Neurotech skills for Claude Code

32 skills and 6 agents that give Claude the established workflows of neural engineering: BCI decoding, EEG/MEG and spike analysis, real-time rigs, brain stimulation and neural devices. Free and open source under the MIT licence.

View on GitHub

Install the Claude Code plugin

/plugin marketplace add Hillock-NeuroTech/ultimate-neurotech-claude-skills
/plugin install neuro-router@hillock

Then add the bundles you work with, for example /plugin install neuro-decoding@hillock. Skills load automatically when a request matches.

What it catches

Four panels: a decoder that scores 0.94 split by trial and 0.57 split by session; a Medtronic Percept stream rebuilt with lost packets as gaps; two ultrasound protocols with the same mechanical index and ten times different Ispta; an EEG spectrum with mains harmonics and one muscle channel.
Each panel is one of the suite's scripts run on synthetic data.

The same decoder scores 0.94 when EEG trials are split at random and 0.57, below chance, when whole recording sessions are held out. With Hillock loaded, Claude evaluates decoders with session- and subject-wise splits, true chance levels and permutation tests by default.

Skill bundles

neuro-router

Entry point that maps any neurotech request to the right workflow; reading every neural file format (EDF, BrainVision, XDF, NWB, Neuralynx, Blackrock, Intan, Open Ephys, SpikeGLX and more); finding public EEG, MEG, iEEG and spike datasets.

neuro-decoding

neuro-signal

neuro-realtime

neuro-stim

neuro-device

Tested scripts and a benchmark

17 Python scripts do the standard calculations, each tested against cases with known answers: leak-free decoder evaluation, clock alignment, LSL stream checks, stimulation charge and density, ultrasound safety indices, Percept stream rebuilds, noise budgets, impedance fitting, FDA and vulnerability lookups. A 46-task benchmark, graded by code, checks any AI assistant on the mistakes these workflows prevent.

Stimulation and device workflows are research and engineering support, not clinical decisions, regulatory sign-off or certification. Built by Hillock Labs. Corrections and reviews from practitioners are welcome on GitHub.