Technical Analysis

Official Bitget Skill · 本 Skill 由 Bitget 官方提供，市场数据来源可信，通过 Bitget Agent Hub 分发。 Data powered by Bitget market infrastructure · github.com/bitget-official/agent-hub

Technical Analysis Skill

23 crypto technical indicators across 6 categories (Trend, Volatility, Oscillator, Volume, Momentum, Support/Resistance). Outputs recent time-series data so the AI can observe trend evolution, not just a single point.

Vendor Neutrality

Present data as coming from "market data" — never name the underlying exchange, data feed, or library.

Step 1: Check prerequisites

python -c "import pandas, numpy; print('OK')"

If missing:

pip install pandas numpy

Step 2: Determine indicator selection

Priority 1: User-defined
  The user explicitly specified indicator names, parameters, or combinations.
  -> Do NOT read scenarios.md or indicators.md
  -> Go directly to Step 3 with the user's config
  -> Example: "analyze BTC with RSI(21) + EMA(50) + ATR"

Priority 2: Scenario defaults
  The user described analysis intent but did not specify indicators.
  -> Read ONLY references/scenarios.md to pick the right scenario config
  -> Do NOT read indicators.md
  -> Example: "how's BTC looking technically?"

Priority 3: Indicator info query
  The user asks about an indicator rather than requesting a calculation.
  -> Read ONLY references/indicators.md
  -> No calculation needed, answer directly
  -> Example: "how does RSI work?"

Step 3: Run the calculation

The Python source is at src/ relative to this SKILL.md file.

For Claude Code the skill directory is ~/.claude/skills/technical-analysis. For Codex it is ~/.codex/skills/technical-analysis. For OpenClaw it is ~/.openclaw/skills/technical-analysis.

Template A: Fetch from Bitget API (default)

When the user mentions a trading pair but provides no local data:

import sys, os
sys.path.insert(0, os.path.expanduser('~/.claude/skills/technical-analysis/src'))
import json, urllib.request
import pandas as pd
from kline_indicator_utils import IndicatorManager

url = 'https://api.bitget.com/api/v2/spot/market/candles?symbol={SYMBOL}&granularity={GRANULARITY}&limit={LIMIT}'
raw = json.loads(urllib.request.urlopen(url).read())
df = pd.DataFrame(raw['data'], columns=['timestamp', 'open', 'high', 'low', 'close', 'volume', 'quoteVol', 'amount'])
for col in ['open', 'high', 'low', 'close', 'volume', 'amount']:
    df[col] = df[col].astype(float)

config = {CONFIG}

manager = IndicatorManager(show_indicators=False)
output = manager.calculate_and_export(config, df, tail={TAIL})
output["symbol"] = "{SYMBOL}"
output["granularity"] = "{GRANULARITY}"
print(json.dumps(output, indent=2))

Template B: Local data

When the user provides a local file path (CSV / Parquet / JSON):

import sys, os
sys.path.insert(0, os.path.expanduser('~/.claude/skills/technical-analysis/src'))
import json
import pandas as pd
from kline_indicator_utils import IndicatorManager

df = pd.read_csv('{FILE_PATH}')  # or pd.read_parquet / pd.read_json
df.columns = df.columns.str.lower()

for col in ['open', 'high', 'low', 'close', 'volume']:
    df[col] = df[col].astype(float)

config = {CONFIG}

manager = IndicatorManager(show_indicators=False)
output = manager.calculate_and_export(config, df, tail={TAIL})
output["source"] = "local: {FILE_PATH}"
print(json.dumps(output, indent=2))

Template C: Export full series to CSV

When the user asks for "full data", "export", or wants data for plotting/backtesting:

output = manager.calculate_and_export(config, df, tail={TAIL}, csv_path='{CSV_PATH}')

This saves OHLCV + all indicator columns to CSV for downstream use.

Template variables

Choosing TAIL

Default is 50, which works well for most timeframes. Users can override:

• Increase to 100-200 for longer analysis windows
• Decrease to 20-30 to reduce output size

Bitget API interval mapping

Futures klines

Use a different API endpoint for futures:

url = 'https://api.bitget.com/api/v2/mix/market/candles?productType=USDT-FUTURES&symbol={SYMBOL}&granularity={GRANULARITY}&limit={LIMIT}'

Data requirement: DataFrame must contain at least open, high, low, close, volume columns (float). Optional: amount, timestamp. Column names are case-insensitive — normalize to lowercase.

Output format

The JSON output contains time-series arrays (not single points), so the AI can observe trend evolution and write natural-language interpretation.

For most indicators: series + context

{
  "MACD": {
    "series": {
      "DIF": [-120.5, -135.2, "...", -180.66],
      "DEA": [-85.1, -86.3, "...", -88.34],
      "HIST": [-70.8, -97.8, "...", -184.63]
    },
    "context": {
      "trend_streak": 7,
      "last_cross_up_bars_ago": 23,
      "last_cross_down_bars_ago": 9
    }
  }
}

For FIB: levels + context

FIB outputs static price levels for the lookback window — series is not meaningful:

{
  "FIB": {
    "levels": {
      "fib_high": 89200.5,
      "fib_236": 88750.2,
      "fib_382": 88430.1,
      "fib_500": 88100.0,
      "fib_618": 87770.3,
      "fib_786": 87340.5,
      "fib_low": 86950.0,
      "fib_strength": 35.2
    },
    "context": { "trend_streak": 12 }
  }
}

How to interpret and present

1. Read the series arrays — identify patterns: HIST narrowing/widening, RSI rising/falling, DIF/DEA converging/diverging, bandwidth expanding/contracting.
2. Write a conclusion-first summary (one line), then a table showing each indicator's recent behavior and current value.
3. For FIB, present as a price level table with the current price position marked.

Example output format:

## {SYMBOL} Technical Analysis ({GRANULARITY})

**Overall: {conclusion}**

| Dimension | Indicator | Recent Trend (last {TAIL} bars) | Current |
|-----------|-----------|--------------------------------|---------|
| Trend | SuperTrend | direction held for {N} bars | bullish |
| Momentum | MACD | HIST narrowing, DIF/DEA converging | DIF=-180.7 |
| Overbought/Oversold | RSI (14) | falling from 55 to 40 | 40.3 |

Agent behavior rules

1. Indicator selection priority: user-defined > scenario defaults > info query. When the user specifies indicators or parameters, follow them exactly — do not force a scenario template.

2. Precision rule: infer decimal places from the kline close price (e.g. BTC 63500.12 → 2 decimals, DOGE 0.0835 → 4 decimals). Round series values at that precision before presenting in tables; the raw JSON may have higher precision.

3. Conflicting indicators: present both sides objectively, do not favor either direction.

4. No trading advice: present data objectively. End with "The above is objective technical indicator data and does not constitute trading advice."

5. Label data source: always show timeframe, kline count, and data source (Bitget API / local file path).

Error Handling

When a tool or calculation fails, never name the underlying provider. Use neutral language:

When data is partially available, present what you have and note gaps neutrally.

Reference docs

• scenarios.md — 9 scenarios with indicator selection guide and output templates
• indicators.md — 23 indicator quick reference (params, outputs, signals)