Simmer Weather Trader

Automated weather prediction market trading skill for Simmer/Polymarket. Cross-references 4 weather sources (NOAA, Open-Meteo, Wunderground, NVIDIA FourcastN...

Security Scan

Overall the skill mostly matches its stated purpose (weather-driven trading) but contains multiple implementation inconsistencies that could cause unexpected behavior (missing dry-run gating, mismatched dependency / env documentation, scheduler metadata), so proceed with caution.

Purpose & Capability

Name/description (Simmer weather trading using NOAA/Open-Meteo/Wunderground/FourcastNet) aligns with the code: the package fetches Simmer markets, queries three weather APIs + NVIDIA FourcastNet, scores and executes trades. Requested credentials (SIMMER_API_KEY, NVIDIA_API_KEY, TELEGRAM_BOT_TOKEN) are expected for these integrations. Minor mismatch: clawhub.json lists pip deps (simmer-sdk, numpy, netCDF4) that are not all present in simmer_weather_bot/requirements.txt.

Instruction Scope

SKILL.md claims the bot defaults to dry-run and that --live is required for real trades, and marks TELEGRAM_BOT_TOKEN as optional. In the provided code: config.py unconditionally reads TELEGRAM_BOT_TOKEN (os.environ[...]) which will crash if not set, and I see no CLI parsing or handling of a --live flag in main.py or the visible files — execute_trade posts directly to the Simmer trade endpoint. That means the claimed dry-run safeguard is not obviously implemented in the visible code, so the bot may be capable of making live trades when run as-is. The code also instructs/executes heavy operations (Playwright scraping, netCDF zip extraction, writing temp files) — these are consistent with stated functionality but worth noting for resource/anti-scraping implications.

Install Mechanism

This is instruction-and-code-only (no explicit install spec). The repo expects pip packages and Playwright/Chromium. There are inconsistencies between clawhub.json (pip list including simmer-sdk, httpx, python-telegram-bot, numpy, netCDF4) and simmer_weather_bot/requirements.txt (python-telegram-bot, httpx, playwright, python-dotenv). The SKILL.md lists additional pip installs (netCDF4, playwright, playwright install chromium). No external arbitrary URL downloads were detected, but Playwright requires a browser install step which the SKILL.md documents.

Credentials

Requested secrets (SIMMER_API_KEY and NVIDIA_API_KEY) are proportionate to trading + FourcastNet use. However SKILL.md labels TELEGRAM_BOT_TOKEN optional while config.py forces it (will raise KeyError if absent) — an incoherence. The code reads only those env vars declared (and some optional SIMMER_BASE_URL, SIMMER_VENUE), so no unrelated credentials are requested, but the mandatory-vs-optional mismatch is problematic because missing token will crash startup or the token may be forced into the environment unexpectedly.

Persistence & Privilege

always: false (no blanket always-on privilege). However clawhub.json includes a cron schedule (*/30 * * * *) and automaton.managed: true, which indicates the skill is intended to be scheduled/managed by the platform every 30 minutes. For a trading bot that is expected, this is reasonable — but combined with the missing dry-run enforcement it increases the blast radius (scheduled runs could execute trades).

What to consider before installing

This skill appears to implement the trading functionality it claims, but several inconsistencies are concerning and should be resolved before installing or granting keys. Key points to check before use: - Dry-run gating: The README/SKILL.md state the bot defaults to dry-run and requires --live for real trades, but I could not find CLI parsing or a --live check in the visible code and execute_trade directly calls the Simmer trade endpoint. Ask the author for the exact mechanism that prevents live trades by default; do not provide a live trading key until you confirm a working dry-run switch. - TELEGRAM_BOT_TOKEN mismatch: SKILL.md marks Telegram token optional, but config.py unconditionally reads TELEGRAM_BOT_TOKEN (will raise an exception if absent). If you don't want Telegram, request a code change to make that env truly optional or provide a dummy token in a safe test environment. - Dependencies mismatch: clawhub.json, SKILL.md and requirements.txt disagree on which Python packages are required (numpy, netCDF4, simmer-sdk appear in some manifests but not others). Ensure the install process you run installs the exact packages required (netCDF4, numpy, Playwright + browser). Test in an isolated environment (container/VM) before pointing real keys at it. - Scheduled runs: The registry metadata includes a cron schedule (every 30 minutes). If you install this as a managed automaton, it may run automatically on that schedule. Combined with the dry-run uncertainty, that could lead to unintended live trades. Consider disabling scheduled/autonomous execution until you have confirmed the dry-run behavior. - Keys and permissions: The bot needs your SIMMER API key (used to place trades) and NVIDIA API key (FourcastNet). Only provide API keys with tightly scoped permissions and in a separate test account if possible. Rotate or revoke keys after testing. - Operational impacts: The bot uses Playwright to scrape Wunderground which is resource intensive and may run into anti-bot / rate-limit measures; it writes temp netCDF files when parsing FourcastNet outputs (files are deleted but running on shared hosts may cause issues). Logs are written to bot.log. If the author can demonstrate (or you can confirm) the following, my confidence would move to high/benign: a visible safe dry-run implementation that blocks execute_trade unless --live is passed or a clear env toggles live mode; corrected dependency manifests; and config.py changed to make TELEGRAM_BOT_TOKEN truly optional. Otherwise treat this skill as suspicious and run only in an isolated test environment with non-production API keys.

An automated trading bot for Simmer weather prediction markets. Fetches active weather markets, cross-references temperature forecasts from 4 independent sources, and only trades when the consensus is strong.

How it works

1. Market Discovery — fetches active weather markets from Simmer via the SDK
2. Multi-Source Forecast — gets high temperature predictions from:
   • NOAA (US government weather API)
   • Open-Meteo (free global weather API)
   • Wunderground (scraped via Playwright for broader coverage)
   • NVIDIA FourcastNet (physics-based atmospheric model)
3. Confidence Scoring — computes a 0–100 score based on:
   • Source agreement (all within ±1°F)
   • Market bucket fit
   • Simmer edge recommendation
   • Time to resolution
4. Execution — only trades when score reaches 100 (maximum confidence)

Default signal

The default strategy is conservative multi-source consensus:

• All 3 weather sources must agree within ±1°F
• FourcastNet must confirm the bucket
• Simmer edge must recommend TRADE
• Only YES trades (betting the temp falls within the market bucket)

This is a template. The default signal uses 4 weather models. Remix it by:

• Adding more weather sources (AccuWeather, Weather.com, etc.)
• Adjusting the agreement threshold (currently strict ±1°F)
• Adding NO trades (betting the temp falls outside the bucket)
• Using ML models trained on historical forecast accuracy per source

Setup

Environment variables

SIMMER_API_KEY=your_key        # Required — from simmer.markets
NVIDIA_API_KEY=your_key        # Required — for FourcastNet
TELEGRAM_BOT_TOKEN=your_token  # Optional — for Telegram UI
TRADE_AMOUNT=10.0              # Optional — default $10
CONFIDENCE_THRESHOLD=100       # Optional — default max confidence only
SIMMER_BASE_URL=https://api.simmer.markets  # Optional
SIMMER_VENUE=sim               # Optional — default "sim"

Dependencies

pip install httpx python-telegram-bot python-dotenv numpy
pip install netCDF4              # For FourcastNet output parsing
pip install playwright           # For Wunderground scraping
playwright install chromium      # Required for Wunderground

Supported cities

New York, Los Angeles, Chicago, Miami, Houston, Phoenix, Philadelphia, San Francisco, Seattle, Denver, Boston, Atlanta, Dallas, Minneapolis, Las Vegas, Detroit, Portland, San Antonio, San Diego, Milan, Madrid, Tel Aviv, London, Paris, Berlin, Tokyo.

Add more in city_map.py.

Remix guide

Swap in your own signals:

• Different weather sources: Replace or add forecast functions in the simmer_weather_bot/ folder
• Different scoring: Modify compute_confidence() in strategy.py
• Add NO trades: Extend the strategy to also bet against consensus
• ML-based: Train a model on historical forecast accuracy and replace the simple agreement check

The plumbing (market discovery, trade execution, Telegram UI, health checks) stays the same.

Hard rules

• Always defaults to dry-run. Pass --live for real trades.
• Always tags trades with source and skill_slug for tracking.
• Always includes reasoning with the weather data used.
• Reads API keys from env — never hardcodes credentials.