Supervised recompute of stored ACWR history in fitnesshistory.json under the rolling 7d:28d primary model — dry-run review, backup, then apply. Use when historical snapshot ACWR values need recomputing (e.g. after a model change or backfilled loads), never as a routine step.
Refresh the real Garmin API test fixtures (testfixtures.json) — an OWNER-RUN-ONLY step agents and CI must never execute. Use when Garmin response shapes changed, fixture-driven tests look stale, or the user asks to recapture fixtures.
Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).
Add or change an MCP tool in coach-mcp the repo way — parser, annotations inventory, clock-discipline allowlist, FakeGarminClient tests, doc hygiene, e2e goldens. Use when creating a new @mcp.tool, renaming/consolidating tools, or changing snapshot/push behavior.
Cut and publish a garmin-coach-mcp release — version triple-bump, tag push (CI publishes PyPI + GitHub release), MCP registry publish, and the two-account git push auth dance. Use when the user asks to release, publish, cut a version, or bump the package.
Adaptive AI training coach MCP server backed by Garmin Connect. Runs locally from the garmin-coach-mcp Python package. Needs 3 environment variables to run.