Discover patterns in health data, answer questions about correlations, and guide structured self-experiments with observation, hypothesis, check-ins, analysis, and next-step recommendations.
Use this skill when:
/insights (legacy shortcut)Pattern discovery mode:
When the user asks about patterns or correlations, proactively analyze all available local data (Apple Health metrics, nutrition logs, experiment check-ins) to find correlations. Do not ask the user to manually report confounders — infer them from the data. Present findings as specific, data-backed observations, for example:
This proactive pattern discovery from data is a core differentiator. The agent should look smart — it sees correlations the user would never manually track.
Experiment mode:
Rules:
Start every /insights session by calling the experiments tool:
{ "command": "gap_report" }
If the user wants to start an experiment:
title, domain, hypothesis, null_hypothesis, intervention, primary_outcome, and optional secondary_outcomes, windows, and questions.experiments tool:{
"command": "create",
"input_json": {
"title": "...",
"domain": "...",
"hypothesis": "...",
"null_hypothesis": "...",
"intervention": "...",
"primary_outcome": "..."
}
}
For a daily check-in:
experiments tool:{
"command": "checkin",
"input_json": {
"experiment_id": "<id>",
"compliance": 0.9,
"primary_outcome_scores": { "metric": 7 },
"confounders": [],
"note": "..."
}
}
For experiment review, call the experiments tool:
{ "command": "analyze", "experiment_id": "<id>" }
When the script says more data is needed:
/insights after enough data exists