科研 Skill
引文与出处核查
auditing-citations
Checks citations, quotations, page numbers, and bibliography entries for accuracy and source fit. Use before submission to verify source-supported claims.
SKILL.md
Auditing Citations
Purpose
Audit the evidentiary chain between a manuscript and its sources. This skill does not beautify a bibliography or infer missing publication details. It creates a traceable record showing what was checked, against which source, with what result, and what remains unresolved.
Citation accuracy has at least two distinct dimensions. Bibliographic verification asks whether the cited object exists and whether author, title, date, edition, journal, volume, issue, pages, DOI, archival identifier, and translator information are accurate. Claim–source verification asks whether the cited passage actually supports the sentence, interpretation, historical statement, or quotation to which it is attached. A record can pass the first test and fail the second.
Contract
Accepts a manuscript, notes, bibliography, citation export, quotation list, source files, authoritative catalogue records, or a prior CitationRecord set.
Requires the text containing the citations and enough source access to distinguish checked items from inaccessible ones. Direct quotations require the exact edition or a clearly identified equivalent edition.
Produces normalized CitationRecord objects, an itemized audit table, discrepancy notes, submission-blocking issues, and a citation gate report.
May produce a corrected bibliography proposal, a source-access request list, duplicate-record map, or a list of claims that need weaker wording or different evidence.
Fails when the cited source cannot be identified, a quotation cannot be located, an edition or pagination conflict prevents reliable matching, or the requester asks the agent to invent or conceal missing details.
Guarantees that every audited item receives an explicit status and that inaccessible or mismatched items are not silently represented as verified.
Does not guarantee that inaccessible sources are accurate, that the bibliography is exhaustive, that a verified source makes the manuscript's interpretation correct, or that a journal will accept the citation style.
When to use
Use this skill before submission, after AI-assisted drafting, after importing references from a manager, when direct quotations have been translated or paraphrased, when a reviewer questions a source, or whenever references were assembled from secondary mentions rather than primary consultation.
Do not wait until the final day when the manuscript contains many direct quotations. Citation auditing is most reliable when repeated after major drafting rounds and again at the submission gate.
Inputs required
At minimum provide:
- The manuscript or the passages containing citations.
- The bibliography or reference list, if one exists.
- Accessible source texts, scans, database records, or stable catalogue metadata.
- The target citation style or journal instructions, if format is in scope.
- A statement of which editions, translations, or archival versions the researcher actually used.
When any input is missing, record it as a limitation. Do not treat a citation string as evidence that the underlying work was consulted.
Procedure
1. Create the audit inventory
Extract every in-text citation, footnote, endnote, direct quotation, epigraph, translated passage, data attribution, archival reference, URL, DOI, and bibliography entry. Assign a stable CitationRecord ID. Reconcile obvious duplicates without deleting provenance.
2. Classify citation function
Label what each citation is being asked to do:
- identify a source;
- support a factual statement;
- support an interpretive claim;
- establish a concept's lineage;
- document a direct quotation;
- provide background or further reading;
- support a methodological choice.
A source used only as background MUST NOT be reported as direct proof of a contested claim.
3. Verify bibliographic identity
Check author or editor, title, container title, publication year, edition, translator, volume, issue, page range, publisher, DOI, URL, archive, collection, and item number as relevant. Prefer the source itself, publisher metadata, DOI registry, library catalogue, or archival finding aid over copied citations from unrelated webpages.
Record discrepancies exactly. If two legitimate editions differ, preserve both identities and identify the edition used in the manuscript.
4. Verify direct quotations
Compare wording, spelling, punctuation, ellipses, brackets, emphasis, and surrounding context. Confirm page, paragraph, section, timestamp, image number, folio, or stable location. For translations, distinguish:
- published translation quoted directly;
- researcher's translation;
- AI-assisted draft translation requiring human review.
Do not silently modernize or improve quoted language.
5. Verify claim–source fit
Read enough surrounding context to determine whether the source supports the attached claim. Classify fit as:
- direct support;
- qualified support;
- contextual relevance only;
- support for a narrower claim;
- contradiction;
- unable to determine.
A shared keyword is not sufficient support. A citation to a theorist's name does not validate every later use of that theorist.
6. Check citation placement and scope
Determine whether a citation applies to one sentence, several sentences, or a whole paragraph. Flag ambiguous citation placement, citation clusters where no source is mapped to a specific claim, and paragraph-final citations that appear to support more than they actually do.
7. Reconcile bibliography and manuscript
Identify cited-but-unlisted works, listed-but-uncited works, duplicate identities, inconsistent author names, year suffix errors, broken cross-references, missing translators, and mismatched titles. Formatting corrections come after identity is secure.
8. Assign verification state
Use only explicit states:
verified-primaryverified-secondaryresearcher-suppliedinferenceunverifiedunknowndisputed
Do not use “verified” when only a search snippet or another bibliography was checked.
9. Determine blocking status
Submission-blocking issues include fabricated or absent sources, unverifiable direct quotations, materially wrong page numbers, citations that contradict the attached claim, and unresolved identity conflicts affecting the argument. Minor punctuation differences in a citation style are normally non-blocking.
10. Issue the gate report
Return PASS only when every citation has a status and no blocking item remains. Use CONDITIONAL PASS when non-blocking format issues or clearly listed access limitations remain. Return FAIL when any central quotation, evidentiary citation, or bibliographic identity is unverified or contradicted.
Stop signals
Stop and request researcher action when:
- the source is inaccessible and no authoritative record is available;
- the manuscript cites a work that cannot be located;
- quotation wording materially differs from the source;
- edition or translation identity is unclear;
- a page number belongs to another edition;
- a source supports only a narrower claim than the manuscript makes;
- a secondary citation is presented as direct consultation;
- the requester asks for plausible missing details.
Do not “repair” these problems through guessing.
Completion criteria
The skill is complete only when:
- every citation and bibliography item has a stable record;
- every direct quotation has a checked location or unresolved flag;
- every claim-bearing citation has a claim–source-fit status;
- bibliography and in-text references reconcile;
- discrepancies are separated into blocking and non-blocking groups;
- researcher decisions are identified;
- the final gate follows the documented evidence.
Anti-fabrication rules
Never invent an author, title, page, DOI, issue number, archive identifier, quotation, or source content. Never use search-result snippets as substitutes for reading the relevant passage. Never convert a likely match into a verified identity. Never hide a failed citation by deleting it without reporting the affected claim. Never translate a quotation and present it as an established published translation unless that translation was checked.
When source access is partial, say exactly what was checked. “Metadata verified; passage not checked” is preferable to a polished but false assurance.
Output format
Use templates/citation-audit.md and include:
- Audit scope and source-access statement.
- Citation inventory with record IDs.
- Bibliographic verification result.
- Quotation or passage verification result.
- Claim–source-fit result.
- Required correction.
- Blocking status.
- Counts by verification state.
Gate report
Gate: Citation Audit
Status: PASS | CONDITIONAL PASS | FAIL
Evidence checked:
Blocking issues:
Non-blocking risks:
Researcher decisions required:
Required next action:
Next skill:
Good invocation
Audit all citations and direct quotations in this manuscript against the supplied PDFs. Distinguish metadata verification from claim–source fit and fail the gate if a central quotation cannot be checked.
Bad invocation
Make the references look complete, supply plausible page numbers, and do not tell the reviewer which sources I could not access.
The bad invocation requests concealment and fabrication. The skill MUST refuse it and preserve the unresolved state.
Next skills
checking-terminology-consistencywhen translated names or concepts vary.responding-to-peer-reviewwhen the audit addresses reviewer concerns.verifying-before-submissionafter all blocking citation issues are resolved.structuring-humanities-argumentwhen a claim must be narrowed after source mismatch.
Friction triggers
Create citation/support friction when source existence, locator, quotation fidelity, claim–source fit, or evidential standing diverge.
Friction checks
Audit these five checks separately. A verified source that does not support the claim receives INSUFFICIENT, CONTEXT_ONLY, QUALIFIES, or CONTRADICTS, never automatic support.
Judgment boundary
The agent may recommend narrowing, replacing, or withdrawing a claim. The researcher decides the substantive revision; no citation repair authorizes an invented decision.
Evidence ledger updates
Create or update a ledger entry for material evidence attached to a central or changed claim, including verification method, locator, limitations, counterevidence, and evidential standing. Routine citation identity and formatting checks may remain in the compact citation-audit table when they do not change scholarly standing.
Interpretation history impact
If quotation context changes an interpretation, append a proposed version and route it for researcher review rather than rewriting the accepted reading.
Possible epistemic return
New source content that changes an earlier premise may trigger epistemic return. A false locator, absent source, or misfit citation is defect repair and uses rollback.
Productive refusal
When source access cannot establish a citation or first-use claim, state exactly what was checked, what remains unverified, permitted limited wording, and the source access needed.
Gate impact
Central mismatch, contradiction, or unverified quotation forces FAIL. Material Claim or EvidenceItem change marks dependent argument and submission gates for recheck.
Limitations
An audit is bounded by source access, edition identity, language competence, and the quality of authoritative metadata. It can establish that a passage exists and is represented accurately; it cannot decide by itself whether a complex interpretation is persuasive. Specialist review may be required for rare languages, manuscripts, archival conventions, legal restrictions, or contested editions.
示例提示词
帮我核查这篇书稿里所有引文的页码、元数据和原文是否准确对应。
所属插件
安装该插件会一并获得同领域的 62 个 Skill / MCP。
同插件的其他科研 Skill
- 学术文献获取下载Auto-downloads open-access PDFs for graded sources and writes a manual worklist for paywalled items. Use after literature-review, before ingest-source.
- 单篇文献入库管理Adds one verified PDF to a shared bibliography library, computing its citation key and keywords. Use to register a single document outside the full workflow.
- 数字人文计算方法Surveys computational DH methods: text mining, NLP, corpus linguistics, archive standards, GIS, network analysis, and OCR. Use for an overview of available DH tools.
- 考古地理信息系统分析Runs archaeological GIS analysis with GeoPandas: catchment analysis, viewshed, kernel density, spatial statistics. Use for mapping archaeological site locations.
- 论证结构提取可视化Extracts and visualizes argument structure: Toulmin reconstruction, support/attack graphs, circularity detection. Use to diagram an argument’s logical structure.
- 考古遗物类型学分析Performs artifact analysis: typological classification, ceramic attribute coding, lithic reduction analysis. Use when classifying excavated artifacts statistically.
- 研究选题构思设计Explores research intent, field state, methodology, and ethics, then writes an approved design doc. Use before starting a new research question or chapter.
- 术语译名一致性检查Audits a manuscript for terms, translations, abbreviations, and names whose forms must stay stable. Use when terminology drift could undermine an argument.
- 计算语用学分析Runs computational pragmatics: speech act classification, implicature detection, discourse coherence, politeness analysis. Use for language-use-and-meaning studies.
- 文献综述学术对话构建Turns verified sources into a structured scholarly conversation among positions, not a chronological summary. Use when synthesizing literature into a debate narrative.
- 语料库语言学分析Performs corpus analysis with NLTK/spaCy: collocation, keyword analysis, KWIC concordance, word2vec, diachronic trends. Use for quantitative text-corpus research.
- 数字档案馆资源检索Queries digital archive REST APIs: Europeana, Chronicling America, Internet Archive, HathiTrust, DPLA, OAI-PMH. Use when searching or harvesting heritage metadata.
- 学术手稿章节撰写Drafts a chapter or article section by section from synthesized wiki content, agreeing each section’s argument first. Use to turn stable research synthesis into prose.
- 研究项目漂移状态检测Checks a shared source library and registered projects for drift: un-indexed PDFs, unmerged keywords, bibkey collisions. Use to audit a multi-project research setup.
- EMI课堂语料转写分析Segments and annotates classroom transcripts, coding language choice and discourse function with anonymised output. Use to quantify classroom talk or translanguaging.
- 人工智能伦理分析Analyzes AI ethics: algorithmic fairness metrics, bias auditing, value alignment, moral philosophy reasoning. Use to evaluate an AI system’s ethical dimensions.
- 研究计划任务执行Executes a ready research plan task by task, gating acquisition before ingest and dispatching analysis. Use after writing-research-plan to run the planned work.
- 研究项目结项清单Runs the final completion checklist: rendering, bibliography, wiki, data, reproducibility, archiving. Use when a manuscript is finalized and ready to publish.
- 形式逻辑自动推理Applies formal logic tools: Z3 SMT solver, modal logic S5, SAT encoding, Lean4 proof verification. Use to formalize or mechanically check a logical argument.
- 研究问题凝练构建Turns a broad topic or intuition into a precise, contestable, researchable question. Use at the start of a project when no defensible question yet exists.
- 历史地理信息系统分析Performs historical GIS work: georeferencing old maps, digitizing boundaries, temporal territory overlays, population interpolation. Use for historical spatial analysis.
- 历史语言学比较分析Analyzes historical linguistics in Python: sound correspondences, cognate detection, phylogenetic trees, Swadesh lists. Use when reconstructing language relationships.
- 历史社会网络分析Builds historical network analyses: correspondence networks, prosopography, temporal community detection, Gephi export. Use to map relationships among historical actors.
- 历史文献OCR识别Transcribes printed and handwritten historical documents with Tesseract/Kraken OCR, confidence filtering, spell correction. Use for digitizing scanned historical text.
常见问题
引文与出处核查能做什么?
Checks citations, quotations, page numbers, and bibliography entries for accuracy and source fit. Use before submission to verify source-supported claims。
如何安装引文与出处核查?
在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「引文与出处核查」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。
引文与出处核查需要哪些权限?
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。
引文与出处核查经过官方审核吗?
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。