NeuroKit2 Physiological Research Workflow

What it does

Guides Claude through method-aware, reproducible biosignal research using NeuroKit2 pinned at 0.2.13.

• Enforces a data contract: sampling rate, physical units, clock origin and drift, missing/saturated/flat samples, and whether event onsets are samples or seconds.
• Locks preprocessing order: preserve raw → verify time base → segment at gaps → clean per modality at native rate → detect peaks/decompose → inspect quality → log peak corrections → derive features → align to a common grid → epoch and analyze.
• Treats schemas as runtime observations: records returned columns and info keys together with version, method, and sampling rate instead of assuming a universal column list.
• Ships six bounded CLI helpers: synthetic fixture generator, signal inspector, ECG-HRV pipeline, EDA pipeline, sample-exact epoch planner, and multimodal manifest validator. They reject URLs, symlinks and path traversal, never pickle, and require --deidentified for real data.
• Interpretation guardrails: no LF/HF-as-sympathovagal-balance claims, PPG PRV ≠ ECG HRV, no VLF/ULF interpretation from short records, RSA is not context-free vagal tone.

Who it's for

• Psychophysiology, cognitive neuroscience and wearable researchers working with ECG/HRV, EDA/SCR, respiration, EOG/EEG.
• Graduate students and lab engineers who need reproducibility: pinned versions, recorded environments, auditable exclusion logs.
• Data scientists reviewing or auditing someone else's NeuroKit2 code.

Not for diagnosis, patient monitoring, or medical-device validation. Keep PHI out of prompts, logs, and fixtures.

Examples

1. Pre-flight before HRV: "Inspect this 250 Hz ECG CSV, then compute time, frequency and nonlinear HRV." → inspect_signal.py flags non-monotonic timestamps and flat runs, then ecg_hrv_pipeline.py runs with duration warnings for VLF/ULF metrics.
2. Explicit EDA decomposition: the skill clarifies that default eda_process(method="neurokit") uses high-pass tonic/phasic separation (not cvxEDA), that cvxEDA needs cvxopt, and that amplitude_min is relative, not an absolute µS threshold.
3. Multimodal alignment audit: validate a strict local manifest with validate_multimodal.py before calling bio_process(), avoiding index-concatenation NaNs and checking that synchronized ECG+RSP exist before trusting RSA output.

Install the skill I found on Claude Skill Mart.
Copy the skills/neurokit2 folder from the GitHub repo K-Dense-AI/scientific-agent-skills into my ~/.claude/skills/neurokit2/.
When it's done, tell me in one line what this skill can do.