A Claude Code skill that turns Claude into a domain expert for EEG/MEG analysis, BCI development, and neuroscience signal processing — from raw recordings to production-grade models.
ClaudeEEG loads a comprehensive neuroscience knowledge base into Claude's context, covering the full EEG analysis stack:
| Domain | Coverage |
|---|---|
| Theory | Frequency band biology, ERP components, cross-frequency coupling, volume conduction, artifact physics |
| Preprocessing | Gold-standard pipeline ordering — filtering, ICA, ASR, AutoReject, re-referencing |
| Libraries | MNE-Python, EEGLAB/MATLAB, SciPy, scikit-learn, braindecode, pyRiemann |
| ML / DL | CSP+LDA, Riemannian geometry, EEGNet, ShallowConvNet, foundation models (LaBraM, BIOT, SignalJEPA) |
| Statistics | Permutation tests, FDR/Bonferroni correction, cluster-based inference, mixed-effects models |
| BCI Paradigms | Motor Imagery, P300, SSVEP, neurofeedback |
| Clinical | Sleep staging, seizure detection, resting-state connectivity |
| Data Formats | EDF, BrainVision, EEGLAB .set, FIF, CNT, EGI/MFF, BIDS |
npx skills add https://github.com/Krish-mal15/ClaudeEEGOnce installed, the skill is available in any Claude Code session. Invoke it explicitly or just use a trigger keyword in your message:
/ClaudeEEG help me preprocess this EEG dataset
"I have 64-channel EEG from a motor imagery task. Walk me through a full preprocessing pipeline."
"My data has heavy ocular artifacts. How should I set up ICA in MNE?"
"Should I use ASR before or after ICA for a mobile EEG recording?"
"Help me implement the dual-dataset ICA trick to preserve slow ERPs."
"Extract and plot P300 components from my oddball task data."
"What baseline correction window should I use for a CNV paradigm?"
"My N200 amplitudes look wrong after filtering — what's going on?"
"Build a real-time motor imagery classifier using CSP + LDA."
"Compare Riemannian MDM vs EEGNet for cross-subject motor imagery — which should I use?"
"Set up a P300 speller pipeline with xDAWN spatial filtering."
"Implement an SSVEP decoder using CCA for a 4-target paradigm."
"Train EEGNet on BCI Competition IV 2a using braindecode."
"My classifier has 95% within-session accuracy but drops to 60% cross-session. How do I fix this?"
"Fine-tune LaBraM on my small dataset."
"Set up proper leave-one-subject-out cross-validation to avoid data leakage."
"Compute alpha ERD/ERS during motor preparation using Morlet wavelets."
"Measure theta-gamma phase-amplitude coupling in my working memory data."
"Calculate wPLI connectivity between frontal and parietal channels."
"Stage sleep from polysomnography data using AASM criteria."
"Detect interictal epileptiform discharges in a long EEG recording."
"Compute individual alpha frequency (IAF) per subject."
"My ICA components all look like noise — what went wrong?"
"Walk me through interpreting these ICLabel labels and probabilities."
"Why does my source localization look wrong?"
"Critique this preprocessing pipeline and flag any ordering errors."
pip install mne mne-icalabel braindecode pyriemann autoreject meegkit scikit-learn torch| Package | Purpose |
|---|---|
mne >= 1.6 |
Core EEG/MEG processing |
mne-icalabel |
Automated IC classification |
braindecode |
Deep learning for EEG |
pyriemann |
Riemannian geometry classifiers |
autoreject |
Automated epoch rejection |
meegkit |
Zapline / DSS line-noise removal |
scipy, scikit-learn |
Signal processing and ML |
torch |
PyTorch for DL models |
ClaudeEEG/
├── SKILL.md # The skill itself — loaded into Claude's context
├── README.md # This file
├── scripts/ # Runnable Python pipeline examples
└── references/
├── MNE Resources/ # MNE tutorials (preprocessing, epochs, inverse, stats)
└── NumPy Resources/ # NumPy reference docs
These behaviors are baked into the skill and activate automatically:
- Gamma skepticism. Scalp gamma (30–80 Hz) fully overlaps the muscle artifact range. The skill flags uncritical gamma claims and requires rigorous EMG rejection before accepting gamma findings.
- ICA ordering enforcement. The skill warns if you filter after epoching, run ICA on unfiltered data, or skip ASR for mobile recordings.
- CV hygiene. ML pipelines always use subject-stratified splits. Random-shuffle splits on sliding windows are flagged as data leakage.
- 2024 literature. Includes the Kang et al. finding (ICA may hurt DL decoding) and the Callan et al. ASR-before-ICA result for mobile EEG.
The skill activates on any of: EEG · MEG · electrophysiology · BCI · brain-computer interface · ICA · ERP · epochs · artifact removal · EEGLAB · MNE · braindecode · NeuroPype · neural decoding · oscillations · frequency bands · power spectral density · source localization · motor imagery · P300 · SSVEP · neurofeedback · neuro
