MRFOUNDRY IN PRACTICE01 / 10

Give your agent MRI working guidance.

01 · INSTALL FOR YOUR AGENT

npx skills add KeWang0622/MRFoundry -g

Choose your skills and supported agent in the installer.

02 · ASK IN CODEX OR CLAUDE CODE

Use MRFoundry. Design a golden-angle radial GRE in PyPulseq. Simulate the exported sequence, reconstruct the signal, and check the result.

Skills are guidance. Scientific applications are installed separately as needed.

MRFOUNDRY IN PRACTICE02 / 10

You set the goal. The agent does the work.

Researcherquestion + constraintsCodex / Claude Codereasons · writes code · executesScientific toolsPyPulseq · KomaMRI · SigPyMRFoundry skillsmethods · conventions · checksguidanceoutputs → review → next action
The seven skills are reusable instructions—not seven separate models.

MRFOUNDRY IN PRACTICE03 / 10

The guidance becomes a sequence file.

MRFOUNDRY GUIDANCEpulse-sequence-designUse PyPulseq.Check timing,limits and trajectory.Export the sequence.Paraphrased current instructionsAGENT ACTIONConfigure → export → read backpython design_sequence.pyRFGxGygolden_angle_radial_gre.seqTiming passed · exported trajectory checkedView report ↗
Recorded workflow mapped to current skills · waveform points from the original run

MRFOUNDRY IN PRACTICE04 / 10

Execute the exported sequence in KomaMRI.

MRFOUNDRY GUIDANCEpulse-sequence-designUse an establishedBloch simulator.Test its upstreamexample first.Paraphrased current instructionsAGENT ACTIONRead the .seq into KomaMRIpython simulate_koma.pyOfficial smoke testPASSFull exported trainRUNkoma_signal.npz8,192 finite complex ADC samplesView report ↗
Recorded workflow mapped to current skills · KomaMRI via its official Python interface

MRFOUNDRY IN PRACTICE05 / 10

Carry the right conventions across tools.

MRFOUNDRY GUIDANCEmri-reconstructionUse the actualnon-Cartesiantrajectory.Inspect conventionsand residuals.Paraphrased current instructionsAGENT ACTIONConnect signal + coordinatespython reconstruct_sigpy.pyKomaMRI signalExported k-spaceSigPyNUFFT + CGcoords = (ky, kx) × FOVreconstruction.npz + report.jsonComplex image · prediction · data residualView report ↗
Recorded workflow mapped to current skills · established SigPy operators and solver

MRFOUNDRY IN PRACTICE06 / 10

A runnable result still needs your judgment.

MRFOUNDRY GUIDANCEmri-research-workflowEvaluate the resultagainst the goal.Investigate failure.Keep conclusionswithin the evidence.Paraphrased current instructionsRESEARCHER REVIEWInitial image qualitywas rejected.AGENT NEXT ACTIONTest phase and axes.Check solver stability.Refine phantom sampling.
Documented human intervention · condensed description, not a quoted chat transcript

MRFOUNDRY IN PRACTICE07 / 10

Turn feedback into the next action.

Researcher feedbackrevise the experimentAgent + guidancediagnose · change · rerunNew evidencefailed + passed checks10.10% → 5.48% → scoped 1.60% checkSAVE THE CONTEXTparameters · failures · references · next steps
Saved run artifacts are observed; .mrfoundry/ formalizes reuse in the current skills.

FOR FUN / SAME WORKING PATTERN08 / 10

The same workflow can start with a playful idea.

ASK: “WRITE MRI IN K-SPACE”PyPulseq↓SigPyactual exported trajectory20 rotated wordsrequest → skill guidance → code → checked artifacts
FOR FUN · static Fourier model · more shots take more time · not a clinical sequence

MRFOUNDRY IN PRACTICE09 / 10

What you add to an already capable agent.

01InstructionsRead again at the next task.02ConventionsUnits, axes, formats, tool choices.03ChecksWhat to test before trusting a result.04Sources + notesWhere the method and decisions came from.
Reusable MRI working practices · no claim of greater intelligence or proven baseline superiority

MRFOUNDRY IN PRACTICE10 / 10

Try it. Inspect what happens. Iterate.

Start with your own question.

Read the skills ↗Inspect this run ↗Open the sequence ↗Run the fun example ↗

Keep the tools, assumptions, checks and revisions visible.

MRFoundry · Ke Wang · source-linked workflow demonstration