Update to Fit File Forge

SKILL.md text changes

− removed · + added

--- before
+++ after
@@ -5,14 +5,14 @@
 
 To debrief a session:
 
-1. Find it with `find_activities` (by date, sport or name) unless the user gave one. If more than one fits, ask which.
-2. Call `get_athlete` once per conversation for the athlete's thresholds, units and what they allowed when connecting.
-3. Read it fully: `get_activity` with depth `full` (summary, laps, zones, best efforts), then `get_activity_splits` (by `km` or `mi` for steady sessions, by `lap` for intervals).
-4. If the athlete allowed detailed data, call `get_activity_series` for the channels the question needs (heart rate, pace or power, elevation) and chart heart rate against pace or power and elevation where you can. Compute from the data rather than estimating.
+1. Find it with `find_activities` (by date, sport or name) unless the user gave one. If more than one fits, ask which. Each hit's `capabilities` says what the session has: a `summary_only` one has only Garmin's summary, so there are no laps, zones, splits or series to read; say so and don't infer them.
+2. Use the athlete's thresholds from `get_athlete` when zones or targets matter.
+3. Match the depth to the question. A quick question ("did it go OK?", "what was my pace?") needs `get_activity` at depth `summary`. A full debrief reads `get_activity` with depth `full` (summary, laps, zones, best efforts) together with `get_activity_splits` (by `km` or `mi` for steady sessions, by `lap` for intervals).
+4. For a full debrief, if the athlete allowed detailed data, `get_activity_series` for the channels the question needs (heart rate, pace or power, elevation) gives the detail for a chart of heart rate against pace or power and elevation. `columns_meta` and `coverage` say how each column was built and how complete it is; time-weight with `dt_s`, and use the pace columns rather than deriving pace from `speed_mps`. Compute from the data rather than estimating.
 5. If a workout was planned that day (`get_calendar`), compare what was done with what was planned, rep by rep.
-6. Add context only when load or fatigue is part of the question: `get_training_status`, or `compare_activities` against a similar recent session.
+6. Add context only when load or fatigue is part of the question: `get_training_status`, or `compare_activities` against a similar recent session (its notes say when the two weren't a like-for-like effort).
 
-Read `dataNotes` before interpreting: they say, for example, that cadence is in steps per minute, or that Garmin sent only a summary (then there are no laps, zones, splits or series, so say so and don't infer them).
+Read `metrics_meta` and `dataNotes` before interpreting. `metrics_meta` gives moving and elapsed time and pace on one rule (moving pace for pacing, GAP or NGP for effort on hills) and marks numbers that don't apply to this session (`reliable: false` with a reason, such as efficiency on an interval session): leave those out and say why in one clause. `dataNotes` say, for example, that cadence is in steps per minute, or that Garmin sent only a summary.
 
 Answer in this order: one sentence with the verdict, then the evidence with units (reps, splits, paces, heart rate, power), then one practical suggestion. Use the athlete's own thresholds and zones, never age-based formulas. Fit File Forge has no sleep, HRV or stress data. Pain or injury questions get a training-load answer and a pointer to a professional, never a diagnosis.