running-plan-kit

The tooling behind stacktrace.run — an injury-first, heart-rate-capped running program built and coached in conversation with an AI, synced to intervals.icu and a Garmin watch, and published as a blog. Background reading: how this works.

What's here

prompt-template.org      The intake prompt. Fill in YOUR data, paste it into
                         an AI conversation (e.g. Claude), get your own plan.
                         This encodes the whole coaching methodology.
scripts/
  sync_plan_intervals.py   Plan -> intervals.icu structured workouts; pulls
                           completed runs, daily wellness (RHR/HRV/sleep) and
                           an org agenda back; strength load from RPE.
  sync_strength_garmin.py  Strength sessions as native Garmin watch workouts
                           (animations, rep counting), scheduled on the
                           calendar with race-weekend skips.
  generate_training_pages.py  Turns the org data into blog pages: run log
                           with per-run execution analysis, program calendar,
                           live week table, per-run posts. Hugo-flavored.
skills/                  Claude Code skills wrapping the workflows. Copy the
                         directories into your project's .claude/skills/ and
                         you get /sync-runs, /plan-sync, /garmin-strength,
                         /update-blog.

Quick start

1. Read prompt-template.org → fill the YOUR INTAKE section with real data (a Strava/intervals.icu export beats guesses) → paste the whole file into an AI conversation. Answer its follow-ups honestly. You'll get a plan document plus the org files these scripts expect.
2. Skim each script's header — configuration is environment variables (INTERVALS_API_KEY or Emacs auth-source, INTERVALS_EASY_MIN, shift knobs, …). They assume your plan lives in ~/org.
3. Optional: install the skills and let your AI drive the loop — "I ran today" → pull, analysis, blog update.

Notes

• Built for one runner's setup, in conversation with an AI, and shared as-is: read before you run, adapt freely.
• No credentials live in this repo or these scripts — keys come from your environment; the Garmin OAuth token stays in your home directory.
• This produces a self-coached training plan, not medical advice. It is deliberately conservative and defers to a physio/physician for anything that looks like a real injury.