Skill instructions
act-on-new-data1.47 KB
View saved version →
---
name: act-on-new-data
description: Use when the user asks ChatGPT to do something each time new data arrives in freddy, such as after each night's sleep, each workout or each weigh-in. Not for a task at a fixed time, and not for a one-time question.
---
# Act on new data
This skill uses the freddy app's MCP tools (list_metrics, query_metrics) and freddy's data.arrived event to act on the user's own recorded wellness data when it arrives. It is a factual presentation of recorded numbers, not medical advice, diagnosis, or a clinical assessment.
Create a task that runs when new data arrives in my freddy account.
1. Call list_metrics once to find the exact metric names and sources for the data I named. Never guess names.
2. Subscribe to freddy's data.arrived event with those metrics (at most 10), with the provider when I named a source. Use match "all" when the task needs each metric for the day, and "any" when one is enough. Do not make a timed schedule in place of the event: the event arrives when the data does.
3. If freddy refuses the subscription, tell me what freddy said. If this chat cannot subscribe to events, tell me that events work in Work chats.
4. When the event arrives, call query_metrics for its day and its metrics, then do the task I asked for with those values.
5. Tell me that the event shows in freddy, under AI, where I can stop it.
6. Present facts only. Do not assess my health status, offer diagnoses, or prescribe changes. Do not mention plans or prices.
build-a-page1.4 KB
View saved version →
---
name: build-a-page
description: Use when the user asks for a chart, a dashboard or a page of charts of the data recorded in their freddy account, or wants to see, change or delete their freddy pages. Not for a one-time answer in the chat.
---
# Build a page of charts
This skill uses the freddy app's MCP tools (list_metrics, preview_chart, list_pages, edit_page, open_pages) to chart the user's own recorded wellness data. It is a factual presentation of recorded numbers, not medical advice, diagnosis, or a clinical assessment.
Build or change a page of charts from my freddy data.
1. Call list_metrics once to find the exact metric names and their sources. Never guess names.
2. To change, open or delete a page I already have, call list_pages first and use its page_id and version.
3. Draw each new chart with preview_chart and show it to me. If I did not say what the page should show, suggest charts for the data I have and ask before you save.
4. Save with edit_page. Send the whole page, with at most 12 charts. The result says what each chart shows: tell me about a chart that has no data.
5. Call open_pages to show the saved page. The page is also in the freddy app and on freddy.coach.
6. Delete a page only when I ask for it, with edit_page and delete: true.
7. Chart only the values that my sources recorded. Do not assess my health status, offer diagnoses, or prescribe changes. Do not mention plans or prices.
connect-a-source1.46 KB
View saved version →
---
name: connect-a-source
description: Use when the user says connect my health source, wants to connect a wearable, ring, CGM, training platform, gym log, Apple Health, or Health Connect to freddy, or asks why a newly connected source shows no data yet. Not for disconnecting or an ordinary data query.
---
# Connect my health source
This skill uses the freddy app's MCP tools (get_profile, list_metrics, query_metrics, connect_source, sync_source). Follow the workflow below.
Connect my health source to freddy.
1. Call get_profile to see what is already connected.
2. If I haven't named a source, ask which app or device I use. Once I answer, immediately call connect_source with that provider; do not guess.
3. If this connection interrupted another health-data question, pass that exact question as original_question. Never add provider credentials, health data, or anything else to that field.
4. Use the interactive card when available. OAuth approval and credentials happen on freddy or the provider; Apple Health and Health Connect finish in the freddy phone app. Keep the tool's text instructions as the fallback, and never ask me to paste credentials into chat.
5. After authorization, the first sync/backfill may take a few minutes. The card checks for real synced data and offers one action to continue the original question. If no card is available, re-check with get_profile or query_metrics after 30-60 seconds; call sync_source only if the source is connected but still stale.
get-started3.27 KB
View saved version →
---
name: get-started
description: Use when the user runs setup for freddy, has just installed freddy, or asks how to get started with freddy. Not for a data question from a user who already uses freddy, and not for disconnecting a source.
---
# Get started with freddy
This skill uses the freddy app's MCP tools (get_profile, list_metrics, query_metrics, connect_source, sync_source) to present the user's own recorded wellness data back to them. It is a factual data summary, not medical advice, diagnosis, or a clinical assessment.
Help me get my first answer from my own freddy data. Start with what I want to know, then set up what is missing.
1. Call get_profile to see what is already connected.
2. If I already asked a data question in this conversation, use it. If I already asked to connect a named app or device, skip this question and go to step 4 with it. Otherwise ask me what I want to know first from my own data. Give two or three short example questions about recent days, such as "How did I sleep last night?" or "How much did I train this week?". If I have no question yet, continue without one.
3. If a source is connected, go to step 6.
4. If no source is connected and I haven't named one, ask which app or device I use for that data. Once I answer, call connect_source with that provider; do not guess. If I asked a question, pass it unchanged as original_question. Never add provider credentials, health data, or anything else to that field. If the tool says that source connects through another app, such as Apple Health or Health Connect, offer that route. If it says the source is not available, ask which other app or device I use.
5. Use the interactive card when available. OAuth approval and credentials happen on freddy or the provider; Apple Health and Health Connect finish in the freddy phone app. Keep the tool's text instructions as the fallback, and never ask me to paste credentials into chat. The first sync may take a few minutes; the card checks for real synced data and offers one action to continue with my question. If I gave no question, or no card is available, ask me to tell you when I have finished, then check with get_profile; call sync_source only if the source is connected but still stale.
6. If a connected source has no data yet, tell me the sync state that get_profile reports. If its first sync is still running, ask me to tell you when a few minutes have passed, then check again with get_profile; call sync_source only if the source is connected but still stale. If get_profile reports an error, or a finished sync with no data, tell me that and do not ask me to wait. Once a connected source has data, call list_metrics once. If I have a question, call query_metrics with the exact metric names for it, and answer from the data the source gave: the values, their dates, and the period they cover. If I have no question, offer two or three questions that my data can answer, and call query_metrics when I choose one.
7. If my data has nothing for my question, say so in one sentence, name the kinds of data I do have, and offer two or three questions that it can answer.
8. Present facts only. Do not assess my health status, offer diagnoses, or prescribe changes; if I want guidance beyond what my numbers show, suggest a qualified professional. Do not mention plans or prices.
long-term-trends1.61 KB
View saved version →
---
name: long-term-trends
description: Use when the user asks how a metric recorded in their freddy account changed over months or years — history charts, progress since a date, seasonal patterns, year-over-year comparisons. Not for a single week (use weekly-health-review).
---
# Long-term trends
This skill uses the freddy app's MCP tools (get_profile, list_metrics, query_metrics, connect_source, sync_source) to chart the user's own recorded wellness data over time. It is a factual presentation of historical numbers, not medical advice, diagnosis, or a clinical assessment.
Show how my freddy data changed over time.
1. Call list_metrics to find the exact metric names — never guess generic English names. Prefer a daily metric that the source itself reports over a per-reading metric; for per-reading data, granularity returns figures that freddy computes from the readings.
2. For multi-month or multi-year windows, call query_metrics with granularity: "month" (or "week" for finer detail) and explicit start / end dates (YYYY-MM-DD). This returns avg/min/max/count buckets so years of data fit in one response.
3. If a response reports truncation or omitted buckets, follow its continuation hint — it names the exact start and end to pass next.
4. Present the numbers factually: the overall direction, notable changes with their dates, seasonal patterns if visible, and — when enough history exists — the most recent 3 months side by side with the same period a year earlier. State which aggregation level the numbers come from. Do not characterize the data as a health status and do not draw medical conclusions from it.
weekly-health-review1.68 KB
View saved version →
---
name: weekly-health-review
description: Use when the user asks for a weekly summary or check-in of the sleep, heart, training, or body data recorded in their freddy account. Not for multi-month or multi-year questions (use long-term-trends) or single-metric lookups.
---
# Weekly health review
This skill uses the freddy app's MCP tools (get_profile, list_metrics, query_metrics, connect_source, sync_source) to present the user's own recorded wellness data back to them. It is a factual data summary, not medical advice, diagnosis, or a clinical assessment.
Summarize my week from my freddy data.
1. Call get_profile to see which sources are connected and whether a sync is running, then list_metrics once to learn the exact metric names available for my account.
2. Fetch the most recent 7 full days with explicit start / end dates (YYYY-MM-DD; end = yesterday, start = 6 days before that) for the areas that exist in my data: sleep (duration, stages), heart data (HRV, resting heart rate), training (workouts, load), and body measurements (weight) if present.
3. Fetch the 7 full days immediately before that window, again with explicit start / end dates, as the comparison baseline — two equal-length, non-overlapping weeks.
4. Present a factual side-by-side summary: how this week's numbers compare with the previous week, with the actual values. Call out gaps — days without data or a source that stopped syncing. Do not assess health status, offer diagnoses, or prescribe changes; if the user wants guidance beyond what their numbers show, suggest they consult a qualified professional.
Stay grounded in the data returned — no invented values, and no conclusions from metrics that were not fetched.
