科研 Skill

知识图谱与SPARQL检索

knowledge-graph-sparql

Builds knowledge graphs and runs SPARQL queries over RDF and Wikidata for entity linking in research and library data. Use when modeling research entities as a linked graph.

发布者 Codar安全审核通过v0.1.0更新于 2026-11-18

SKILL.md

Knowledge Graph and SPARQL Analysis

When to Use This Skill

Use this skill when you need to:

  • Build RDF knowledge graphs from structured or tabular data
  • Query Wikidata, DBpedia, or local triple stores with SPARQL
  • Perform entity linking and disambiguation using identifiers
  • Construct ontologies and apply reasoning rules
  • Visualize knowledge graph topology and relationship patterns
  • Integrate heterogeneous data sources through linked data principles
  • Answer complex multi-hop queries over scholarly metadata

Trigger keywords: SPARQL, RDF, knowledge graph, triple store, Wikidata, DBpedia, ontology, OWL, entity linking, linked data, URI, predicate, subject, object, triples, CONSTRUCT query, federated query, LOD, schema.org, FOAF, Dublin Core, library catalog, authority file.

Background & Key Concepts

RDF Data Model

The Resource Description Framework (RDF) represents information as triples:

$$\langle \text{subject},\ \text{predicate},\ \text{object} \rangle$$

For example: <Person:Einstein> <property:birthPlace> <Place:Ulm>. Nodes are either URIs (resources) or literals (strings, numbers). Collections of triples form a directed labeled graph.

SPARQL Query Patterns

A basic SPARQL SELECT query:

SELECT ?author ?title
WHERE {
  ?paper a schema:ScholarlyArticle ;
         schema:author ?author ;
         schema:name ?title .
  FILTER(?year > 2020)
}
LIMIT 100

Key query types: SELECT (tabular), CONSTRUCT (new graph), ASK (boolean), DESCRIBE (entity description).

Graph Patterns

  • Optional patterns: OPTIONAL { ?s :p ?o } — left outer join
  • Filters: FILTER(LANG(?label) = "en")
  • Aggregates: GROUP BY, COUNT, SUM, AVG
  • Property paths: ?a :knows+ ?b — transitive closure

OWL Reasoning

OWL (Web Ontology Language) adds axioms: owl:subClassOf, owl:equivalentClass, owl:inverseOf. A reasoner can infer new triples: if A subClassOf B and x type A, then x type B.

Entity Linking

Map text mentions to canonical URIs using:

  1. String similarity matching
  2. Context disambiguation (co-occurring entities)
  3. Lookup in authority files (VIAF, GND, ORCID, ROR)

Environment Setup

pip install rdflib>=6.3 requests>=2.31 pandas>=2.0 networkx>=3.2 \
            matplotlib>=3.7 numpy>=1.24
from rdflib import Graph, Namespace, URIRef, Literal, BNode
from rdflib.namespace import RDF, RDFS, OWL, XSD, FOAF, DCTERMS
import requests
import pandas as pd
print("Knowledge graph environment ready")

Core Workflow

Step 1: Build an RDF Knowledge Graph with rdflib

from rdflib import Graph, Namespace, URIRef, Literal, BNode
from rdflib.namespace import RDF, RDFS, OWL, XSD, FOAF, DCTERMS
import pandas as pd
import numpy as np

# -----------------------------------------------------------------
# Define namespaces
# -----------------------------------------------------------------
EX  = Namespace("http://example.org/research/")
SCHEMA = Namespace("https://schema.org/")
BIBO   = Namespace("http://purl.org/ontology/bibo/")

# -----------------------------------------------------------------
# Create a graph with sample scholarly data
# -----------------------------------------------------------------
g = Graph()
g.bind("ex",     EX)
g.bind("schema", SCHEMA)
g.bind("foaf",   FOAF)
g.bind("bibo",   BIBO)
g.bind("dcterms",DCTERMS)

# --- Authors ---
authors_data = [
    ("A001", "Alice Zhang",    "Tsinghua University", "0000-0001-2345-6789"),
    ("A002", "Bob Smith",      "MIT",                 "0000-0002-3456-7890"),
    ("A003", "Carol Johnson",  "Oxford University",   "0000-0003-4567-8901"),
    ("A004", "David Li",       "Tsinghua University", "0000-0004-5678-9012"),
]

for aid, name, affil, orcid in authors_data:
    author_uri = EX[f"author/{aid}"]
    inst_uri   = EX[f"institution/{affil.replace(' ', '_')}"]

    g.add((author_uri, RDF.type,    FOAF.Person))
    g.add((author_uri, FOAF.name,   Literal(name)))
    g.add((author_uri, SCHEMA.affiliation, inst_uri))
    g.add((author_uri, SCHEMA.identifier,
           Literal(f"https://orcid.org/{orcid}", datatype=XSD.anyURI)))
    g.add((inst_uri,   RDF.type,    SCHEMA.Organization))
    g.add((inst_uri,   SCHEMA.name, Literal(affil)))

# --- Papers ---
papers_data = [
    ("P001", "Deep Learning for Climate", 2022, ["A001", "A002"], ["C01", "C02"]),
    ("P002", "Transformer Models Survey", 2021, ["A002", "A003"], ["C02", "C03"]),
    ("P003", "Graph Neural Networks",     2023, ["A001", "A004"], ["C01", "C04"]),
    ("P004", "Federated Learning Privacy",2022, ["A003", "A004"], ["C03", "C05"]),
]
concepts_data = {
    "C01": "Deep Learning", "C02": "Natural Language Processing",
    "C03": "Privacy",       "C04": "Graph Theory",
    "C05": "Federated Learning",
}

for pid, title, year, author_ids, concept_ids in papers_data:
    paper_uri = EX[f"paper/{pid}"]
    g.add((paper_uri, RDF.type,           BIBO.AcademicArticle))
    g.add((paper_uri, DCTERMS.title,      Literal(title)))
    g.add((paper_uri, DCTERMS.date,
           Literal(str(year), datatype=XSD.gYear)))
    for aid in author_ids:
        g.add((paper_uri, DCTERMS.creator, EX[f"author/{aid}"]))
    for cid in concept_ids:
        concept_uri = EX[f"concept/{cid}"]
        g.add((paper_uri, SCHEMA.about, concept_uri))
        g.add((concept_uri, RDF.type,    SCHEMA.DefinedTerm))
        g.add((concept_uri, SCHEMA.name, Literal(concepts_data[cid])))

# --- Citation relationships ---
citations = [("P002", "P001"), ("P003", "P001"), ("P003", "P002"), ("P004", "P002")]
for citing, cited in citations:
    g.add((EX[f"paper/{citing}"], BIBO.cites, EX[f"paper/{cited}"]))

print(f"Graph has {len(g)} triples")
print(f"Unique subjects: {len(set(g.subjects()))}")

# Serialize to Turtle format
turtle_str = g.serialize(format="turtle")
print("\n--- First 1000 chars of Turtle serialization ---")
print(turtle_str[:1000])

# Save graph
g.serialize("research_graph.ttl", format="turtle")
print("\nGraph saved to research_graph.ttl")

Step 2: SPARQL Queries Over the Knowledge Graph

from rdflib import Graph
import pandas as pd

# Reload the graph
g = Graph()
g.parse("research_graph.ttl", format="turtle")

# -----------------------------------------------------------------
# Query 1: All papers with author names and years
# -----------------------------------------------------------------
q1 = """
PREFIX dcterms: <http://purl.org/dc/terms/>
PREFIX foaf: <http://xmlns.com/foaf/0.1/>
PREFIX bibo: <http://purl.org/ontology/bibo/>

SELECT ?title ?author_name ?year
WHERE {
  ?paper a bibo:AcademicArticle ;
         dcterms:title ?title ;
         dcterms:date  ?year ;
         dcterms:creator ?author .
  ?author foaf:name ?author_name .
}
ORDER BY ?year ?title
"""
results1 = g.query(q1)
df1 = pd.DataFrame(results1, columns=["title", "author_name", "year"])
print("=== Query 1: Papers and Authors ===")
print(df1.to_string(index=False))

# -----------------------------------------------------------------
# Query 2: Co-author pairs
# -----------------------------------------------------------------
q2 = """
PREFIX dcterms: <http://purl.org/dc/terms/>
PREFIX foaf: <http://xmlns.com/foaf/0.1/>
PREFIX bibo: <http://purl.org/ontology/bibo/>

SELECT DISTINCT ?author1 ?author2 (COUNT(?paper) AS ?collab_count)
WHERE {
  ?paper a bibo:AcademicArticle ;
         dcterms:creator ?a1 ;
         dcterms:creator ?a2 .
  ?a1 foaf:name ?author1 .
  ?a2 foaf:name ?author2 .
  FILTER(?a1 < ?a2)
}
GROUP BY ?author1 ?author2
ORDER BY DESC(?collab_count)
"""
results2 = g.query(q2)
df2 = pd.DataFrame(results2, columns=["author1", "author2", "collaborations"])
print("\n=== Query 2: Co-author Pairs ===")
print(df2.to_string(index=False))

# -----------------------------------------------------------------
# Query 3: Citation network (who cites whom via paper author)
# -----------------------------------------------------------------
q3 = """
PREFIX dcterms: <http://purl.org/dc/terms/>
PREFIX bibo:   <http://purl.org/ontology/bibo/>
PREFIX foaf:   <http://xmlns.com/foaf/0.1/>

SELECT ?citing_paper_title ?cited_paper_title
WHERE {
  ?citing_paper bibo:cites ?cited_paper .
  ?citing_paper dcterms:title ?citing_paper_title .
  ?cited_paper  dcterms:title ?cited_paper_title .
}
"""
results3 = g.query(q3)
df3 = pd.DataFrame(results3, columns=["citing", "cited"])
print("\n=== Query 3: Citation Relationships ===")
print(df3.to_string(index=False))

# -----------------------------------------------------------------
# Query 4: Concept co-occurrence in papers
# -----------------------------------------------------------------
q4 = """
PREFIX schema:  <https://schema.org/>
PREFIX dcterms: <http://purl.org/dc/terms/>
PREFIX bibo:    <http://purl.org/ontology/bibo/>

SELECT ?c1_name ?c2_name (COUNT(?paper) AS ?co_count)
WHERE {
  ?paper a bibo:AcademicArticle ;
         schema:about ?c1 ;
         schema:about ?c2 .
  ?c1 schema:name ?c1_name .
  ?c2 schema:name ?c2_name .
  FILTER(?c1 < ?c2)
}
GROUP BY ?c1_name ?c2_name
HAVING (?co_count >= 1)
ORDER BY DESC(?co_count)
"""
results4 = g.query(q4)
df4 = pd.DataFrame(results4, columns=["concept1", "concept2", "co_count"])
print("\n=== Query 4: Concept Co-occurrence ===")
print(df4.to_string(index=False))

Step 3: Wikidata SPARQL Queries

import requests
import pandas as pd
import time
import os

WIKIDATA_ENDPOINT = "https://query.wikidata.org/sparql"

def wikidata_sparql(query, timeout=30):
    """Execute a SPARQL query against Wikidata.

    Args:
        query: SPARQL query string
        timeout: request timeout in seconds
    Returns:
        DataFrame with results, or synthetic fallback on error
    """
    headers = {
        "User-Agent": "ResearchBot/1.0 (academic research)",
        "Accept": "application/sparql-results+json",
    }
    try:
        response = requests.get(
            WIKIDATA_ENDPOINT,
            params={"query": query, "format": "json"},
            headers=headers,
            timeout=timeout
        )
        response.raise_for_status()
        data = response.json()
        vars_ = data["head"]["vars"]
        rows = []
        for binding in data["results"]["bindings"]:
            row = {v: binding.get(v, {}).get("value", "") for v in vars_}
            rows.append(row)
        return pd.DataFrame(rows), True
    except Exception as e:
        print(f"Wikidata query failed: {e}")
        return pd.DataFrame(), False

# Query: Top universities founded before 1600 with their location
query_universities = """
SELECT ?university ?universityLabel ?founded ?countryLabel
WHERE {
  ?university wdt:P31 wd:Q3918 .         # instance of: university
  ?university wdt:P571 ?founded .         # founded date
  ?university wdt:P17  ?country .         # country
  FILTER(YEAR(?founded) < 1600)
  SERVICE wikibase:label { bd:serviceParam wikibase:language "en" }
}
ORDER BY ?founded
LIMIT 20
"""

df_univ, success = wikidata_sparql(query_universities)

if success and len(df_univ) > 0:
    print("=== Historical Universities (from Wikidata) ===")
    print(df_univ[["universityLabel", "founded", "countryLabel"]].to_string(index=False))
else:
    # Synthetic fallback
    print("=== Historical Universities (synthetic data) ===")
    synthetic = pd.DataFrame({
        "universityLabel": ["University of Bologna", "University of Oxford",
                            "University of Cambridge", "University of Salamanca",
                            "University of Paris"],
        "founded": ["1088", "1096", "1209", "1218", "1257"],
        "countryLabel": ["Italy", "UK", "UK", "Spain", "France"],
    })
    print(synthetic.to_string(index=False))

# Query: Nobel laureates in physics with their alma mater
query_nobel = """
SELECT ?person ?personLabel ?year ?universityLabel
WHERE {
  ?person wdt:P166 wd:Q38104 .            # award: Nobel Prize in Physics
  ?person p:P166 ?statement .
  ?statement pq:P585 ?year .
  OPTIONAL { ?person wdt:P69 ?university . }
  SERVICE wikibase:label { bd:serviceParam wikibase:language "en" }
}
ORDER BY DESC(?year)
LIMIT 20
"""

df_nobel, success = wikidata_sparql(query_nobel)
if success and len(df_nobel) > 0:
    print("\n=== Recent Nobel Physics Laureates ===")
    print(df_nobel[["personLabel", "year"]].head(10).to_string(index=False))
else:
    print("\n(Wikidata offline – skipping Nobel query)")

# -----------------------------------------------------------------
# Entity lookup by ORCID
# -----------------------------------------------------------------
def lookup_researcher_wikidata(orcid_id):
    """Find Wikidata entity for a researcher by ORCID.

    Args:
        orcid_id: ORCID string (e.g. "0000-0002-1825-0097")
    Returns:
        dict with Wikidata QID and labels, or empty dict on failure
    """
    query = f"""
    SELECT ?person ?personLabel ?employerLabel
    WHERE {{
      ?person wdt:P496 "{orcid_id}" .
      OPTIONAL {{ ?person wdt:P108 ?employer . }}
      SERVICE wikibase:label {{ bd:serviceParam wikibase:language "en" }}
    }}
    """
    df, ok = wikidata_sparql(query)
    if ok and len(df) > 0:
        return df.iloc[0].to_dict()
    return {}

# Example ORCID (Tim Berners-Lee: 0000-0003-1279-3709)
researcher = lookup_researcher_wikidata("0000-0003-1279-3709")
if researcher:
    print(f"\nWikidata entity found: {researcher}")
else:
    print("\nEntity lookup: API offline or ORCID not found in Wikidata")

Advanced Usage

Federated SPARQL Query

-- Example: Link local data with Wikidata via federated query
-- (Run this in a SPARQL endpoint that supports SERVICE)

PREFIX owl:    <http://www.w3.org/2002/07/owl#>
PREFIX schema: <https://schema.org/>

SELECT ?localAuthor ?wd_label ?wd_birthdate
WHERE {
  # Local graph
  ?localAuthor a schema:Person ;
               owl:sameAs ?wikidataURI .

  # Federated to Wikidata
  SERVICE <https://query.wikidata.org/sparql> {
    ?wikidataURI wdt:P569 ?wd_birthdate .
    SERVICE wikibase:label {
      bd:serviceParam wikibase:language "en" .
      ?wikidataURI rdfs:label ?wd_label .
    }
  }
}

OWL Reasoning with rdflib + OWL-RL

from rdflib import Graph, Namespace, URIRef
from rdflib.namespace import RDF, RDFS, OWL

def apply_rdfs_closure(g):
    """Apply basic RDFS inference rules (subClassOf, subPropertyOf).

    Args:
        g: rdflib Graph
    Returns:
        g: graph with inferred triples added
    """
    changed = True
    while changed:
        changed = False
        new_triples = set()

        # Rule: subClassOf transitivity
        for s, _, sc in g.triples((None, RDFS.subClassOf, None)):
            for _, _, sc2 in g.triples((sc, RDFS.subClassOf, None)):
                t = (s, RDFS.subClassOf, sc2)
                if t not in g:
                    new_triples.add(t)

        # Rule: type propagation via subClassOf
        for s, _, type_ in g.triples((None, RDF.type, None)):
            for _, _, superclass in g.triples((type_, RDFS.subClassOf, None)):
                t = (s, RDF.type, superclass)
                if t not in g:
                    new_triples.add(t)

        for t in new_triples:
            g.add(t)
            changed = True

    return g

# Example: Add subClassOf axioms and infer types
g_owl = Graph()
EX2 = Namespace("http://example.org/")
g_owl.bind("ex", EX2)

# Ontology axioms
g_owl.add((EX2.Professor, RDFS.subClassOf, EX2.AcademicStaff))
g_owl.add((EX2.AcademicStaff, RDFS.subClassOf, EX2.Person))
g_owl.add((EX2.alice, RDF.type, EX2.Professor))

print(f"Before inference: {len(g_owl)} triples")
g_owl = apply_rdfs_closure(g_owl)
print(f"After RDFS closure: {len(g_owl)} triples")

for s, p, o in g_owl.triples((EX2.alice, RDF.type, None)):
    print(f"  alice rdf:type {o.split('/')[-1]}")

Graph Visualization with NetworkX

import networkx as nx
import matplotlib.pyplot as plt
from rdflib import Graph

def rdf_to_networkx(g, predicate_filter=None):
    """Convert rdflib Graph to NetworkX DiGraph for visualization.

    Args:
        g: rdflib.Graph
        predicate_filter: optional set of predicate URIs to include
    Returns:
        G: networkx DiGraph
    """
    G = nx.DiGraph()
    for s, p, o in g:
        s_label = str(s).split("/")[-1].split("#")[-1]
        p_label = str(p).split("/")[-1].split("#")[-1]
        o_label = str(o).split("/")[-1].split("#")[-1]

        if predicate_filter and str(p) not in predicate_filter:
            continue
        G.add_edge(s_label, o_label, predicate=p_label)
    return G

g2 = Graph()
g2.parse("research_graph.ttl", format="turtle")

from rdflib.namespace import DCTERMS
G_nx = rdf_to_networkx(g2, predicate_filter={
    str(DCTERMS.creator),
    "http://purl.org/ontology/bibo/cites",
})

print(f"Visualization graph: {G_nx.number_of_nodes()} nodes, "
      f"{G_nx.number_of_edges()} edges")

fig, ax = plt.subplots(figsize=(10, 7))
pos = nx.spring_layout(G_nx, seed=42)
edge_labels = {(u, v): d["predicate"] for u, v, d in G_nx.edges(data=True)}
nx.draw_networkx(G_nx, pos, ax=ax, node_color="lightblue",
                 node_size=1200, font_size=7, arrows=True)
nx.draw_networkx_edge_labels(G_nx, pos, edge_labels, font_size=6, ax=ax)
ax.set_title("Research Knowledge Graph")
ax.axis("off")
plt.tight_layout()
plt.savefig("knowledge_graph_viz.png", dpi=150, bbox_inches="tight")
plt.close()
print("Figure saved: knowledge_graph_viz.png")

Troubleshooting

ProblemCauseFix
SPARQL parse errorMissing prefix declarationAdd PREFIX declarations at top of query
rdflib parse failureWrong serialization formatCheck file extension; specify format="turtle" or "xml"
Wikidata timeoutComplex query or busy endpointAdd LIMIT; break into smaller queries; use OPTIONAL not required triples
Empty SPARQL resultsNamespace mismatchVerify namespace URIs match what was used when inserting data
rdflib reasoning too slowLarge graphUse owlrl package for OWL 2 RL reasoning instead
Unicode in literalsNon-ASCII charactersrdflib handles UTF-8 natively; ensure Literal() wraps the string

External Resources

Examples

Example 1: Import CSV Data into RDF Graph

import pandas as pd
from rdflib import Graph, Namespace, URIRef, Literal
from rdflib.namespace import RDF, XSD, DCTERMS, FOAF

def csv_to_rdf(df, base_uri="http://example.org/"):
    """Convert a publications CSV to RDF triples.

    Args:
        df: DataFrame with columns: id, title, year, authors, doi
        base_uri: base URI for entities
    Returns:
        rdflib Graph
    """
    EX = Namespace(base_uri)
    BIBO = Namespace("http://purl.org/ontology/bibo/")
    g = Graph()
    g.bind("ex", EX); g.bind("bibo", BIBO)
    g.bind("dcterms", DCTERMS); g.bind("foaf", FOAF)

    for _, row in df.iterrows():
        paper_uri = EX[f"paper/{row['id']}"]
        g.add((paper_uri, RDF.type,        BIBO.AcademicArticle))
        g.add((paper_uri, DCTERMS.title,   Literal(str(row["title"]))))
        if pd.notna(row.get("year")):
            g.add((paper_uri, DCTERMS.date,
                   Literal(str(int(row["year"])), datatype=XSD.gYear)))
        if pd.notna(row.get("doi")):
            g.add((paper_uri, BIBO.doi, Literal(str(row["doi"]))))

        for i, author_name in enumerate(str(row.get("authors", "")).split(";")):
            author_name = author_name.strip()
            if author_name:
                author_uri = EX[f"author/{author_name.replace(' ', '_')}"]
                g.add((author_uri, RDF.type,    FOAF.Person))
                g.add((author_uri, FOAF.name,   Literal(author_name)))
                g.add((paper_uri, DCTERMS.creator, author_uri))
    return g

import numpy as np
np.random.seed(42)
sample_df = pd.DataFrame({
    "id": [f"P{i:04d}" for i in range(10)],
    "title": [f"Research Article {i}" for i in range(10)],
    "year": np.random.randint(2018, 2024, 10),
    "authors": ["Alice; Bob", "Carol", "David; Alice", "Eve; Frank",
                "Grace", "Henry; Alice", "Iris", "Jack; Eve",
                "Kate; David", "Liam"],
    "doi": [f"10.1234/j.2023.{i:04d}" for i in range(10)],
})

g_imported = csv_to_rdf(sample_df)
print(f"Imported graph: {len(g_imported)} triples")
g_imported.serialize("imported.ttl", format="turtle")
print("Saved to imported.ttl")

Example 2: SPARQL-Based Authority File Reconciliation

from rdflib import Graph, Namespace, URIRef, Literal, OWL
from rdflib.namespace import RDF, FOAF
import pandas as pd

def reconcile_authors(g, external_authority_df, name_col="name", uri_col="wikidata_uri"):
    """Add owl:sameAs links from local author nodes to external authority URIs.

    Args:
        g: rdflib.Graph
        external_authority_df: DataFrame with name → external URI mapping
        name_col, uri_col: column names
    Returns:
        g: graph with sameAs triples added
    """
    name_to_uri = dict(zip(external_authority_df[name_col],
                           external_authority_df[uri_col]))
    added = 0
    for author, _, name in g.triples((None, FOAF.name, None)):
        name_str = str(name)
        if name_str in name_to_uri:
            ext_uri = URIRef(name_to_uri[name_str])
            g.add((author, OWL.sameAs, ext_uri))
            added += 1
    print(f"Added {added} owl:sameAs links")
    return g

# Simulate authority file
authority = pd.DataFrame({
    "name": ["Alice Zhang", "Bob Smith", "Carol Johnson"],
    "wikidata_uri": [
        "https://www.wikidata.org/entity/Q999991",
        "https://www.wikidata.org/entity/Q999992",
        "https://www.wikidata.org/entity/Q999993",
    ]
})

g3 = Graph()
g3.parse("research_graph.ttl", format="turtle")
g3 = reconcile_authors(g3, authority)
g3.serialize("reconciled.ttl", format="turtle")
print("Reconciled graph saved to reconciled.ttl")

示例提示词

我想用Wikidata构建一个研究实体的知识图谱并写SPARQL查询。

所属插件

跨学科科研方法跨学科通用科研方法:系统综述与元分析、预注册、实验与问卷设计、文献矛盾与研究空白、科学计量与专利分析、可复现研究、答辩与审稿回复。

安装该插件会一并获得同领域的 54 个 Skill / MCP。

同插件的其他科研 Skill

常见问题

知识图谱与SPARQL检索能做什么?

Builds knowledge graphs and runs SPARQL queries over RDF and Wikidata for entity linking in research and library data. Use when modeling research entities as a linked graph。

如何安装知识图谱与SPARQL检索?

在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「知识图谱与SPARQL检索」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。

知识图谱与SPARQL检索需要哪些权限?

该科研 Skill没有声明额外权限;执行具体操作时仍由 Agent 按自身规则请求确认。

知识图谱与SPARQL检索经过官方审核吗?

是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。