Readiness check

Before a hard session, you want a straight answer: train as planned, modify, or back off. This recipe makes the assistant weigh wellness against current fatigue and give a green/amber/red call — while handling the two things AI assistants get wrong most often: subjective scales and not-yet-synced data. To make those scale and staleness rules stick across Claude chats, add the Claude Project instructions.

When to use this

• In the morning, before a planned interval or long session.
• When TSB is deep negative and you are unsure whether that is productive fatigue or a warning.
• Any day you “feel off” and want the data to confirm or contradict it.

The recipe

Act as my coach and tell me whether I should train hard today. Use icuvisor
with my intervals.icu data.

1. Read my athlete profile.
2. Pull my wellness for the last 14 days: Intervals readiness when present,
   sleep duration, sleep quality, sleep score, HRV, resting heart rate, fatigue,
   soreness, stress, feel, mood, motivation, and any `_native` provider fields.
3. Pull my current fitness, fatigue, and form (CTL / ATL / TSB).
4. If a hard session is planned today, use `get_today` to read planned-event
   `workout_status`, tags, the planned `indoor` flag, weather
   availability/provenance, and any completed warm-up activity tags before
   judging context.

Then:
- Give today's training-readiness guidance in one word — green, amber, or red — and why.
- If the Intervals readiness field is missing or null, say that before using fallback signals.
- Name the one or two signals driving that call.
- Recommend whether to keep, modify, move, or execute indoors/outdoors today, naming the planned-event `workout_status`, any relevant tags, `indoor` flag, weather availability, or activity tags returned by icuvisor.

Rules: sleep quality is a 1-4 scale, sleep score is 0-100 when device-imported,
and feel is a 1-5 scale — use the labels icuvisor returns, do not rescale them
to 0-10. If readiness is missing, do not invent or backfill a readiness score
from Garmin Body Battery or any other native/supporting field; treat HRV,
resting HR, sleep, subjective scales, and `_native` fields as cautious evidence.
If today's wellness row has not synced yet (common with a timezone offset), tell
me the latest data is stale rather than guessing today's values. Do not invent
HRV or sleep numbers. Use weather only when `get_today.weather` or completed
activity weather fields provide it with provenance; if `get_today.weather.status`
is `forecast_unavailable`, say that weather is unavailable from icuvisor rather
than inventing conditions. If you suggest doing the workout indoors instead of
outdoors, keep it as advice or a preview unless I approve a calendar edit; do not
create a second active workout for the same session.

What icuvisor does

analyze_correlation can be added if you want to know whether a wellness metric tracks load over a longer window.

A good answer looks like

Readiness guidance: amber.

Intervals readiness is missing for today (_meta.missing_fields includes readiness), so I am not treating Garmin/native fields as a replacement score. This call uses supporting wellness signals plus current form.

Driving signals: HRV is 48 ms, down from a 14-day average of 58 ms and the lowest in that window. Sleep last night was 6h 10m at quality 2/4 (sleepScore 72/100). Resting HR is +4 bpm, and _native.garmin includes Body Battery min/max on its own 0-100 provider scale.

Form: TSB is -14 (get_fitness) — you are carrying real fatigue.

Recommendation: Modify today’s session. A suppressed HRV plus short, poor sleep on top of a -14 TSB is not the day for VO2 work. Do the planned session as endurance (Z2), or move the intervals to tomorrow and reassess once today’s wellness syncs.

Variations

• Pre-race morning: “…tell me whether I am ready to race today” and ask for the one thing to manage during the event.
• Specific session: paste the planned workout — “I have 4x8min at threshold planned; is today the day?”
• Indoor alternative: “If the outdoor session looks risky from available weather or logistics, suggest an indoor trainer version, but do not change my calendar unless I approve it.”
• Trend only: “Just show me the 14-day HRV, resting HR, and sleep trends with no recommendation.”

Why this prompt works

• Explicit scale reminder. Assistants routinely treat sleep quality as out of 10 and call a 3/4 “poor”. Stating the scale stops a good night being read as a bad one.
• Missing-readiness fallback. Garmin and other providers can expose useful HRV, resting HR, sleep, subjective, and _native wellness fields even when Intervals readiness itself is null. The prompt tells the assistant to explain that absence first and use fallback fields cautiously instead of creating a synthetic score.
• Staleness instruction. intervals.icu wellness often lags by a day because of timezone offsets. Telling the assistant to flag stale data prevents it from inventing today’s HRV.
• One-word verdict first. Forcing green/amber/red keeps the answer decision-shaped instead of a hedge.
• Weather provenance and one active workout. get_today says whether weather is sourced or unavailable, so the assistant can recommend an indoor alternative without making up a forecast. Keeping the alternative as a preview avoids double-counting planned load by placing both outdoor and indoor versions on the active calendar.