Skill guide: /analyze — inspect, plan, analyze, interpret

data-analysis is the end-to-end analysis skill for an uploaded dataset — a spreadsheet, logs, a recording or a video. It composes twelve method skills and always follows the same order: inspect, plan, analyze, interpret.

How to call it

text

/analyze Inspect the uploaded pss_trial.xlsx (two groups, PSS-10 pre/post). Check data quality and baseline balance, compute change scores, and deliver a reproducible analysis report with plots.

Attach the file with the + button first; it stays in the session for later steps.

What happens

Inspect: structure, missing values, ranges, group balance — reported before anything is computed.
Plan: an explicit analysis plan you can read (and object to).
Analyze: code runs in the sandbox, each step visible and expandable; every artifact records the code and data snapshot behind it.
Interpret: a report with data quality, plan, assumption checks, results with effect sizes and CIs, interpretation and limitations.

Six artifacts from one message; the trajectory plot with group means and 95% CIs.

Method skills it composes

descriptives · correlation · regression (linear · logistic) · mixedlm (mixed-effects) · abtest
transcription (local whisper) · acoustic (formants · MFCC) · voice (emotion features)
face_expression (image · video) · pose (image · video) · rppg (heart rate from face video) · eeg_suite (preprocess · band power · topomap · ERP · ERD/ERS · SSVEP)

A video in, facial action units out — the same skill handles recordings and biosignals.

A usability A/B export in, a four-panel dashboard out — SUS, NASA-TLX, task time and errors.

What you get

The analysis report (Markdown) and a structured AnalysisReport (JSON).
Plots and tables as artifacts, each with its Reproducibility panel (code, data, seed).

Tips

Say what the columns mean; the plan gets better and the report names the variables correctly.
For hypothesis tests with assumption audits and power, follow with /stats — the file is already there.