weather-nyc-forecast - SKILL.md Agent Skill

name: weather-nyc-forecast description: Composite 10-day Manhattan weather report that scrapes the Google Weather card, calls OpenWeather, WeatherAPI, and optionally Visual Crossing, then writes a single markdown summary. Use when you need a multi-source picture of Manhattan/NYC weather and want the report generated automatically on a schedule (every 6h, cron, or a runner job).

Weather NYC Forecast

Overview

Collects the Google Weather card, OpenWeather One Call, WeatherAPI forecast, and (if you supply it) Visual Crossing.
Blends highs, lows, rain chances, and humidity to build a markdown table + per-day narrative.
Designed to produce a reusable forecasts/manhattan-weather-forecast.md that you can push into dashboards, Telegram, or a status page.

Quick start

Install runtime dependencies (inside the environment that will run the cron job):
```
pip install requests beautifulsoup4 pytz
```

Export the API keys:

export OPENWEATHER_API_KEY=...
export WEATHERAPI_KEY=...
export VISUALCROSSING_API_KEY=...  # optional but gives better 10-day coverage
export GOOGLE_CONSENT_COOKIE=YES+1  # optional, allows scraping if Google redirects to consent

Run the script manually to verify:
```
cd /home/gianpaj/github/openclaw
python skills/weather-nyc-forecast/scripts/fetch_manhattan_forecast.py --output forecasts/manhattan-weather-forecast.md
```
The file is created with a table plus per-day details; existing files are overwritten so the latest 10 days are always in one place.

Scheduling (every 6 hours)

Use cron or your orchestration tool to run the script every six hours. Example cron entry:

0 */6 * * * cd /home/gianpaj/github/openclaw && /usr/bin/python skills/weather-nyc-forecast/scripts/fetch_manhattan_forecast.py --output /home/gianpaj/github/openclaw/forecasts/manhattan-weather-forecast.md >> /home/gianpaj/github/openclaw/logs/weather-nyc.log 2>&1

If you prefer systemd timers or a runner (GitHub Actions, CI, cron runner), schedule the same command on a 6-hour cadence and point --output to the canonical markdown path.

Output and consumption

The markdown file includes a header, summary table, and detailed bullet list per day.
Use the table for quick ingestion (Slack/Telegram summary) and the per-day section for publishing long-form announcements.
When integrating with other automation, read the table header to confirm the file was updated within the last 6 hours.

Key variables

Environment variable	Purpose	Required?
`OPENWEATHER_API_KEY`	OpenWeather One Call v3 credentials	✅
`WEATHERAPI_KEY`	WeatherAPI 10-day forecast access	✅
`VISUALCROSSING_API_KEY`	Optional cross-check for day-by-day conditions	⚪️
`GOOGLE_CONSENT_COOKIE`	Set to `YES+1` if Google redirects to consent; allows scraping the weather card	⚪️
`GOOGLE_USER_AGENT`	Override the UA string the scraper sends	⚪️

Troubleshooting

Google often returns the consent gate (/ml?continue=...). If you continuously see Google weather fetch failed, set GOOGLE_CONSENT_COOKIE=YES+1 (or copy the real consent cookie from the browser) and rerun.
Missing API keys will skip each provider but the script still tries to aggregate whatever data is available.
Logs are emitted at INFO level by default but you can pass --quiet to suppress them in the cron job.

Script reference

scripts/fetch_manhattan_forecast.py: fetches from every provider, normalizes temperatures to °F, averages highs/lows/pop/humidity, and writes forecasts/manhattan-weather-forecast.md. Read it if you want to add another provider, change formatting, or send the markdown to Slack/Telegram after generation.