Market Crypto

Crypto skill

Routing — use recommend by default

For ANY user question about trading, direction, entry, buy/sell, "analisa", "rekomendasi", "gimana arahnya", "mau beli/jual", "setup", "sinyal", "pump", "dump" — call recommend. It is the engine that combines TA + SMC + macro + news + Telegram sentiment and outputs the actionable verdict.

Only call analyze if the user explicitly asks for a plain TA snapshot, "chart aja", or "ringkasan teknikal" without asking for a decision.

Output format — ALWAYS use --tg

Append --tg to every recommend and analyze call. It makes the skill emit a pre-rendered plain-text Telegram card (verdict + plan + R/S levels + explicit LONG/SHORT trigger conditions + 3 reasoning bullets + macro tags + NFA). The card is the message body — relay it verbatim to the user.

DO NOT re-narrate the card into "Current Snapshot / Critical Levels / Last 7 Candles / Skenario / Rekomendasi" essays. The card is intentionally terse so the user gets actionable info at a glance.

# Recommended invocation (used for ALL market queries):
python3 -m skills.crypto.analyze recommend  --tf 1d --tg
python3 -m skills.crypto.analyze analyze  --tf 1d --tg   # legacy, also emits card via --tg

Skip --tg only when the user explicitly asks for a chart attachment or a full data drill-down (e.g. "kasih chart" / "jelasin SMC-nya panjang").

Commands (recommend = default for trading queries)

| Command | Example | What it returns | | --- | --- | --- | | recommend | recommend BTC --tf 1d | verdict (LONG/SHORT/WAIT) + confidence + score + entryzone + stoploss + takeprofit[3] + riskreward[3] + suggestedleveragex + reasoning + warnings + SMC detectors (OB/FVG/sweeps/zone) + news_summary (RSS + Telegram) + narrative | | backtest | backtest BTC --tf 1d --risk balanced | walk-forward metrics (trades, win%, total R, profit factor, max DD, expectancy) | | analyze | analyze SOL --tf 1d | legacy: snapshot + chart PNG + news + narrative (NO verdict). Prefer recommend. | | price | price BTC | live USD/IDR, mcap, 24h vol, 24h change | | ta [--tf 1d] | ta ETH --tf 4h | raw RSI/MACD/BB/EMA/ATR/pivot | | news [symbol] | news BTC | latest CryptoPanic posts | | compare | compare BTC ETH | side-by-side TA snapshot | | watchlist add/rm/list [sym] | watchlist add BTC | manage persistent watchlist |

recommend flags

• --tf {1h|4h|1d|1w} (default 1d)
• --risk {conservative|balanced|aggressive} (default conservative: |score|≥4 & RR≥2)
• --mtf enable multi-timeframe confluence (1w / 1d / 4h)
• --no-smc opt-out of Smart Money Concepts overlay (default ON)
• --no-news / --no-telegram opt-out of news/Telegram sentiment
• --no-narrative skip LLM narrative

Output is JSON to stdout. Charts are saved to $HERMES_CHART_OUT (default /tmp/hermes-charts/) and the absolute path is included in the JSON as chart_path.

Symbols

Use the bare ticker (BTC, ETH, SOL, DOGE). The skill maps it to:

• CoinGecko ID (bitcoin, ethereum, solana, …) for price + market data.
• CCXT pair /USDT for OHLCV via Binance public REST.

Custom mappings can be added in feeds.py:_COINGECKO_IDS. For coins not in the map, the lowercased ticker is tried as a CoinGecko ID directly.

Data sources

• CoinGecko — /simple/price, /coins/{id}. Free Demo tier or anonymous. Set COINGECKO_API_KEY to lift the rate limit.
• Binance via CCXT — fetch_ohlcv on spot pairs. No auth, no key.
• DeFiLlama — /tvl/{protocol}, chain TVL. No auth.
• CryptoPanic — /posts/. Requires CRYPTOPANIC_API_KEY (free 200/day).
• alternative.me — Fear & Greed /fng/?limit=1. No auth.

Hermes autonomy

Read-only by design. No trade execution, no signing, no on-chain calls. Safe to run without confirmation. The agent should still apply rate-limit prudence (max ~10 analyze calls/min).

Caveats

• Stablecoins (USDT, USDC) work for price queries but TA is meaningless.
• Memecoins outside the curated ID map need either the explicit CoinGecko slug or the CCXT pair name (PEPE/USDT).
• TwelveData and Glassnode are NOT used — keeping the crypto skill 100% free tier.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: jagres0039
• Source: jagres0039/hermes-market-skills
• License: MIT

Install and usage instructions live in the source repository linked above.