科研 Skill
知识库内容一致性审查
semantic-wiki-review
Audits a research wiki’s content for contradictions, outdated claims, and weak support. Use for a meaning-level content review distinct from structural linting.
SKILL.md
Semantic Wiki Review
scripts/lint-wiki.py proves the wiki is structurally well-formed:
frontmatter present, wikilinks resolve, no orphans. It cannot tell you
whether two source pages contradict each other, whether a synthesis still
holds after a newer source was ingested, or whether a claim lacks citation
backing. That work needs reading — and that is what this skill is for.
Announce at start: "Using semantic-wiki-review to audit the wiki for content issues."
Not a gate. Manual trigger only. Runs alongside the workflow, never blocks it.
When to use
- User asks: "are the wiki pages consistent?", "any contradictions?", "which synthesis pages are stale?"
- After a batch of new sources was ingested and earlier syntheses may be stale.
- Before promoting a
status: reviewpage tostatus: stable. - As a recurring sanity pass (monthly, before manuscript submission).
NOT for: structural lint (use wiki-lint), single-page proofreading,
or fact-checking against external sources (that is a literature task).
Scope
Pick a scope before reading — full wiki audit is expensive. Defaults:
| Scope | When |
|---|---|
All status: stable pages | Before manuscript submission |
| All pages touching topic X | After ingesting a major new source on X |
| All syntheses citing source Y | After source Y was substantially revised |
| Recently-changed pages (last N days) | Periodic check |
State the scope to the user before reading.
Checklist
- Determine scope — confirm with user if ambiguous.
- List in-scope pages — read each fully (no skim).
- Build a claim ledger as you read — one line per substantive claim,
tagged with
[[source-key]]and page-of-origin. - Flag issues in the categories below. Each has a recording channel — where
the finding is written so downstream skills (lint, drafting) can act on it:
- Contradictions — two pages assert incompatible things. → Record as a
typed
relations:edge withtype: contradictson the pages themselves (both directions),confidence: inferredunless a verbatim quote grounds it, plus abecause. This is NOT areview_flag(a flag is about one page's own content; a contradiction is a relation between two). - Stale syntheses — a synthesis page relies on sources that a later
source has complemented or refuted. Later means later published (the
year in the bibkey, or
orig_year), not later ingested: a 1996 paper ingested last week is still the older position. Start from theSUPERSEDED-CITEDlines oflint-wiki.py'sResearch chronologysection. →review_flagskind: stale. - Flattened chronology — a page sets positions from different years side
by side as if all were held today: an older view in the present tense with
no word about what came after, or a researcher cited for a view they later
gave up (a
revisesedge exists, or a later source by the same author on the page says otherwise). →review_flagskind: stale, with adetailnaming both sources and their years. Where the later source's answer is only in prose, also write the missingresponds-to/supersedes/revisesedge on the later page (confidence: inferredunless a quote grounds it). Later is not the same as right: flag the flattening, do not decide the debate. - Misdirected edges —
ANACHRONISMlines in the lint output: a stance edge from an earlier source to a later one. → Move it to the later page with the fitting type; keep thebecause. - Unsupported claims — assertions without a
[[source]]link or@citekeyin the text. →review_flagskind: weak-support(ormissing-citationwhen a specific citation is simply absent). - Overstatement — a claim stronger than the source supports (superlatives,
precise figures without consensus). →
review_flagskind: overstatement. This includes over-generalisation, the most frequent failure of model-written summaries: a source's hedge (probably, suggests) or scope (one site, stratum, period, corpus, sample) has fallen away, so the wiki states as general what the source stated as local or tentative. Compare the claim with its cited page; name the dropped qualifier indetail. - Generic conclusions — a synthesis conclusion that would be just as true
of a neighbouring topic (swap the site, text or period and it still holds):
a complex picture emerges, plays an important role, further research
is needed, or a conclusion with no concrete anchor and nothing that could
make it wrong. See
executing-research-plan→ Writing a synthesis. →review_flagskind: weak-support,detailstarting withgeneric:and quoting the sentence. (Weak support, because a claim that names no evidence cannot be checked against any.) - Missing cross-references — pages discussing the same entity / concept without linking each other. → Report only (a linking suggestion, not a content defect); the user or a follow-up adds the wikilink.
- Suspect / aggregator citations — citations whose source URL points to aggregator hosts (academia.edu, researchgate.net) or sources flagged as suspect/pseudoscientific. → Report only. See MCP Optimisation below.
- Contradictions — two pages assert incompatible things. → Record as a
typed
- Record structured findings on the affected pages — for the content
categories above, append a
review_flagsentry to the flagged page's frontmatter (see "Recording review flags" below), and add/strengthen therelations: contradictsedges for contradictions. This is the ONLY editing the audit does: frontmatter metadata only, never the page prose. Existing open flags on a page are left in place (dedupe bykind+detail); do not silently reopen astate: resolvedflag. - Write a review report to
knowledge/_meta/semantic-review-<YYYY-MM-DD>.md(frontmatter:type: synthesis,status: draft,author: llm) — the human-readable digest of the same findings, plus the report-only categories. - Never touch page prose or
status. The flags are advisory metadata; the user (or a follow-up skill invocation) decides which issues to fix, and only the user promotesstatus. Resolving a flag means setting itsstate: resolved(with aresolved:date) once the concern is addressed — not deleting it.
Report format
---
title: "Semantic Wiki Review — <scope> — <date>"
type: synthesis
created: YYYY-MM-DD
updated: YYYY-MM-DD
status: draft
author: llm
---
## Scope
<one sentence>
## Pages audited
- [[page-1]]
- [[page-2]]
- ...
## Contradictions
| Pages | Issue | Suggestion |
|-------|-------|------------|
| [[a]] vs [[b]] | A claims X, B claims ¬X | Resolve in synthesis or note dispute |
## Stale syntheses
| Page | Outdated by | Why |
## Flattened chronology
| Page | Earlier position (year) | Later position (year) | What the page does |
## Unsupported claims
| Page | Claim | Suggested source / mark as opinion |
## Missing cross-references
| Page | Should link | Why |
## Suspect / aggregator citations
| Page | URL | source_class | Primary source known? | Suggested action |
## Summary
<2-3 sentences: how healthy is the wiki, what to prioritise>
Recording review flags
Single-page content findings are written to the flagged page's own frontmatter
as a review_flags entry (schema: schema/knowledge-frontmatter.schema.json).
This is a distinct axis from status (human-owned maturity — never touched
here) and from relations: contradicts (page↔page conflicts). A page can be
status: stable and carry an open flag at the same time — that is exactly the
case worth surfacing: a blessed page a newer source now undercuts.
# appended to knowledge/synthesis/chronologie-debatte.md frontmatter
review_flags:
- kind: overstatement # overstatement | weak-support | stale | missing-citation | open-question
detail: "Dates the destruction 'securely' to 925 BCE; the cited source says 'probably'."
raised_by: semantic-wiki-review
detected: 2026-07-03
state: open # open | resolved
Why on the page and not only in the dated report:
- It gates drafting.
drafting-manuscriptrefuses to draft from a page with an open flag without a logged override — the finding cannot be silently skipped. A detached report can. - It travels with the page — visible in Obsidian and to any later reader,
not buried in a
_meta/file dated weeks ago. - It is auditable and resolvable in place —
state: resolved+ aresolved:date closes it while keeping the trail;wiki-lintreports the open ones.
The dated report (step 6) stays as the human-readable digest and is the only place the report-only categories (missing cross-references, suspect citations) live. Findings and report are two views of the same audit — keep them consistent.
Web-source classification
Classify every cited URL per
docs/source-classes.md. (Until v0.45 the optionaldao-searxng-mcpdid this; it is discontinued — the labels are unchanged.)
Procedure:
- Collect every external URL from the audited source / synthesis pages (frontmatter
sources:or body links). - Per URL: fetch the page (
WebFetch/webfetch) and assign itssource_class(primary_publisher|academic_repository|preprint_server|aggregator|suspect|grey_lit_or_unknown) by the rubric indocs/source-classes.md. - URLs with
source_class=aggregatororsuspectgo into the new "Suspect / aggregator citations" issue category, with a note on whether the primary source (DOI) is known. - No automatic fix — the report is advisory; the user decides replacement or retention with rationale.
Red Flags
| Thought | Reality |
|---|---|
| "I can check this in my head" | Above 5 pages impossible — keep a ledger. |
| "The linter already checked" | The linter checks structure, not content. Different job. |
| "I'll fix issues as I find them" | No — the audit repairs nothing. It records review_flags (frontmatter) and writes the report; fixing prose is a separate step. |
| "Stable means correct" | Stable means human-reviewed at time X. It can still be stale — flag it, don't demote it. |
| "A contradiction is a page flag" | No — a conflict between two pages is a relations: contradicts edge; a review_flag is about one page's own content. |
Key Principles
- Metadata-only, never prose. The audit records
review_flagsandrelationson pages and writes the dated report — it never edits page prose and never changesstatus. Prose repair is a separate, later step. - One report per run, dated, in
knowledge/_meta/; flags live on the pages. - Scope before depth. 10 pages honestly read beats 100 skimmed.
- No gate. This is a tool, not a precondition for other skills.
示例提示词
帮我检查知识库里有没有互相矛盾的论断或被新文献推翻的旧结论。
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.
- 引文与出处核查Checks citations, quotations, page numbers, and bibliography entries for accuracy and source fit. Use before submission to verify source-supported claims.
- 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.
常见问题
知识库内容一致性审查能做什么?
Audits a research wiki’s content for contradictions, outdated claims, and weak support. Use for a meaning-level content review distinct from structural linting。
如何安装知识库内容一致性审查?
在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「知识库内容一致性审查」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。
知识库内容一致性审查需要哪些权限?
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。
知识库内容一致性审查经过官方审核吗?
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。