setup

peak-lab setup

You are bootstrapping the peak-lab plugin for a new user. Run through the following checklist, pausing for user input where needed. Be concise — each step is one or two sentences of output unless something is wrong.

1. Verify uv is installed

Run command -v uv via Bash. If missing, tell the user to install it (curl -LsSf https://astral.sh/uv/install.sh | sh on macOS/Linux) and stop here. Do not attempt to auto-install.

2. Verify Garmin Connect tokens exist

Run "${CLAUDE_PLUGIN_ROOT}/bin/garmin-auth" --verify via Bash. Exit code 0 means tokens are valid — skip to step 4. Non-zero means we need to log in (step 3).

3. Walk the user through one-time Garmin Connect authentication

Tell the user to run this in their terminal and wait for confirmation:

"${CLAUDE_PLUGIN_ROOT}/bin/garmin-auth"

(They can copy the absolute path from the message you printed.) The script prompts for email, password, and MFA code if needed, then saves DI OAuth tokens to ~/.garminconnect/garmin_tokens.json. peak-lab never sees the password — only the saved tokens.

If they hit a 429, tell them to wait 15-60 minutes and try again — Garmin throttles by IP. Do NOT retry repeatedly; that extends the cooldown.

After the user confirms the script printed ✓ Login successful, re-run bin/garmin-auth --verify to confirm. If verify still fails, surface the error and stop.

4. Capture the athlete profile

peak-lab works in the current directory. Whatever directory the user ran Claude Code from IS their training project — profile.md, runs/, and garmin-call.log all live right there. No env vars, no overrides, no git-repo-root guessing. If the user wants a different project, they cd and relaunch.

Sanity-check the cwd in one Bash call:

pwd
test ! -d .claude-plugin || { echo "ERROR: $(pwd) is the plugin source directory — cd into a separate training project folder"; exit 1; }
test -f profile.md && echo "PROFILE_EXISTS" || echo "NO_PROFILE"

If the output includes .claude-plugin/ at the current directory, stop and tell the user to cd into a separate training project folder and relaunch.

If PROFILE_EXISTS, read profile.md and print a short summary, then ask the user whether to keep or update it.

If it doesn't exist, keep the ask short. Sports, training hours, and recent race history can be inferred later from the Garmin data pulled in /peak-lab:analyze — do NOT ask for them up front. Only ask for what's genuinely self-reported and unknowable from the data:

1. Current training goals (required, free-form) — e.g. "build base then race a sub-4 marathon in October", "first 70.3 in June".
2. Known injuries or constraints (required, can be "none") — historical issues count too, with notes on whether they're active or latent.
3. Target race time for the A race (optional, but recommended if a race is set) — e.g. "marathon target 4:00:00". This drives VDOT-derived training pace zones when there's no recent real race to anchor on.
4. Upcoming races beyond the A race (optional) — peak-lab also auto-discovers races from Garmin Connect's calendar via get_scheduled_workouts in the analyze flow, so only ask if the user wants to add something that isn't already in Garmin.

Do not ask about sports, typical training hours, or recent race results — those come from the Garmin pull. Explain this briefly so the user understands why the profile is so short.

Write the collected answers to profile.md (in the current directory) using the Write tool. Use this structure:

# Athlete profile

_Last updated: <YYYY-MM-DD>_

## Goals
...

## Constraints / injuries
...

## A race — target time
<e.g. "City Marathon YYYY-MM-DD — target H:MM:SS"; omit the target section if none set>

## Additional races (beyond what Garmin's calendar contains)
| Name | Date | Priority |
| ---- | ---- | -------- |
| ... | ... | ... |

## Auto-derived at analyze time
- Sports: inferred from `activities[]` in `raw.json`
- Training volume: computed from 4-week and 8-week rolling windows over `activities[]`
- Recent race fitness: derived from `personal_records` or race-flagged activities

5. Confirm and hand off

Print: ✓ peak-lab is ready. Run /peak-lab:analyze to generate your first training analysis.

Notes for you, the orchestrator

• The training project IS the current working directory. Don't invent env vars, don't consult git, don't look anywhere else.
• When you use the Write tool to create profile.md, pass an absolute path (e.g. /path/to/my-project/profile.md) — the Write tool expects absolute paths. Use $(pwd)/profile.md resolved to its literal value.
• Never print the user's Garmin password or token contents.
• If any step fails, surface the exact error and do not proceed.
• Data lives in the user's cwd. This is intentional: peak-lab treats the project folder as the single source of truth for the athlete's training knowledge.