SKILL.md

$2a

Tools: write_file, bash, read_file

When to Use Which Chart

Simple price action → Candlestick with no indicators. Good for "show me BTC this month."

Trend analysis → Add EMA/SMA overlays. Good for "is ETH in an uptrend?"

Momentum check → Add RSI or MACD as subplots. Good for "is SOL overbought?"

Full technical view → Candles + Bollinger Bands + RSI + MACD. Good for "give me the full picture on BTC."

Volume analysis → Requires separate fetch from market_chart endpoint (OHLC endpoint has no volume).

Asset comparison → Line chart comparing two assets (BTC vs Gold, ETH vs S&P500, etc.). Use comparison template for normalized or percentage-based comparisons.

How to Build Charts

Read and customize the template scripts in skills/charting/scripts/:

chart_template.py — Baseline candlestick chart with TradingView styling (crypto via CoinGecko)

chart_with_indicators.py — RSI, MACD, Bollinger Bands, EMA/SMA examples (crypto via CoinGecko)

chart_stock_template.py — Stock/forex chart using Twelve Data API

chart_comparison_template.py — Compare two assets (crypto vs commodity, stock vs crypto, etc.)

Copy the relevant template to scripts/, customize the config section (coin, days, indicators), and run it.

The templates handle all data fetching internally with retry logic and error handling.

Note: These templates are for market data visualization (price charts, indicators). For backtest result charts (equity curves, drawdowns, performance dashboards), add matplotlib charting directly to your backtest script — the data is already there, no need to re-fetch or create a separate file.

TradingView Color Palette

Element

Color

Hex

Up candles

Teal

#26a69a

Down candles

Red

#ef5350

Background

Dark

#131722

Grid

Subtle dotted

#1e222d

Text / axes

Light gray

#d1d4dc

MA lines

Blue / Orange

#2196f3 / #ff9800

RSI line

Purple

#b39ddb

MACD line

Blue

#2196f3

Signal line

Orange

#ff9800

Do not deviate from this palette unless the user asks.

Data Source APIs

CoinGecko (Crypto Only)

Endpoint: https://pro-api.coingecko.com/api/v3/coins/{coin_id}/ohlc/range

Auth: Header x-cg-pro-api-key: {COINGECKO_API_KEY}

Use for: BTC, ETH, SOL, and all cryptocurrencies

Example:

url = f"https://pro-api.coingecko.com/api/v3/coins/{COIN_ID}/ohlc/range"

params = {"vs_currency": "usd", "from": from_ts, "to": now, "interval": "daily"}

headers = {"x-cg-pro-api-key": os.getenv("COINGECKO_API_KEY")}

resp = requests.get(url, params=params, headers=headers)

raw = resp.json()  # [[timestamp_ms, open, high, low, close], ...]

Twelve Data (Stocks, Forex, Commodities)

Endpoint: https://api.twelvedata.com/time_series

Auth: Query param apikey={TWELVEDATA_API_KEY}

Use for: Stocks (AAPL, MSFT), Forex (EUR/USD), Commodities (XAU/USD for gold)

Common Symbols:

Stocks: AAPL, MSFT, GOOGL, TSLA, SPY

Forex: EUR/USD, GBP/JPY, USD/CHF

Commodities: XAU/USD (gold), XAG/USD (silver), CL/USD (crude oil)

Intervals: 1min, 5min, 15min, 30min, 1h, 4h, 1day, 1week, 1month

Example:

url = "https://api.twelvedata.com/time_series"

params = {

    "symbol": "XAU/USD",  # Gold spot price

    "interval": "1day",

    "outputsize": 90,  # Number of candles

    "apikey": os.getenv("TWELVEDATA_API_KEY")

}

resp = requests.get(url, params=params)

data = resp.json()

# data["values"] = [{"datetime": "2024-01-01", "open": "2050.00", "high": "2060.00", ...}, ...]

IMPORTANT: Twelve Data returns data in reverse chronological order (newest first). Always reverse the list before creating a DataFrame:

values = data["values"][::-1]  # Reverse to oldest-first

Interval Selection Strategy

The templates now auto-select optimal intervals to minimize data volume while maintaining visual quality:

Time Range

Auto-Selected Interval

Rationale

≤31 days

Hourly

High granularity for short-term analysis

32-365 days

Daily

Sufficient detail, lower data volume

>365 days

Daily

Daily is optimal for long-term trends

Override: Set INTERVAL = "daily" or INTERVAL = "hourly" in the config to override auto-selection.

Key Gotchas

**savefig facecolor**: You MUST set facecolor='#131722' and edgecolor='#131722' in savefig, or the saved PNG reverts to white background.

Title spacing: Prefix titles with \n to add spacing from the top edge.

**returnfig=True**: Use when you need post-plot customization (price formatting, annotations). When using it, call fig.savefig() manually — don't pass savefig to mpf.plot().

No volume in OHLC: CoinGecko OHLC endpoint returns [timestamp_ms, open, high, low, close] only. Use volume=False or fetch volume separately from coin_chart endpoint.

Panel ratios: Set panel_ratios when adding indicator subplots. E.g., (4, 1, 2) for candles + volume + one indicator, (5, 1, 2, 2) for two indicators.

Figure size: Default (14, 8). Increase to (14, 10) or (14, 12) when adding subplots.

Rules

Paths are relative to workspace. Write to scripts/foo.py, not workspace/scripts/foo.py. The bash CWD is already workspace.

**Always save to output/ directory.** Use os.makedirs("output", exist_ok=True).

**Always run the script with bash("python3 scripts/<name>.py")** to verify it works.

**Always call read_file on the generated PNG, then use markdown image syntax to display it:** ![Chart](output/filename.png)

Scripts must be standalone. Use requests + os.getenv(). No internal imports, no dotenv.

CRITICAL: Do NOT use proxied_get() in chart scripts. Chart scripts are standalone and run in the workspace - they cannot import from core.http_client. Always use requests.get() and requests.post() directly. This is an exception to the PLATFORM.md proxy rules because these scripts execute outside the main Star Child process. The templates demonstrate the correct pattern.

Env vars are inherited. os.getenv("COINGECKO_API_KEY") works directly.

Default to dark theme unless user asks for light.

Filename should describe the chart. e.g. btc_30d_candles.png, eth_7d_rsi_macd.png.

Data sources: Use CoinGecko API for crypto (BTC, ETH, etc). Use Twelve Data API for stocks, forex, and commodities (AAPL, EUR/USD, XAU/USD for gold). Never mix APIs - keep scripts focused on one data source.

Think about what you're measuring: Before creating a chart, ask yourself: "What question is the user trying to answer?" A normalized chart (all start at 100) shows relative trends but hides actual gain magnitude. If the user wants to know "which gained more" or is comparing investment performance, they need the actual multipliers (e.g., 50x vs 10x), not just lines that look similar.

Troubleshooting

401 Unauthorized Errors

Templates auto-configure proxy from PROXY_HOST/PROXY_PORT env vars. If 401 errors occur:

Check environment:

bash("env | grep -E 'PROXY|REQUESTS_CA'")

Expected vars:

PROXY_HOST / PROXY_PORT - Proxy address (templates use these to set HTTP_PROXY/HTTPS_PROXY)

REQUESTS_CA_BUNDLE - Proxy CA cert for SSL

COINGECKO_API_KEY / TWELVEDATA_API_KEY - Can be fake in proxied environments

If vars are missing, this is an environment configuration issue, not a script issue.

charting