Metforge analysis
Perform reproducible atmospheric-science calculations and statistical diagnostics. Use when the agent needs to calculate established or custom climate indices, derive meteorological variables, compute trends, detrend or filter time series, perform EOF/PCA, regression, correlation, composites, bootstrap or significance tests, or build coordinate-aware xarray workflows from NetCDF/GRIB/Zarr data. Prefer validated implementations from MetPy, xarray, SciPy, statsmodels, xclim, xeofs/eofs, xskillscore, or another established package before implementing formulas manually. Not for diagnosing why a numerical model or dynamical-core test fails; use metforge-model-diagnose for that.From its SKILL.md
npx -y skills add yuanruichen/MetForge --skill metforge-analysisAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its file declares
Copied from the file, not written here
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
5.3 KB, as published. Nobody here has run it
MetForge Analysis
Turn atmospheric data into a traceable scientific diagnostic. Make the mathematical definition, sampling assumptions, units, execution mode, and validation visible.
Analysis contract
Before computing, establish:
- scientific quantity or hypothesis
- exact formula, published definition, or package function
- variables, units, coordinates, domain, levels, period, season, and baseline
- weighting, anomaly, standardization, missing-data, and event-selection rules
- output variables, dimensions, units, metadata, and acceptance checks
Do not silently choose among competing index definitions. For a named index, recover the authoritative definition and identify any required reference EOFs, climatology, or external coefficients.
Computation workflow
- Inspect files, coordinates, units, calendar, chunking, missing values, and existing project code.
- Choose an established implementation before writing formulas:
- use MetPy with unit-aware inputs for supported meteorological diagnostics
- use xarray for labeled reductions, grouping, rolling, resampling, and alignment
- use SciPy/statsmodels for filters and statistical models
- use xclim for supported climate indices and xeofs/eofs for EOF analysis
- use xskillscore or established statistical libraries for supported metrics
- Read references/indices-and-diagnostics.md, references/trends-and-filtering.md, or references/statistical-methods.md as the task requires.
- Validate the definition on a small subset or synthetic case with a known result.
- Select direct execution or a scheduler using the rules below, then run the smallest valid calculation before scaling out.
- Validate dimensions, coordinates, units, numeric ranges, sample counts, missing-data propagation, and at least one independent or limiting-case check.
- Save self-describing outputs and append an execution record after each completed computation stage.
Scientific safeguards
- Preserve units and coordinate metadata; quantify any loss during format conversion.
- Apply area or mass weighting where the diagnostic requires it.
- Keep preprocessing order explicit. Detrending, anomaly removal, filtering, standardization, and compositing are not interchangeable.
- Do not concatenate separated seasons and then apply a continuous time filter.
- Distinguish regression maps from correlation maps and regression coefficients from composite anomalies.
- Account for serial dependence, multiple testing, and event overlap when they affect inference.
- Treat EOF sign as arbitrary and document weighting, normalization, missing mask, training period, and projection method.
- Never replace a package function with a hand-written approximation only to avoid understanding its assumptions.
Execution mode and processing record
Before running code, inspect the current environment: available CPUs and memory, data volume and chunking, expected runtime, whether sbatch/Slurm is available, whether the session is already inside a Slurm job, and existing project conventions.
- Run directly for metadata inspection, subset validation, interactive exploration, and calculations that fit safely within the current session.
- Use Slurm or the site's scheduler for expensive multi-file calculations, parameter sweeps, memory-heavy operations, or work that should not run on a login node.
- Do not invent account, partition, walltime, or module settings. Infer them from existing scripts/configuration or request the missing value.
- For submitted work, capture the job script, job ID, resources, environment activation, stdout/stderr paths, and terminal status.
After every concrete compute call or completed batch stage, append a record to the project's existing log/provenance location. If none exists, use output/logs/YYYYMMDD-HHMM-<task>.md. Record inputs, software environment, direct/Slurm decision, commands or script, parameters, outputs, validation results, deviations, and status. Never record credentials or secrets. If files cannot be written, return the same record explicitly in the response.
Handoff
Return the calculation definition, code or script, outputs, validation evidence, statistical caveats, and the processing record. Use metforge-figure only when a visual result is needed and metforge-model-diagnose only when the question concerns model failure or numerical behavior.
What ships with it: 4 files
6.6 KB alongside SKILL.md
agents/
- openai.yaml267 B
references/
- indices-and-diagnostics.md2.2 KB
- statistical-methods.md2.1 KB
- trends-and-filtering.md2.1 KB
Gives 0 of the 12 instructions most research analysis skills give
Counted across 1,213 of the 2,113 authors here whose files we hold, read 2026-09-06
- Cite sources for every important claimin 47 of 1213, across 38 files
- Separate facts from inferences and recommendationsin 21 of 1213, across 12 files
- Write findings to a markdown filein 19 of 1213
- Label every insight with a confidence levelin 18 of 1213, across 8 files
- Read product marketing context before asking questionsin 18 of 1213, across 8 files
- Rank themes by frequency and intensityin 16 of 1213, across 6 files
- Establish research mode before proceedingin 16 of 1213, across 6 files
- Segment survey responses by customer tier or tenurein 16 of 1213, across 6 files
- Categorize support tickets before analyzingin 16 of 1213, across 6 files
- Weight research sources from the last twelve monthsin 16 of 1213, across 6 files
- Use at least five data points per segmentin 15 of 1213, across 5 files
- Extract verbatim quotes for all research findingsin 15 of 1213, across 5 files
Said here and by no other author read
- Establish scientific hypothesis and formula before computing
- Use established packages before writing custom formulas
- Validate definitions on small subsets or synthetic cases
- Preserve units and coordinate metadata throughout calculations
- Apply area or mass weighting where required
- Keep preprocessing order explicit
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once.