科研 Skill
同行评审流程发起
requesting-peer-review
Runs a two-stage constructive-then-adversarial review of a draft against discipline checklists. Use when a chapter or grant draft is ready for structured review.
SKILL.md
Requesting Peer Review
Subject a drafted manuscript to structured review before submission. Two passes: constructive (strengthen the argument, improve clarity) and adversarial (find holes, challenge assumptions). Each pass is a fresh subagent with the manuscript and relevant sources — no conversation history pollution.
Announce at start: "Using requesting-peer-review on ."
If unmet: explain to the user which condition is missing, ask for a short
reason, write it to knowledge/_meta/gate-overrides.log, and close out.
Closing a review with open Major Issues is a deliberate decision, not an
oversight — the reason in the log makes that traceable.
When to use
- A manuscript QMD file exists under
output/and has rendered successfully - Before sending to an external reviewer, before submission, before grant deadline
- After major revision, to re-check whether open issues were addressed
NOT for: first-draft feedback (use drafting-manuscript's architecture and per-section stops), copy-editing (separate task), single-paragraph checks.
Checklist
- Confirm target manuscript and render status (
make rendermust succeed) - Identify discipline — Theologie, Biblische Archäologie, Alte Geschichte, Digital Humanities, Mixed Methods — each has specific review criteria
- Select reporting standard — CONSORT (RCT), STROBE (observational), MRC Framework (complex interventions), PRISMA (systematic review), ARRIVE (animal), TRIPOD (prediction models); discipline-specific: stratigraphic documentation (archaeology), Formkritik/Quellenkritik (theology), source criticism (DH)
- Scope the dispatches by length — see Review scope. An article or a single chapter is reviewed whole; a book or any manuscript over ~12,000 words is reviewed per chapter (one constructive and one adversarial dispatch per chapter) plus one structure pass over the whole. Long inputs measurably weaken model judgement of argument; a review of 80,000 words at once is a review of the first and last chapters
- Dispatch constructive reviewer subagent (see
agents/peer-reviewer.md) with roleconstructive - Dispatch adversarial reviewer subagent (fresh context) with role
adversarial - Collate reviews into a single
output/<target>/reviews/<date>-review.mddocument - Classify each issue: Major / Minor / Editorial / Out-of-scope. Keep the reviewer's
[confirm with subject expert]tag on the issue — it travels into the walk-through - Walk the user through each Major and Minor — decision: accept → revise, reject (with rationale), defer (with log). An issue tagged
[confirm with subject expert]is a model's claim about the field, not a finding: the user decides whether they can confirm it from their own expertise or whether it goes to a human colleague first. Present the two sections Missing voices and What this review could not check as well — they are where the review is honest about its own reach - Log decisions in
knowledge/_meta/log.md, one line per pass, in this exact shape —lint-wiki.py'sHALF-REVIEWgate reads it, and an unlogged pass is an unverifiable claim that the review happened:
## [YYYY-MM-DD] review | constructive | <one-line verdict>
## [YYYY-MM-DD] review | adversarial | <one-line verdict>
If a pass is skipped, log it anyway with the deciding reason (review | adversarial | SKIPPED — <reason>, decided by <who>). The gate then still reports the gap, which is correct: the manuscript went out on half a review, and the record should say so.
11. If accepted issues → revisions needed: route back to drafting-manuscript with issue list
Process Flow
digraph peer_review {
"Confirm manuscript + render" [shape=box];
"Identify discipline" [shape=box];
"Select reporting standard(s)" [shape=box];
"Dispatch constructive reviewer" [shape=box];
"Dispatch adversarial reviewer" [shape=box];
"Collate reviews" [shape=box];
"Classify issues" [shape=box];
"Walk issues with user" [shape=box];
"Any major open?" [shape=diamond];
"Route to drafting-manuscript (revision)" [shape=box];
"Log decisions" [shape=box];
"Done" [shape=doublecircle];
"Confirm manuscript + render" -> "Identify discipline";
"Identify discipline" -> "Select reporting standard(s)";
"Scope by length" [shape=box];
"Select reporting standard(s)" -> "Scope by length";
"Scope by length" -> "Dispatch constructive reviewer" [label="per chapter if long"];
"Scope by length" -> "Dispatch adversarial reviewer" [label="per chapter if long"];
"Dispatch constructive reviewer" -> "Collate reviews";
"Dispatch adversarial reviewer" -> "Collate reviews";
"Collate reviews" -> "Classify issues";
"Classify issues" -> "Walk issues with user";
"Walk issues with user" -> "Any major open?";
"Any major open?" -> "Route to drafting-manuscript (revision)" [label="yes"];
"Route to drafting-manuscript (revision)" -> "Dispatch constructive reviewer" [label="after revision"];
"Any major open?" -> "Log decisions" [label="no"];
"Log decisions" -> "Done";
}
Review scope
| Manuscript | Dispatches |
|---|---|
| Article, single chapter, grant narrative (≲ 12,000 words) | 1 constructive + 1 adversarial, whole text |
| Book, thesis, or longer manuscript | per chapter: 1 constructive + 1 adversarial; then 1 structure pass over the whole |
The structure pass (role constructive, fresh context) does not reread every
chapter. It receives the introduction, the conclusion, the outline files with
their claim chains and section reverse outlines (output/<target>/outline/*.md,
written by drafting-manuscript), and the chapter-level review summaries. It
asks only global questions: does the book's thesis survive the sequence of
chapters, are any chapters doing the same job, does the conclusion answer what
the introduction asks, is a key term used in two senses across chapters?
Collate per-chapter reports under one dated review document with a section per
chapter. Log one line per role as before (the HALF-REVIEW gate reads
that), naming the chapters covered in the verdict — e.g.
## [2026-11-13] review | adversarial | ch. 1–6 + structure; 4 Major (ch. 3, 5).
Reviewer Assignment
Two subagents, FRESH context each:
Constructive reviewer (agents/peer-reviewer.md, role: constructive):
- Goal: improve the argument, clarify, strengthen
- Tone: supportive senior colleague
- Prompts: "Where is the argument weakest but most important?" "What evidence would strengthen Claim X?" "What does the reader need that isn't here?"
Adversarial reviewer (same skill, role: adversarial):
- Goal: find holes, challenge interpretations, flag methodological gaps
- Tone: rigorous external reviewer (Reviewer 2)
- Prompts: "Which claim is inadequately supported?" "What alternative explanation isn't addressed?" "Which sources are missing?" "Is the statistical/stratigraphic/philological method defensible?"
Dispatch prompts are neutral. A model's verdict follows the prompt more than the text: the same essay, framed three different ways, is graded anywhere from good to failing. So the dispatch states the role, the materials and the template — and nothing about how good the manuscript is, how much work went into it, what the author hopes to hear, or how close the deadline is. No scores, grades or acceptance probabilities are asked for; the review's output is issues with locations and reasons. The adversarial role is a way of reading (look for the weakest link), not a presumption that the text is weak.
Both subagents receive:
- The manuscript file
- The relevant synthesis and source pages (paths listed)
- The selected reporting standard(s)
- The research plan's Hypothesis + Falsification Criteria
- A blank structured template for their report
Review Report Template
Saved to output/<target>/reviews/<YYYY-MM-DD>-<role>-review.md:
---
title: "<Role> Review — <Manuscript>"
reviewer: <subagent-id>
role: constructive | adversarial
date: YYYY-MM-DD
manuscript: <path>
reporting_standard: <CONSORT|STROBE|...|discipline-specific>
---
## Summary
<2–3 sentences: overall impression>
## Major Issues
1. **<Issue>** — <why it matters, where it lives (section/line), suggested resolution>. Tag `[confirm with subject expert]` when the issue rests on a claim about the field's facts (a date, a reading, a find, what the literature holds) rather than on the manuscript's own logic or its cited sources
...
## Minor Issues
...
## Editorial
(typos, unclear phrasing, citation format)
## Methodological Assessment
<assessment against the selected reporting standard>
## Missing Voices
Which schools, positions, language areas or kinds of evidence does the argument
not engage that a specialist reader would expect? (E.g. German- or
French-language scholarship, Israeli excavation reports, a rival dating model,
epigraphic or iconographic evidence.) Model-assisted research leans towards the
most-cited, recent, English-language mainstream — name what that bias would
have hidden here. "None identified" only after checking.
## What This Review Could Not Check
Be explicit about the review's own reach: claims that need first-hand knowledge
of a site, a collection, unpublished material or a language you cannot verify;
sources you could not open; technical judgements (statistics, stratigraphy,
textual criticism) where your assessment is a plausibility check, not an expert
verdict. A reader of this review should know which of its silences mean
"checked, fine" and which mean "not checkable here".
## Falsification Test
Quantitative/mixed: does the manuscript honestly engage with the pre-registered falsification criteria? Hermeneutic: does it state what evidence or reading would refute its thesis, and engage it? If no, flag.
Critical Thinking — Evidence Audit for Reviewers
Both reviewer roles (constructive and adversarial) apply the evidence framework before any substantive assessment:
- Make the claim explicit — what exactly is being claimed, in one sentence?
- Evidence type — primary data, secondary synthesis, expert opinion, theoretical argument
- Choose the framework:
- GRADE — causal / quantitative claims
- Cochrane ROB — experimental designs
- Quellenkritik — textual-historical evidence (form, redaction, textual criticism)
- Stratigraphic reliability — archaeological findings
- Reproducibility audit — DH / computational
- Confounders — which alternative explanations does the manuscript NOT discuss?
- Logical fallacies — ad hominem, affirming the consequent, equivocation, cherry-picking, generalisation from single finds
- Argument structure (Toulmin) — for each central claim: what is the data (the evidence cited), what is the warrant (the rule that licenses the step from data to claim — a dating convention, a typological assumption, a hermeneutic principle), and is the warrant stated or silently assumed? An unstated warrant that a specialist would contest is a Major Issue
- Falsification test — would the author acknowledge what would refute the thesis, and engage it honestly?
Adversarial reviewers may apply this checklist more harshly (Reviewer 2); constructive reviewers name gaps constructively with a suggested improvement.
Grading-note template (optional appendix to the review report):
## Evidence Grading — <Claim>
**Claim:** <exact statement from the manuscript>
**Framework:** GRADE | Cochrane ROB | Quellenkritik | Stratigraphic | Reproducibility
**Evidence type:** <primary / secondary / theoretical>
**Grade:** strong | moderate | weak | very weak
**Rationale:** Design, Bias, Precision, Consistency, Directness — short
**Confounders addressed in manuscript:** <yes/no, which>
**Logical fallacies:** <list or none>
**Falsifiability:** <testable / not testable>
Discipline-Specific Checklists
Source: research-skills/peer-review/references/ has general checklists. Add:
Biblical Archaeology / Ancient History:
- Stratigraphy documented with Locus/Stratum references
- Ceramic typology tied to named sequences (e.g. Iron IIA, IIB)
- 14C dates with lab numbers and calibration method
- Terminological consistency (Levantine Iron Age vs. Iron II, etc.)
Theology / Biblical Studies:
- Textual basis named (MT / LXX / DSS / Peshitta) with critical edition
- Formgeschichte / Redaktionsgeschichte method stated
- Engagement with German-language scholarship where relevant
- Hermeneutical stance transparent
Digital Humanities:
- Dataset provenance + license
- Reproducibility: code + data availability statement
- Tools and versions named
- Limitations of method explicit
MCP Optimisation (recommended)
If
dao-paper-search-mcp(seedocs/recommended-mcps.md) is available in the project, use it for the Cited Evidence Audit section of the review report. Otherwise, do the spot-checks manually (open the full text, compare the claim with the original passage).
Extended Cited Evidence Audit (5 spot-checks):
| Citation | Claim | Source verifies? | source_class | Note |
|---|---|---|---|---|
[@bibkey, p. X] | yes / no / partial | primary_publisher / aggregator / ... |
- For each cited DOI:
dao-paper-search-mcp.search_crossref(query=<doi>)or equivalent adapter, checking that the hit'sidentifiers.doiis that DOI; compareinline_citation.authoritative_bibliography_linefrom the MCP response against the manuscript bibliography — discrepancies are Major Issues. - For web citations (online sources): fetch the page and classify it per
docs/source-classes.md. Aggregator or suspect classification without an explanatory note in the manuscript is at minimum a Minor Issue. - Without
dao-paper-search-mcp: check DOIs by hand (resolve them, compare author, year and title); keep thesource_classcolumn.
Red Flags
| Thought | Reality |
|---|---|
| "One review pass is enough" | Constructive AND adversarial — both. |
| "Reviewer 2 is unfair, I'll ignore them" | Harsh critique is often the most load-bearing. Classify the entry, don't discard. |
| "I know the literature, external review is overkill" | Blind spots are by definition what you don't see. |
| "Falsification check is only for quantitative work" | Qualitative work also has to say what would refute its thesis. |
| "I'll handle Minor Issues later" | Later means never. Freeze them in the review document. |
| "The book is one manuscript, one dispatch is cleaner" | Long inputs weaken judgement of argument. Per chapter, plus a structure pass. |
| "I'll tell the reviewer this is nearly ready, so it focuses on polish" | Then the verdict is the one you asked for. Neutral dispatch: role, materials, template. |
| "The reviewer says the dating is wrong — accept it" | A tagged field claim is the model's, not a finding. The user confirms it or a colleague does. |
| "Nothing is missing — the review found no gaps" | Then the Missing voices and Could not check sections say why. Silence is not coverage. |
Key Principles
- Two passes, fresh context each — constructive + adversarial, kept separate; per chapter for long manuscripts, plus a structure pass
- Neutral dispatch — no framing of quality, no scores; the prompt must not decide the verdict
- The review states its own reach — missing voices, what could not be checked, field claims tagged for expert confirmation
- Discipline-specific checklist — no universal template
- Falsification check — quantitative: honest engagement with the pre-registered hypothesis; hermeneutic: the manuscript says what would refute its thesis
- Classify before revising — Major / Minor / Editorial / Out-of-scope before editing
- Log per decision — accepted / rejected / deferred all documented
示例提示词
我的章节已完成初稿,帮我按建设性再对抗性的两阶段做同行评审。
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
同行评审流程发起能做什么?
Runs a two-stage constructive-then-adversarial review of a draft against discipline checklists. Use when a chapter or grant draft is ready for structured review。
如何安装同行评审流程发起?
在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「同行评审流程发起」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。
同行评审流程发起需要哪些权限?
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。
同行评审流程发起经过官方审核吗?
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。