Strong naming and descriptions across 14 tools. All tools follow verb_noun pattern (garmin_get_*, garmin_list_*). Descriptions are detailed (150-250 chars typical), explaining WHAT, WHEN, and WHY to use each tool. Parameters have clear types and descriptions. However, output schemas are not explicitly documented in the source code, only input schemas are visible. Error handling is minimal; no recovery guidance or actionable error messages. Tool composition is excellent (single responsibility, proper chaining via IDs). Missing: output schema documentation, error classification, and per-tool error handling patterns.
Call any Garmin Connect API path directly. This is the escape hatch for the ~130 endpoints without a dedicated tool (SpO2, respiration, VO2 max, race predictions, gear, workouts, badges, and so on). Prefer a dedicated tool when one exists -- their output is easier to read and their arguments are validated. Paths look like '/usersummary-service/stats/daily/2026-08-01/2026-08-03'; the response is returned as-is.
Get the full record of one recorded workout by its activity ID: type, timing, distance, pace and speed, heart rate zones, cadence, power, elevation, calories, training effect and recovery time. Find the ID with garmin_list_activities first.
Get body battery energy levels across a date range: the daily high and low, plus charge and drain totals. Body battery rises with rest and falls with activity and stress, so a multi-day range shows whether energy is trending down faster than it recovers.
Get weigh-ins over a date range: weight, BMI, body fat and water percentage, muscle and bone mass, plus the range averages. Only returns data for days with a recorded weigh-in, so an empty result usually means no measurement rather than an error.
Output schemas not documented. Source shows input schemas (date, detail params) but no explicit documentation of what each tool returns. LLMs cannot plan downstream calls or extract fields without knowing response structure.
No error handling or recovery guidance. Tools lack error classification (retryable vs user-fixable vs fatal) and actionable error messages. If a date is invalid or API fails, LLM receives no guidance on next steps.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Get the all-round wellness summary for one day: total and goal steps, distance, floors, calories (active, resting, consumed), intensity minutes, resting heart rate, average stress, and body battery high/low. Start here for broad questions like 'how active was I on Tuesday' -- it answers in one call what would otherwise take several.
Get an all-in-one recovery picture for one day: daily summary, sleep, HRV, body battery and training readiness together. Prefer this for open questions like 'how did I sleep', 'how am I recovering' or 'should I train today' -- it answers in one call what would otherwise take five, and each metric is trimmed to its headline numbers. Metrics your device does not record are reported as errors within the result rather than failing the whole call.
Get heart rate for one day: resting, minimum and maximum heart rate, and the last seven days' resting average. Use detail='full' for the continuous all-day heart rate series. For heart rate during a specific workout, use garmin_get_activity instead.
Get overnight heart rate variability for one day: the weekly average, last night's average, the personal baseline range, and Garmin's HRV status (balanced, unbalanced, low, poor). HRV is the most sensitive single indicator of accumulated stress or illness, so prefer this when asked about recovery trends rather than a single night's sleep.
Get sleep for the night ENDING on the given date: overall sleep score and its component ratings, time in deep/light/REM/awake, sleep and wake times, average overnight SpO2, respiration and resting heart rate. Ask for 'today' to get last night. With detail='full' this also returns per-minute sleep stage, movement and heart rate series, which are large.
Get daily step totals, step goal and distance for each day in a range. Cheap and compact -- prefer this over repeated daily-summary calls when the question is only about step counts over time.
Get stress for one day: the average and maximum stress score, plus how the day divided between rest, low, medium and high stress. Use detail='full' for the underlying per-interval stress series.
Get Garmin's training readiness score for one day (0-100) together with the factors behind it: sleep score and history, recovery time, HRV status, acute training load and stress history. This is the tool for 'should I train hard today' -- it is Garmin's own synthesis of the recovery signals rather than a raw metric.
Get the account profile: name, birth date, height, weight, gender, activity level and -- importantly -- the measurement system. Worth checking once before interpreting units in other tools, since Garmin returns metric values regardless of what the app displays.
List recorded workouts in a date range, newest first, with activity ID, type, name, start time, duration, distance, pace, average and max heart rate, calories, elevation and training effect. Use this to find an activity, then pass its activityId to garmin_get_activity for the full detail of a single session.
Date parameter accepts relative strings ('today', '-7d', '-2w', '-3m') but no validation or format constraints documented. LLMs may pass invalid formats like '-7days' or '7d' without guidance.
detail parameter enum ('summary' vs 'full') lacks guidance on when to use each. Descriptions mention 'full' returns large payloads but do not explain the token cost or when summary is insufficient.
garmin_api_request is a generic escape hatch with minimal validation. Accepts arbitrary API paths and params. No guidance on valid endpoints, error handling, or rate limits. LLMs may construct invalid paths or trigger rate limiting.