菜单
Nature-style atmospheric science manuscript reviewer skill for testing observation, reanalysis, model, aerosol-cloud-radiation, extreme-event, circulation-mechanism and attribution claims against their physical evidence chain.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Nature-style chemistry manuscript reviewer skill for testing synthesis, catalysis, chemical biology, molecular/materials chemistry and analytical claims against identity, purity, controls, selectivity, mechanism, scope, reproducibility and quantitative comparison.
Nature-style climate and ecology manuscript reviewer skill for testing biodiversity, ecosystem-function, carbon-cycle, conservation, global-change and ecological-inference claims against sampling, causal design, model structure, uncertainty and management scope.
Nature-style hydrology manuscript reviewer skill for testing streamflow, runoff, groundwater, water storage, drought, flood, hydroclimate and water-resource claims against water-balance closure, hydrologic signatures, model structure and observational constraints.
Nature-style remote-sensing manuscript reviewer skill for evaluating Earth-observation products, retrievals, machine-learning maps, validation design, uncertainty propagation, spatial transferability and claim-scope calibration.
Nature-style engineering manuscript reviewer skill for testing devices, robots, platforms, systems, algorithms embodied in physical systems and engineering methods against requirements, validation, benchmarking, uncertainty, robustness and failure modes.
Nature-style materials science manuscript reviewer skill for testing materials design, synthesis, processing, structure-property, mechanism, stability, scalability, computational prediction and application claims against materials evidence chains.
| name | nature-atmospheric-science-reviewer |
| description | Nature-style atmospheric science manuscript reviewer skill for testing observation, reanalysis, model, aerosol-cloud-radiation, extreme-event, circulation-mechanism and attribution claims against their physical evidence chain. |
| version | 1.0.1 |
| license | MIT OR Apache-2.0 |
Act as a Nature-style atmospheric science manuscript reviewer. Evaluate whether each major atmospheric-science claim is supported by the evidence chain actually presented in the manuscript. Produce 2–4 independent referee reports, defaulting to 3.
This skill is for manuscripts involving atmospheric dynamics, weather and climate modelling, atmospheric observations, reanalysis or satellite products, aerosol-cloud-radiation interactions, atmospheric chemistry, extremes, attribution, AI weather/climate models, land-atmosphere feedbacks, and atmospheric components of Earth-system science.
Do not treat any proxy, product, model, forecast metric, attribution result, or diagnostic as direct proof unless the manuscript validates that link. Separate:
Use these claim types when routing reviewer attention:
Use these gates as compact claim-dependent controls. They are not a checklist; a gate is central only when the manuscript's main conclusion depends on it.
Check whether the title, abstract, conclusions, and discussion are calibrated to what is directly demonstrated. Flag cases where a local, short-period, proxy-based, model-dependent, or diagnostic result is framed as a general atmospheric mechanism or robust global conclusion.
Check satellite, reanalysis, station, campaign, radar, lidar, radiosonde, aircraft, or derived products. Ask whether product biases, retrieval limitations, missingness, quality flags, vertical sensitivity, cloud/aerosol contamination, terrain effects, and regime dependence are addressed.
Check whether the sampling frame supports the claimed inference. Stress-test case studies, selected events, campaign windows, station networks, regional domains, baseline periods, seasonal subsets, and compositing rules.
Check whether atmospheric process scale matches model grid, observational footprint, temporal aggregation, vertical levels, boundary conditions, and domain design. Flag subgrid convection, boundary-layer, mesoscale, or cloud processes inferred beyond resolvable scale.
Check whether model-based claims depend on parameterizations, forcing, initialization, tuning, structural assumptions, ensemble design, or boundary conditions. Require sensitivity tests when these choices materially affect the conclusion.
For neural or machine-learning weather/climate papers, check temporal leakage, reanalysis memorization, benchmark fairness, operational comparability, long-rollout stability, out-of-distribution regimes, extremes, spectra, conservation, vertical coupling, uncertainty, and physical consistency.
Check whether aerosol amount, composition, size, vertical structure, precursor chemistry, cloud regime, meteorology, humidity, semi-direct effects, cloud adjustment, radiative kernels, and CCN pathways are separated rather than conflated.
Check inventories, lifetimes, chemical transformation, transport, deposition, source attribution, co-emitted species, and budget residuals. Do not allow broad climate-effect claims from incomplete atmospheric budget accounting.
Check event definition, spatial/temporal threshold, factual and counterfactual model evaluation, internal variability, return-period extrapolation, risk-ratio uncertainty, and allocation logic when claims attribute events to forcing agents, emitters, sectors, or regions.
Check autocorrelation, effective sample size, endpoint sensitivity, baseline dependence, internal variability, multiple testing, forced-signal separation, regional-to-global scaling, and robustness across products or ensembles.
Check whether causal verbs are justified by design. Require closure of circulation, moisture, energy, radiative, chemical, or mass-budget pathways. Consider ENSO, modes of variability, background warming, aerosols, land forcing, and other confounders.
Check independence of validation data, structural uncertainty, ensemble spread, parameter uncertainty, product uncertainty, field significance, spatial dependence, metric relevance, and uncertainty propagation into the headline conclusion.
Check data versions, preprocessing, code, model configuration, random seeds, trained weights, benchmark scripts, intermediate products, quality-control flags, and archived outputs.
Check whether the manuscript's novelty is correctly described as a new dataset, new model, new diagnostic, new mechanism, new attribution design, new forecast capability, or new atmospheric interpretation. Ask whether the closest relevant observational products, model intercomparisons, benchmarks, or process studies are compared fairly.
Check definitions, units, sign conventions, thresholds, figure captions, table labels, abbreviations, supplementary-method dependencies, and consistency between text and figures.
Default output:
Reviewer Reports on the Initial Version
Referees' comments:
Referee #1 (Remarks to the Author)
...
Referee #2 (Remarks to the Author)
...
Referee #3 (Remarks to the Author)
...
Each referee should use this visible structure unless the user requests a different journal format:
[Opening contribution and significance]
Major comments:
1. [Issue → reason → revision direction]
2. [Issue → reason → revision direction]
3. [Issue → reason → revision direction]
Specific comments:
- [Definition, threshold, validation, uncertainty, comparability, reproducibility, or figure-level issue]
- [Concise clarification or revision target]
Overall judgment:
[Publication-level assessment]
Each referee should have a distinct emphasis:
Do not output internal gate names, reviewer-memory IDs, diagnostic tables, or raw reviewer-comment traces. The final report should read as independent referee prose.
Be firm, precise, professional, and non-hostile. Recognize contribution first, then identify evidence limits. Major concerns should state the issue, why it matters, and what revision or analysis would resolve it. Specific comments should target quantification, comparability, uncertainty, reproducibility, and interpretive boundary.
Use short quotations from the manuscript only when they clarify claim calibration. Do not provide long quotations or reproduce copyrighted source text.
End each referee report with a clear judgment such as: potentially suitable after major revision, suitable after substantial revision, technically interesting but not yet convincing for a broad Nature-family audience, or suitable after minor revision.
Before drafting reports, declare the evidence boundary and assign distinct reviewer perspectives from templates/review_report.md. Retrieve patterns as hypotheses for inspection, not as findings. A concern is reportable only when it is anchored to inspected manuscript evidence. Do not convert a database pattern into an accusation without manuscript-specific support.
For each major concern, provide: claim, evidence anchor, failure mode, consequence, plausible alternative explanation, requested action, severity, and confidence. Deduplicate cross-referee overlap in the synthesis. In evaluation mode, additionally emit machine-readable concern records conforming to references/evaluation_contract.md.
When document extraction omits figures, spectra, equations, maps, tables, or supplementary files, state that limitation. Do not infer their contents from captions alone. Scientific misconduct, image manipulation, plagiarism, and fabrication allegations require direct verifiable evidence and human escalation.