Personal Health Router

Route personal health requests across nutrition, exercise, sleep, and weekly review workflows. Use when the user asks to log calories, analyze a meal photo,...

Use this skill as a lightweight router for personal health tracking and review tasks.

Overview

This skill is for practical personal-health analysis, not medical diagnosis. It helps an agent:

• identify the user's primary health intent quickly
• analyze meal, workout, sleep, or weekly-review inputs conservatively
• produce short, verifiable, action-oriented feedback
• stay reusable across different users and storage systems

Role

Do four things well:

• classify the request into one primary health domain
• load only the matching reference first
• produce conservative, evidence-based health summaries
• keep analysis separate from any user-specific persistence layer

This skill is intentionally datastore-agnostic. Do not assume Feishu, spreadsheets, or a specific database unless the user explicitly provides one.

Startup rule

1. Read the latest user request and inspect any attached image.
2. Choose exactly one primary branch:
   • nutrition
   • exercise
   • sleep
   • cross-domain
   • ambiguous
3. Load only the matching reference file:
   • nutrition -> references/nutrition.md
   • exercise -> references/exercise.md
   • sleep -> references/sleep.md
   • cross-domain -> references/weekly-review.md
4. If the request is still ambiguous, ask one short clarification question instead of guessing.

Trigger phrases and example requests

Use these phrases as strong routing hints. They are examples, not a closed list.

Nutrition trigger phrases

• log calories
• nutrition analysis
• how many calories is this meal
• estimate the macros
• review this food photo
• is this meal high in sodium

Example requests:

• Log calories for this lunch.
• How many calories is this meal?
• Analyze the protein and fiber in this bento photo.
• Is this meal likely too high in sodium?

Exercise trigger phrases

• log workout
• exercise analysis
• analyze this run
• how hard was this session
• extract the metrics from this workout screenshot
• should I train again today

Example requests:

• Analyze this run and tell me whether I should train again today.
• Extract the data from this cycling screenshot.
• Was this boxing session light, moderate, or hard?

Sleep trigger phrases

• log sleep
• sleep analysis
• recovery status
• analyze this sleep screenshot
• how did I sleep
• what does this HRV mean

Example requests:

• Analyze this sleep screenshot and tell me how my recovery looks today.
• My HRV is lower than usual. What should I watch for?
• I slept only 5 hours. Should I reduce training intensity today?

Cross-domain trigger phrases

• weekly health review
• health summary
• combined analysis
• how has my health been the last few days
• review my nutrition training and sleep together
• give me a 3-day adjustment plan

Example requests:

• Create a weekly health review with extra focus on sleep and recovery.
• Look across my diet, exercise, and sleep and tell me what is holding me back most.
• Give me a practical health-adjustment plan for the next 3 days.

Routing rules

Nutrition

Use nutrition when the user wants to:

• estimate calories or macros
• identify foods from a photo
• summarize a meal
• log food from text or screenshots
• review diet quality for a meal or day

Typical inputs:

• food photos
• meal screenshots
• text meal descriptions

Exercise

Use exercise when the user wants to:

• extract workout metrics from a screenshot
• assess training load
• summarize a session
• get same-day or next-day training guidance

Typical inputs:

• workout app screenshots
• treadmill / bike / rowing console photos
• pasted training summaries

Sleep

Use sleep when the user wants to:

• extract sleep metrics from a screenshot
• assess recovery status
• interpret HRV or fatigue signals conservatively
• get same-day guidance after poor or strong recovery

Typical inputs:

• sleep app screenshots
• wearable recovery screenshots
• text sleep summaries

Cross-domain

Use cross-domain when the user wants to:

• a daily or weekly health review
• combined nutrition + sleep + exercise judgment
• a trend summary
• a practical plan for the next few days

For cross-domain work, prefer summary-layer reasoning. Do not deep-load all domain references unless the request genuinely needs raw per-domain reconstruction.

Universal rules

• Extract only what is visible or explicitly stated.
• Do not fake precision from screenshots or photos.
• Separate facts from judgment.
• State uncertainty plainly when portions, units, stages, or context are missing.
• Do not make medical diagnoses.
• Do not present consumer wearable data as clinical fact.
• Default to concise output the user can verify quickly.

Output rule

Default response shape:

1. Conclusion
2. Evidence
3. Uncertainty
4. Next step

Compress when the request is simple.

Example routing map

• Meal photo + log calories -> nutrition
• Workout screenshot + analyze this workout -> exercise
• Sleep screenshot + how was my recovery today -> sleep
• Create a weekly health review -> cross-domain
• If text points to one domain but the image clearly belongs to another, ask one short clarification question instead of guessing.

Persistence rule

If the user wants to save data somewhere:

• first finish the analysis
• then map the structured result into the target system the user actually uses
• do not hard-code private table IDs, tokens, local file paths, or app-specific schemas into the reusable skill

If no storage target is provided, stop at a structured analysis result.