garmin-coach

Unofficial and unaffiliated. Not affiliated with or endorsed by Garmin. This is a documentation-only Claude Skill — it ships no code that touches your account and makes no network calls of its own.

A Claude Skill that teaches Claude how to correctly read and reason about data from garmin-mcp — a companion MCP server exposing a Garmin Connect account's activities, sleep, HRV, body-weight trend, and more.

This skill is deliberately just the "how to read the data" layer — which tool answers which question, what each field's units and nulls mean, which numbers are directly comparable and which aren't (e.g. HR zone minutes across a zone-config change). It contains no training philosophy, thresholds, or programming rules of its own. Your training plan lives in the conversation with Claude, not baked into this skill — see "The one rule that matters" in SKILL.md.

Why this exists

MCP tools return raw, correct data — but a model reading hr_zones minutes across two activities with different zone_config_ids, or treating avg_hr: null as 0, or comparing get_training_load's TRIMP figure to Garmin Connect's own displayed number without noting they're not the same calculation, will confidently produce wrong conclusions from correct inputs. This skill exists to close that gap — it's tool documentation written for an LLM to act on, not for a human to read once and remember.

Requirements

• A running instance of garmin-mcp, connected to Claude as a custom connector.
• Claude with Skills support (claude.ai, Claude Code, or the Claude app).

Installing

1. Set up garmin-mcp first (see its README) and connect it to Claude.
2. Install this skill via Claude's Skills UI (or drop SKILL.md into your skills directory, depending on how you're running Claude).
3. Ask Claude something like "how should I train today" or "log my weight, 182.4" — it should reach for the right garmin-mcp tool on its own.

Evals

evals/evals.json has a small set of scenario-based checks (does the skill call the right tool, does it correctly flag a null/edge case rather than guessing) — useful if you're modifying SKILL.md and want a sanity check that it still triggers and reasons correctly.

Related

• garmin-mcp — the MCP server this skill is written against. Keep the two in sync: if a tool's return shape changes there, SKILL.md here needs the matching update (this has bitten the project once already — see its git history).