT against the flowleap.* semantic views.

The procedure is served, and it is the source of truth: flowleap patstat docs --workflow guarded-sql. Its steps map to these commands:

Served stepCommand
Verified examples first; reuse a match, or use its promoted_to commandflowleap patstat docs --section examples
Semantic model and interpretation conventions, applied as servedflowleap patstat docs --section semantic-model --part index, then flowleap patstat docs --section semantic-model --view <name> for each view
One SELECT, with the user's question verbatimflowleap patstat query "<SQL>" --question "<question>"
The one retrythe same command plus --retry-of <error code>
flowleap patstat docs --section examples
flowleap patstat docs --section semantic-model --part index
flowleap patstat docs --section semantic-model --view applications
flowleap patstat query "SELECT office, COUNT(DISTINCT family_id) AS inventions FROM flowleap.applications a JOIN flowleap.applicants ap ON ap.application_id = a.application_id WHERE UPPER(ap.name) LIKE 'SIEMENS%' GROUP BY office ORDER BY inventions DESC" --question "where does Siemens hold the most inventions?"

Read the index first, then --view for each view you will query. Read them from the served docs each time, never from memory; this skill does not restate them. The full model (--section semantic-model alone, the YAML at .yaml in --json) is too large for most tool-output limits. A LIMIT is not necessary: past the row cap the backend answers an error, never a truncated table.

You own the retry. The CLI sends patstat query exactly once. It does not resend on a 5xx, a timeout or a connection failure, so every typed error reaches you with error.code, error.message and error.details:

  • patstat_sql_timeout: a cold cache fails like a heavy query. Send the same SQL once more with --retry-of patstat_sql_timeout.
  • Every other patstat_sql_* error: rewrite the SQL once from the message and details, then send it with --retry-of <error code>.
  • patstat_busy or patstat_unreachable: capacity, not a gate verdict (backend ADR 0010). Wait error.details.retry_after seconds (the Retry-After header), then resend the same SQL with --retry-of patstat_busy. This resend does not count as the one retry.
  • A second gate failure after your one rewrite: stop and report the typed error.

Entity disambiguation in guarded SQL has no 422. Probe the candidates first with a cheap SELECT name … LIKE 'X%' GROUP BY name query. When the candidates diverge, let the user pick, as in the portfolio flow.

PATSTAT recipes — chains, legal events, text discovery, INPADOC coverage

Four view families the typed commands do not reach. The worked recipes live in references/patstat-recipes.md, with SQL byte-identical to the backend's verified queries and every measured number, cost and anchor. Read it before writing chain, legal-event, text or extended-family SQL of your own; patstat docs --section examples serves the same SQL with its live status. Four rules decide whether the number is right:

  • Chains (flowleap.priorities, flowleap.continuations) — count over the DOCDB family, walk the chain for descent. Descent has no index, so bound the hops and dedupe on application_id; match on the decoded link_kind, never the raw link_type.
  • Legal events (flowleap.legal_events, flowleap.legal_event_codes) — opposition is not one category. Enumerate the codes from flowleap.legal_event_codes for the office and the concept first, count DISTINCT applications rather than event rows, anchor on a bounded application set, and send the user to the get_legal_status tool because these events are as-of-edition.
  • Text discovery (flowleap.application_texts) — discovery returns identifiers, never document text as the answer. The match idiom is exact: to_tsvector('english', title) @@ plainto_tsquery('english', $q) AND title_lang = 'en'. websearch_to_tsquery may replace plainto_tsquery for phrases, -excluded words and or; read its two traps in the recipes reference first. EXPLAIN does not bound a GIN seed, so bound it yourself with ipr_type = 'PI', a year floor and/or an office. This lookup outranks the app's own CPC reference tables, which are the last-resort fallback for when PATSTAT is unavailable.
  • INPADOC coverage (flowleap.inpadoc_family_members) — count over DOCDB, cover over INPADOC. Counting inventions over the extended family inflates the answer; keep COUNT(DISTINCT family_id) over flowleap.family_members as the invention count and name which family you used.

patstat_unavailable

A backend with no PATSTAT database configured answers patstat_unavailable. Report it plainly ("backend has no PATSTAT dataset configured") and stop. It is a deployment gap, not a transient failure.

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.