GeoGeekLab/nature-reviewer-skills / nature-earth-system-reviewer-skills/skills/nature-atmospheric-science-reviewer-skill
nature-atmospheric-science-reviewer
VerifiedNature-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.
| File | Kind | Size |
|---|---|---|
| LICENSE | other | 1.1 KB |
| MANIFEST.json | config | 368 B |
| README.md | markdown | 7.5 KB |
| SKILL.md | skill md | 10.4 KB |
| checksums.sha256 | other | 1.7 KB |
| docs/SECURITY.md | markdown | 114 B |
| examples/evaluation_prediction.json | config | 260 B |
| references/claim_evidence_protocol.md | markdown | 550 B |
| references/domain_gate.md | markdown | 155 B |
| references/evaluation_contract.md | markdown | 306 B |
| requirements-dev.txt | markdown | 30 B |
| reviewer_db/patterns.csv | other | 17.4 KB |
| reviewer_db/patterns.jsonl | other | 55.9 KB |
| reviewer_db/summary.json | config | 919 B |
| scripts/extract_text_with_anchors.py | script | 1.2 KB |
| scripts/render_review_docx.py | script | 971 B |
| scripts/reviewer_db.py | script | 1.6 KB |
| scripts/validate_package.py | script | 928 B |
| templates/review_report.md | markdown | 953 B |
| tests/test_package.py | script | 640 B |
README.md
View on GitHub# Nature Atmospheric Science Reviewer Skill




**Evidence-centered review for atmospheric observations, reanalysis, numerical and AI models, circulation, chemistry, aerosols, clouds, radiation, extremes, and attribution.**
This package is one of the seven foundational domain reviewer skills in
[Nature Reviewer Skills](https://github.com/GeoGeekLab/nature-reviewer-skills).
> The objective is not language polishing. The objective is to determine whether the manuscript's strongest scientific claims are supported by the evidence actually presented.
## Scientific value
Atmospheric manuscripts often depend on several inferential layers: measurements or retrievals, derived products, model assumptions, diagnostics, mechanism interpretation, and attribution. This skill tests whether those layers form a defensible evidence chain and whether the strongest statements remain within the validated spatial, temporal, and physical domain.
The skill is intended for pre-submission review, internal research-group red teaming, revision planning, reviewer-response preparation, and evidence-centered training for early-career researchers.
## Appropriate scope
Use this skill for:
- satellite, station, radar, lidar, radiosonde, aircraft, and campaign studies
- reanalysis-based diagnostics and multi-product trend studies
- weather, climate, chemistry-transport, Earth-system, and regional models
- AI weather or climate prediction and emulation
- aerosol–cloud–radiation and atmospheric-chemistry studies
- extreme-event attribution, circulation mechanisms, and forced-signal detection
For interdisciplinary manuscripts, combine it with other domain skills or a relevant orchestrator rather than forcing all claims through one disciplinary lens.
## Core evidence gates
The gates are activated according to the manuscript's central claims; they are not applied as a mechanical checklist.
| Review area | What is stress-tested |
|---|---|
| **Observations and products** | Retrieval limitations, quality flags, missingness, vertical sensitivity, terrain and regime dependence |
| **Sampling and representativeness** | Event selection, station coverage, campaign windows, compositing, seasonal and regional scope |
| **Scale and resolution** | Footprint, grid, time aggregation, vertical levels, boundaries, and unresolved processes |
| **Model and AI validity** | Physics, tuning, leakage, benchmark fairness, rollout stability, extremes, spectra, conservation, and OOD behavior |
| **Budgets and mechanisms** | Moisture, energy, radiation, chemistry, transport, circulation, and alternative explanations |
| **Trends and attribution** | Autocorrelation, internal variability, counterfactual design, return periods, risk ratios, and uncertainty |
The package contains **56 abstracted reviewer-reasoning patterns**. They support retrieval and review planning without reproducing raw referee reports.
## Expected review output
A standard run produces **2–4 complementary referee-style reports**, defaulting to three. Each report should:
1. identify the manuscript's central contribution and strongest claims;
2. distinguish direct evidence from derived products, models, correlations, mechanisms, and extrapolations;
3. anchor major concerns to figures, tables, methods, results, supplementary evidence, or explicit missing evidence;
4. explain why each concern matters to the central claim;
5. propose a proportionate resolution: additional analysis, stronger validation, a discriminating experiment, clearer uncertainty, or narrower wording;
6. separate major concerns from local reporting and presentation issues.
A major concern should follow this logic:
```text
claim under review
→ evidence or missing evidence
→ scientific risk
→ alternative explanation or failure mode
→ actionable revision path
```
## Use with an agent runtime
Clone the suite and install the shared runtime:
```bash
git clone https://github.com/GeoGeekLab/nature-reviewer-skills.git
cd nature-reviewer-skills
python -m pip install -e ".[documents]"
```
Point the agent runtime to:
```text
nature-earth-system-reviewer-skills/skills/nature-atmospheric-science-reviewer-skill/SKILL.md
```
For runtimes using `.agents/skills`, copy the complete package directory and keep `SKILL.md`, `reviewer_db/`, `references/`, `templates/`, and `scripts/` together.
Invoke the skill as:
```text
$nature-atmospheric-science-reviewer
```
Example request:
```text
Use $nature-atmospheric-science-reviewer to review the manuscript and supplement. Identify the central atmospheric claims, test observation and product validity, model dependence, mechanism, uncertainty, attribution, and scope, and produce three complementary referee reports with actionable major concerns.
```
## Package contents
```text
README.md user-facing overview
SKILL.md agent instructions and review workflow
MANIFEST.json version, type, runtime, and pattern metadata
reviewer_db/patterns.jsonl canonical reviewer-pattern database
reviewer_db/patterns.csv tabular reviewer-pattern export
reviewer_db/summary.json pattern and gate statistics
references/ evidence protocol, domain gates, and evaluation contract
templates/review_report.md report structure
scripts/ validation, retrieval, extraction, and DOCX rendering wrappers
tests/ package-level regression tests
checksums.sha256 package integrity manifest
```
## Validation and reviewer-memory search
From the repository root:
```bash
python scripts/validate_all.py
python -m pytest
```
Validate this package directly:
```bash
nature-reviewer-validate "nature-earth-system-reviewer-skills/skills/nature-atmospheric-science-reviewer-skill"
```
Search its reviewer memory:
```bash
nature-reviewer-search \
--root "nature-earth-system-reviewer-skills/skills/nature-atmospheric-science-reviewer-skill" \
--query "describe the scientific concern to stress-test" \
--limit 5
```
## v2.1 at a glance
Compared with the v1 generation, v2.1 provides:
- a shared, typed `nature_reviewer_core` runtime instead of duplicated package logic;
- field-weighted BM25 retrieval with phrase support, query expansion, confidence reporting, and diversity-aware selection;
- standardized manifests, package validation, checksums, tests, and benchmark contracts;
- safer manuscript extraction and report generation with explicit input and evidence boundaries;
- clearer multi-referee roles, evidence anchoring, severity calibration, and non-fabrication requirements.
These improvements strengthen consistency and auditability. They do not replace expert scientific judgment.
## Reliability boundary
This skill is a research-assistance and manuscript-quality-control tool. It does not:
- replace qualified domain experts, editors, or formal peer review;
- guarantee correctness, novelty, acceptance, or reproducibility;
- justify unsupported research-misconduct allegations;
- treat a model, retrieval, proxy, correlation, or benchmark as direct proof without validation.
The reviewer memory contains abstracted, non-verbatim reasoning patterns. Raw referee reports are not redistributed.
This project is not affiliated with Nature Portfolio or Springer Nature.
## License
Released under the [MIT License](LICENSE).
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |