How to use the nmrglue NMR 1D processing agent skill

What the skill makes the agent do

An agent skill is a folder of instructions and helper scripts that an AI agent can load when a task matches the skill's description. This one covers processing of calibrated one-dimensional complex NMR FIDs with nmrglue. The skill's description lists the typical jobs: raw 1D NMR processing, ppm-axis verification, apodization, Fourier transformation, manual phasing, baseline correction and reproducible spectral integration.

The skill does not identify compounds or assign resonances. Its output is a processed spectrum, a list of positive peak candidates and signed areas for regions you specify. Interpreting those peaks remains your job.

• Accepts a NumPy .npz file containing exactly one complex fid array, or a canonical 1D complex time-domain NMRPipe file.
• Applies exponential line broadening, zero filling, Fourier transformation and manual phase angles.
• Optionally fits a linear baseline, but only across ppm regions you mark as signal-free.
• Writes spectrum.csv (descending ppm, real and imaginary spectra, fitted baseline) and report.json (input hash, settings, package versions, peak candidates, region areas).

Install the skill

Download the skill ZIP from this page. Then choose one of two routes depending on how you use your agent.

For Claude Code, unzip the file into your user skills folder at ~/.claude/skills/, or into a project-specific .claude/skills/ folder inside the project you are working in. A project folder keeps the skill with that project only.

In the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills instead.

• Option 1, personal: unzip into ~/.claude/skills/
• Option 2, single project: unzip into .claude/skills/ inside the project folder
• Option 3, Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Check what you need before processing

The skill needs Python, nmrglue, NumPy and SciPy. The compatibility note in the skill sets the minimum Python version. The run command uses uv to fetch the packages, so you also need network access for installation. Processing itself is local and needs no credentials.

Before you start, gather the acquisition facts. The skill will not guess them. You need the spectral width in Hz, the positive observation frequency in MHz, the carrier position in ppm, the observed nucleus, and the sign convention of the data. You also need to know whether digital-filter or group-delay removal has already been applied.

• Spectral width (Hz) and observation frequency (MHz)
• Carrier position (ppm) and observed nucleus
• Complex frequency sign convention and whether digital filtering was already removed
• Input as .npz with one complex fid array, or a canonical 1D NMRPipe FID

Run the processing and read the outputs

Copy the supplied processing settings file, assets/processing.json, and replace its example values with your measured parameters and chosen processing options. The skill's documented run command is shown below. Paths assume the skill folder sits at the collection root, so adjust them if you installed it elsewhere. The output directory must be new, which keeps earlier runs reviewable.

After the run, inspect the real and imaginary spectra. If the phase looks wrong, adjust phase0_deg and phase1_deg and run again. Phase is applied after the transform, and there is no automatic phase estimate, so manual checking is required.

• Use nonnegative line broadening in Hz and a zero-filled size at least as large as the acquired FID.
• Only add a baseline when your baseline regions are truly signal-free, with at least two regions.
• Negative areas are kept on purpose. They often point to phase or baseline problems.

Ask the agent for processing in natural language

You do not need to name the skill or its script. Describe the job and the files, and include the facts the skill needs. The skill's description matches requests about raw 1D NMR processing, ppm-axis checks, apodization, Fourier transformation, manual phasing, baseline correction and spectral integration.

Good requests state the input, the settings and the output you want. Vague requests such as 'analyse my NMR' give the agent too little to work with.

• 'Process fid.npz as a 1D complex FID at 400 MHz with 4000 Hz spectral width and carrier at 5 ppm, using 1 Hz line broadening. Give me the peak list and a report.'
• 'Check the ppm axis of this NMRPipe FID against its header, then run the processing.'
• 'Integrate the regions 2.8 to 3.2 ppm and 6.9 to 7.2 ppm and report the signed areas and their ratio.'
• 'Refine the phase of this spectrum, and tell me if the baseline needs correction in the signal-free regions 9.5 to 10 ppm and 0 to 0.5 ppm.'

Understand the limits and safety notes

The skill runs scripts and uses the network for installation. Read the scripts and settings before running them on data you care about. The skill's own notes state that Bruker, Varian and JEOL import paths are not verified by its test suite. A converted file opening correctly does not prove the original acquisition was decoded correctly.

The skill is not a quantitative NMR tool by itself. For quantitative work you must separately establish relaxation delay, pulse angle, saturation, receiver behaviour and the reference amount. The helper does not calculate concentrations or correct unequal relaxation.

• Experimental Bruker, Varian and JEOL data: verify the reader and acquisition metadata first.
• Processed frequency-domain files are rejected by the FID pipeline, but an incorrectly labelled complex spectrum can still pass.
• Peak prominence is a fraction of the largest positive real intensity, not a noise-based detection limit. Strong solvent peaks can hide weak candidates.