v0.3.1

Added

• --db option and MI_FITNESS_DB_PATH environment variable to override the SQLite database location for sync, export, serve, and doctor; precedence is CLI option > environment variable > default platform location. This enables test and multi-database isolation, which the platformdirs default cannot provide (it ignores LOCALAPPDATA on Windows).
• "How to get user_id and passToken" guidance in the README configuration sections (browser-cookie method and QR-login tooling), referenced from doctor output when no configuration file is found.
• Git Bash virtualenv activation command in the installation instructions.
• Export-sensitivity note in docs/export-format.md and the README privacy sections: exports contain no passToken but do carry plaintext user_id identifier columns and must be handled as sensitive personal data.
• Warn after successful CLI exports when selected datasets contain zero rows; exported JSON/CSV bytes are unchanged (#7, thanks @pollychen-lab).

Changed

• load_config() no longer writes a default config.json on first run; configuration is persisted only by the explicit setup command.
• README test-suite counts updated to the current 46 tests.

v0.3.0 — agent-safe-series/v1 workout_series contract

Added

• Agent-safe workout_series MCP tool: auto-downsamples workout time series under a hard max_points cap (default 400, max 500) using fixed time-bucket means aggregated in SQLite, reports honest downsampled/source_points/returned_points/method metadata, and always includes full-resolution stats (avg/min/max/quantiles) and heart-rate time-in-zone.
• data_quality field (coverage days, missing metrics, last sync time) on query_workouts and get_daily_summary responses.
• Synthetic 3-hour ride fixture (10,800 1 Hz heart-rate points with known ground-truth stats) and regression tests for the downsampling pipeline.
• workout_series contract agent-safe-series/v1: top-level start_time, t_unit (seconds_from_start), unit, contract_version, and requested_resolution_seconds (alongside the effective resolution_seconds); stats.percentile_method (linear_interpolation); time_in_zone.reference_source (activity_recorded_max / observed_max / caller_provided); data_quality.actual_samples, data_quality.sample_interval_seconds, and data_quality.coverage_anchor.
• Optional reference_max_hr input on workout_series so agents can normalize time-in-zone against a consistent reference when comparing activities.
• Duration-anchored coverage in workout_series: expected_samples is computed from the activity's nominal duration (workout duration_minutes, else the recorded start/end), so samples missing at the start or end of an activity surface as coverage_ratio < 1.0 instead of looking like a shorter, fully-sampled workout; falls back to the first-to-last sample span (coverage_anchor: "sample_span") when no nominal duration is recorded.
• Garmin-layout counterpart of the synthetic 3-hour ride fixture (tests/garmin_fixtures.py, metricDescriptors/activityDetailMetrics shape) plus cross-format regression tests proving both layouts yield identical stats and downsampled points, including a sensor-gap scenario (garmin-mcp issue #19).

Changed

• Breaking: workout_series points[].t is now a numeric offset in seconds from start_time instead of an ISO timestamp string.

• Hide passToken entry in the interactive setup flow.

• Close the experimental cloud adapter reliably after CLI diagnostics and synchronization.

• Use cn consistently as the default cloud region.

• Reject malformed or reversed export date ranges before reading the local database.

• Apply configured HTTP limits, retries, pagination, sync chunking, and operation timeouts to CLI diagnostics and synchronization.

• Reject invalid synchronization lookback and chunk sizes instead of risking non-terminating chunk loops.

• Report partial and failed CLI synchronization results accurately instead of displaying them as successful.

• Return a non-zero process status when diagnostics are not ready or any synchronization is partial/failed.