| name | personal-health-pulse |
| description | Use when a user wants an agent to ingest personal health check-ins from any supported channel, maintain local structured records, produce warm coaching replies, plan reminders, score days, create weekly reviews, or design a Health Pulse-style workflow for fat loss, training, sleep/recovery, alcohol-risk control, and self-review. |
Personal Health Pulse
Use this skill to help an agent become a personal health "pulse" system: it listens through whatever channel the host agent already supports, records user check-ins into local structured files, and replies with concise nutrition-coaching guidance.
This skill is channel-agnostic. Do not assume any specific chat app, inbox, webhook, or CLI. Treat the channel as an adapter that can receive user messages, send replies, and optionally pass image/file resources.
Core Workflow
- Identify the user's goal, constraints, and preferred data store.
- Create or inspect the Health Pulse data files. Use
references/data-schema.md.
- Route each incoming message:
- simple question: reply directly with recent context;
- record update: write daily/workout/life data, then reply;
- rich review: use image, brand, restaurant, scoring, or weekly-review logic;
- system request: modify the user's Health Pulse project.
- Apply coaching principles from
references/coaching-principles.md.
- Use
references/agent-workflow.md for state, routing, retry, and review lifecycle.
- Use
references/reminder-planning.md when the user wants morning, check-in, evening, weekly, or risk-triggered reminders.
- Use
references/channel-adapter.md to map the current agent channel into the workflow.
Operating Rules
- Keep health records local by default unless the user explicitly asks for a remote store.
- Treat generated advice as coaching and self-review, not medical diagnosis.
- Do not shame the user. Reflect, then offer small next actions.
- Avoid hidden state in conversation memory. Persistent rules, preferences, and long-term facts should be written into project docs/data.
- Prefer structured files over free text when the data will be reused for scoring or summaries.
- Use current tools for channel I/O. If the active environment has a messaging connector, use that connector; otherwise generate the reply for the caller to send.
- Never hard-code user IDs, chat IDs, local usernames, private paths, or personal examples into reusable artifacts.
Common Failure Modes
- Giving one-off advice without writing reusable health state.
- Scoring an open day before late-night food, alcohol, and sleep context exist.
- Turning reminders into noisy lectures instead of small timely interventions.
- Treating brand, menu, or restaurant guesses as exact without reliable evidence.
- Responding to medical, eating-disorder, severe alcohol, or self-harm risk as if it were ordinary coaching.
Message Handling
For each incoming message, decide whether it needs persistence.
- If it reports food, training, sleep, alcohol, body metrics, mood, urges, or body measurements, write to the appropriate data file.
- If it asks "why", "what should I do", or "is this okay" without new recordable facts, answer using recent context.
- If it contains an image or package/menu/restaurant clue, preserve resource metadata and use available vision/search tools only when needed.
- If it mentions a specific brand, packaged product, restaurant, or named dish, verify against reliable sources when possible and record uncertainty.
- If confidence is low, ask one short clarification question instead of pretending precision.
Scoring And Reviews
Use scores as behavioral reviews, not moral judgments.
- Do not automatically score a day until the day is closed: late-night food/alcohol and sleep context should be available.
- Manual score requests may produce a provisional score if data is incomplete; label it clearly.
- Weekly reviews should summarize trends, largest recovery debts, training execution, alcohol pattern, food quality, and next-week focus.
- For detailed scoring and coaching patterns, read
references/coaching-principles.md.
Reminder Planning
When the user wants reminders, design the reminder intent first. Ask about wake time, training windows, work schedule, sleep target, and high-risk food/alcohol windows. Recommend the smallest useful cadence, then let the active agent/platform decide implementation.
Do not assume the environment supports scheduled jobs. If it does, create reminders through the available adapter. If it does not, produce a portable reminder spec or setup instructions. Read references/reminder-planning.md for the full chain.
Resources
references/data-schema.md: recommended local files, fields, and JSONL records.references/coaching-principles.md: ADIME, recovery debt, standard drinks, scoring bands, and tone.references/agent-workflow.md: routing, state lifecycle, light/heavy review, retries, and context windows.references/reminder-planning.md: channel-agnostic reminder design, personalization, and adapter handoff.references/channel-adapter.md: how to integrate this workflow with any agent-supported conversation channel.
