[April 20, 2026] — Experiment Closed
DegenClaw Bot has been shut down. The Railway deployment is offline and all remaining funds have been withdrawn to my personal wallet.
Why: The infrastructure cost (Railway + Anthropic API credits for the 4-agent QuantAgent pipeline) outpaced the account size (~$17). Running a multi-agent LLM stack on a $17 trading account is fundamentally uneconomical — API calls alone ate into the edge before any trade was placed. I ran out of API credits and decided this was the natural stopping point rather than topping up again.
What worked: The ACP integration held up, the QuantAgent signal pipeline was directionally coherent (10 SHORTs / 4 LONGs across 14 trades), and the admin dashboard gave good visibility. The architecture is sound — it just needs a larger account to be viable.
What didn't: Cost efficiency. Claude Haiku at $0.25/MTok sounds cheap, but 4 agents × every 5-minute tick adds up fast on Railway's always-on runtime. The 10x leverage overhaul came too late to compound meaningfully.
Thanks to everyone who followed along. The code stays public — fork it, scale it up, or use it as a reference for your own competition runs.
Virtuals Protocol DegenClaw Competition — S1 March 2026
An autonomous BTC perpetuals trading bot built for the DegenClaw $100K weekly competition on Hyperliquid. Uses QuantAgent — a 4-agent LangGraph framework — to generate trading signals, executes them through the DegenClaw ACP (Agent Commerce Protocol), and posts rationales to the competition forum.
https://degen.virtuals.io/agents/337 https://app.virtuals.io/virtuals/68511
- Total trades: 14 (13 closed, 1 currently open)
- Direction breakdown: 10 SHORTs, 4 LONGs
┌─────────────────────────────────────────────────────────────────┐
│ DegenClaw Bot │
│ │
│ Hyperliquid QuantAgent Pipeline ACP Layer │
│ ────────── ───────────────────── ───────────── │
│ OHLCV candles → IndicatorAgent (RSI, → acp job create │
│ (4h, 50 bars) MACD, ROC, Stoch, %R) acp job pay │
│ ↓ │
│ PatternAgent (chart Competition │
│ formations) → Leaderboard │
│ ↓ (score tracked │
│ TrendAgent (S/R, via ACP jobs) │
│ channels, trendlines) │
│ ↓ │
│ DecisionAgent → Forum post │
│ LONG / SHORT + R:R (rationale) │
└─────────────────────────────────────────────────────────────────┘
| Decision | Reasoning |
|---|---|
| All trades via ACP | Competition only counts jobs routed through the DegenClaw ACP agent |
| Claude Haiku for all agents | Fast (~3s per signal), cost-efficient at $0.25/MTok input |
| Temperature 0.1 | Deterministic, consistent signals — avoids random LLM variance |
| ATR-based stops | Adapts to volatility; 0.5× ATR stop distance, R:R set by QuantAgent (1.2–1.8) |
| 10% risk per trade | Raised from 2% — at ~$17 account size, 2% risk produced negligible notional and ~1x leverage regardless of the leverage cap; 10% allows the sizing formula to reach meaningful leverage |
| 10x max leverage | Raised from 5x — combined with tighter 0.5× ATR stops and 10% risk, each trade now targets |
| 30% drawdown circuit breaker | Halts trading if equity falls 30% from peak |
| Direct HL API for monitoring | Avoids SDK 429 rate limits on Railway restarts |
1. Query Hyperliquid → current equity + open positions
2. Circuit breaker check (30% drawdown halt)
3. If position open → monitor price vs TP/SL → close via ACP if hit
4. If no position → run QuantAgent pipeline:
a. Fetch 50 × 4h candles from Hyperliquid
b. Calculate ATR (14-period)
c. Run 4-agent LangGraph analysis
d. Parse LONG/SHORT direction + risk-reward ratio
e. Compute ATR-based stop-loss (0.5× ATR) and take-profit
5. Execute via ACP job → approve payment → position live
6. Post rationale to DegenClaw forum
| Metric | Weight |
|---|---|
| Sortino Ratio | 40% |
| Return % | 35% |
| Profit Factor | 25% |
Strategy is calibrated to balance return and downside risk protection.
degenclaw-agent/
├── src/
│ ├── bot.py # Main loop — tick, monitor, execute
│ ├── bridge.py # QuantAgent ↔ execution bridge
│ ├── data_feed.py # Hyperliquid OHLCV fetcher
│ ├── trade_tracker.py # CSV trade logger
│ ├── admin.py # Admin portal (equity, position, trade history)
│ └── risk.py # Position sizing + drawdown circuit breaker
├── config/
│ ├── settings.py # All tuneable parameters (loaded from env)
│ └── quantagent_config.py # LLM model config
├── openclaw/ # ACP integration (git clones — see Setup)
│ ├── openclaw-acp/ # ACP CLI
│ └── dgclaw-skill/ # DegenClaw skill scripts
├── quantagent/ # QuantAgent submodule (Y-Research-SBU/QuantAgent)
├── logs/ # Runtime logs (gitignored)
├── run_acp.sh # ACP CLI wrapper (tsx, no npm link)
├── nixpacks.toml # Railway build config
├── Procfile # Railway process config
├── requirements.txt
└── .env.example
- Python 3.11+
- Node.js 20+
ta-libC library (brew install ta-libon macOS)- Anthropic API key
- Hyperliquid account
- DegenClaw registration (Import Champion → generate API key)
git clone https://github.com/your-username/degenclaw-agent.git
cd degenclaw-agent
# QuantAgent (4-agent LangGraph framework)
git clone https://github.com/Y-Research-SBU/QuantAgent.git quantagent
pip install -r quantagent/requirements.txtmkdir -p openclaw && cd openclaw
git clone https://github.com/Virtual-Protocol/openclaw-acp.git
cd openclaw-acp && npm install
cd ..
git clone https://github.com/Virtual-Protocol/dgclaw-skill.git
# Follow dgclaw-skill README to configure .env with your DGCLAW_API_KEY
cd ..python3 -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r requirements.txtcp .env.example .env
# Edit .env with your keys — see .env.example for all required variablespython -m src.botRailway handles the full deployment including ta-lib (via Nix) and Node.js for the ACP CLI.
# Install Railway CLI
npm install -g @railway/cli
railway login
# Create project and set environment variables
railway init
railway variables set ANTHROPIC_API_KEY=... HL_PRIVATE_KEY=... # etc.
# Deploy
railway up --service degenclaw-agentThe nixpacks.toml handles ta-lib, Node.js, and npm install automatically.
Note: Set all environment variables in the Railway dashboard under your service's Variables tab before deploying. The bot will crash on startup if any required key is missing.
All risk parameters are configurable via environment variables:
| Parameter | Default | Description |
|---|---|---|
RISK_PER_TRADE |
0.10 |
10% of equity at risk per trade |
MAX_LEVERAGE |
10 |
Hard cap on leverage |
MAX_CONCURRENT_POSITIONS |
1 |
One position at a time (BTC only) |
CHECK_INTERVAL_SECONDS |
300 |
How often to check for new signals |
| Drawdown circuit breaker | 30% | Halts all trading if equity drops 30% from peak |
size_usd = (equity × RISK_PER_TRADE) / stop_distance × entry_price
size_usd = min(size_usd, equity × MAX_LEVERAGE)
leverage = min(size_usd / equity, MAX_LEVERAGE)
Where stop_distance = 0.5 × ATR(14). At typical BTC volatility (~800 ATR) and a $17 account, this targets ~$170 notional (~10x leverage) with a maximum loss of ~$1.70 (10% of equity) if the stop is hit.
Position sizing overhaul (10x leverage)
The original strategy used 2% risk per trade and a 1.5× ATR stop multiplier. With a ~$17 account and BTC ATR around 800–1100, the stop distances were so large relative to account size that the position sizer never approached the leverage cap — every trade was effectively 1x regardless of the MAX_LEVERAGE setting. Changes made:
RISK_PER_TRADEraised from 2% → 10%: more equity is put at risk per trade, increasing the required notional to achieve that risk amountMAX_LEVERAGEraised from 5x → 10x: higher ceiling for the position size capDEFAULT_STOP_ATR_MULTIPLIERtightened from 1.5× → 0.5× ATR: narrower stops mean a smaller price move triggers the stop, so a larger position is needed to risk the same dollar amount — this is what pushes sizing up to the leverage cap- Removed the hard 90%-of-equity cap on position size: previously
size_usdwas capped atequity × 0.9regardless of leverage, preventing any meaningful leverage from being applied. Now capped atequity × MAX_LEVERAGE
Net effect: next trade will be ~$170 notional on a $17 account (~10x leverage), up from ~$17 (1x).
Admin dashboard improvements
- Replaced the raw CSV text dump with a proper trade history table showing all columns (direction, entry, size, leverage, SL, TP, R:R, status, P&L) with color-coded direction tags and P&L
- Added Net P&L metric card (current equity vs $18.99 starting balance)
- Added Peak Equity metric card (sourced from live bot state)
- SL and TP levels now shown in the Unrealized P&L card
peak_equityadded tobot_state.jsonso the admin portal always reflects the bot's tracked peak
Experiment closed April 20, 2026. 14 total trades (13 closed, 1 open at shutdown). Remaining funds withdrawn to personal wallet.
- QuantAgent — multi-agent LangGraph trading framework
- Virtuals Protocol — DegenClaw competition infrastructure
- Hyperliquid — on-chain perpetuals exchange
This is an experimental competition bot. Not financial advice. Use at your own risk.