# Start Cold - a layer plan for one fall or winter hike, from your forecast and the clothes you own

> Cold-weather hiking layering planner. Enter the trailhead forecast (temperature, wind, what is falling), the elevations, the hours, the effort and whether you run cold or hot, and list the clothes you own one per line ("Merino 150 long sleeve, 170g", "Down puffy jacket", "Rain shell with pit zips"). The page works out four phases - leaving the trailhead, the climb, the stop at the top, the descent - and stacks your layers for each, free in the browser. A metered plan adds what to wear per phase, swaps, missing gear, a verdict (ready, adjust, rethink), a turn-back line and trail tips.

URL: https://start-cold.skillsafe.ai/
API: https://start-cold.skillsafe.ai/api.html

Built on the SkillSafe skill @mvanhorn/pp-weather-goat (turn a forecast into an activity recommendation: can I hike given this weather, and what does it take).

## What runs free in the browser (layerkit.js, stackview.js)

- Phases: start (trailhead air and wind), climb (the average of trailhead and top), top stop (top air and wind, not moving), descent. Wind chill is the North American index (metric form, applied at or below 10 C with wind of 4.8 km/h or more).
- Estimates, always labelled: the temperature at the top when not given (6.5 C per 1000 m of climb), the wind at the top when not given (1.5x the trailhead), garment weights a line does not state, and warmth points (about one point per 6 C of shortfall; base layer 0.6, light fleece 1.0, down jacket 2.5, rain shell 0.5).
- The draft stack: start cold (the trailhead stack is the climb stack), the puffy for the stop, the shell on when it is wet or windy, no cotton when anything else fills that layer, hat, gloves, neck and leg layers by threshold.
- Flags: conditions (freezing rain, a top wind chill below -25 C, top wind above 65 km/h), must (cotton worn, no waterproof jacket on a wet day, a phase the closet cannot warm, no warm hat or gloves for a cold top, cotton-only legs or socks), nice (no puffy, no traction on snow, no spare base on a long wet day, no headlamp for a late finish).
- Exports: Markdown, a closet CSV (id, item, layer, materials, grams, estimated, bring, worn per phase) and a pack list (wear from the car / in the pack).

## The metered task (model gpt-terra)

- `kit` - inputs `today`, `units` (us or metric), `hike`, `phases`, `closet`, `page_reads`, `page_verdict`, optional `notes` (object fields JSON-encoded as strings). Returns `title`, `verdict` (ready|adjust|rethink), `summary`, `phases` [{phase: start|climb|top|down, wear: [closet ids], why, vent}], `swaps` [{out, in, why}], `gaps` [{slot, item, why, must}], `turn_back`, `trail_tips` (three to five), `caveats`.
- The page's check: ids must be in the closet; cotton the plan still wears without a swap is called out; a wet forecast with the waterproof jacket left off the climb or the top is flagged; a top stop dressed colder than the page's need (when the closet can do better) is flagged; a "ready" over a must-fix the plan did not address, or anything but "rethink" over a conditions flag, is flagged.
- House rules: only the input's numbers, in the input's units; no brands, prices or shops; never calls a hike safe and never suggests leaving the shell, warm layer, hat or gloves behind on a cold or wet day.

## Notes

- Not a forecast, an avalanche bulletin or a trail report. Check those before you go. Sample hikes are made up.
- Saved hikes go to the signed-in user's own collection (`hikes`), only when they press save.