The Autopilot SDK (@shoot/autopilot) lets traders configure autonomous trading playbooks that execute around the clock within Adrena prop challenges. Traders pick a playbook, tune its parameters, and the autopilot handles execution — generating continuous volume without requiring the trader to be online.
Why Autopilot?
Adrena’s competitions generated massive volume — but only during human trading hours. The autopilot flips this: traders configure a playbook (strategy + risk guardrails), register it on-chain as an agent, and the FlightController executes ticks 24/7. This keeps volume flowing around the clock while keeping humans in the loop for strategy design.
Architecture
┌──────────────────────────────────────────────────┐
│ FlightController (cockpit/) │
│ Tick loop: fetch → assess → validate → execute │
├──────────────────────────┬───────────────────────┤
│ Playbooks (playbooks/) │ RiskHarness │
│ TrendSurfer │ Ceiling leverage │
│ FadeTrader │ Max exposure fraction │
│ RangeSniper │ Cut-loss / lock-gain │
│ FundingArb │ Cooldown enforcement │
│ GridRunner │ │
├──────────────────────────┼───────────────────────┤
│ Indicators (indicators/) │ Feed (feed/) │
│ VWAP · ATR · MACD │ OracleTap (Pyth) │
│ Stochastic · Keltner │ ReplayTap (test) │
├──────────────────────────┴───────────────────────┤
│ On-chain (onchain/) │
│ ShootProgram · deriveAgentPda · PerpBuilder │
└──────────────────────────────────────────────────┘
Playbooks
Each playbook implements the Playbook interface with an assess() method that returns a Verdict:
interface Playbook {
readonly label: string;
readonly summary: string;
assess(bars: Bar[], exposure: Exposure | null): Verdict;
}
type Verdict =
| { kind: "buy"; conviction: number; allocation: number }
| { kind: "sell"; conviction: number; allocation: number }
| { kind: "exit"; memo: string }
| { kind: "pass" };
TrendSurfer
MACD histogram zero-line crossover. Enters when the MACD histogram crosses from negative to positive (buy) or positive to negative (sell). Exits when histogram reverses against the position direction. Uses histogram momentum — a derivative signal — rather than raw line crossover.
| Parameter | Default | Description |
|---|
fastLen | 12 | Fast MACD period |
slowLen | 26 | Slow MACD period |
signalLen | 9 | Signal line smoothing |
FadeTrader
Keltner Channel + Stochastic confirmation. Fades price when it breaches the outer Keltner envelope AND the Stochastic %K confirms oversold/overbought. Requires dual confirmation from two independent indicators. Exits when price returns to the Keltner basis (center EMA).
| Parameter | Default | Description |
|---|
keltnerPeriod | 20 | Keltner Channel EMA period |
atrMultiplier | 2.0 | ATR band width |
stochK | 14 | Stochastic %K period |
stochD | 3 | Stochastic %D smoothing |
stochOB | 80 | Overbought threshold |
stochOS | 20 | Oversold threshold |
Keltner Channels use EMA + ATR for the envelope. This is mathematically
different from Bollinger Bands (SMA + standard deviation). The ATR-based
envelope adapts to true price range rather than just close-price variance.
RangeSniper
ATR contraction/expansion detector with VWAP directional bias. Detects volatility squeezes (ATR declining for N consecutive bars), then enters when ATR suddenly expands and price moves away from VWAP. This captures volatility regime changes rather than static price level breaks.
| Parameter | Default | Description |
|---|
atrPeriod | 14 | ATR calculation period |
contractionBars | 5 | Consecutive bars of declining ATR |
expansionMult | 1.5 | Required ATR expansion ratio |
FundingArb
Perpetual funding rate exploitation via price deviation proxy. Unique to perps — estimates “implied funding” from the spread between current price and recent average price. When shorts are overcrowded (negative implied funding), goes long expecting mean reversion. Conservative allocation (counter-trend).
| Parameter | Default | Description |
|---|
fundingThreshold | 0.01 | Entry threshold (1% deviation) |
exitThreshold | 0.003 | Exit threshold (0.3% deviation) |
lookbackBars | 20 | Recent price window |
FundingArb is the only playbook specifically designed for perpetual futures
mechanics. It exploits the tendency of overcrowded positions to generate
funding rate pressure that eventually mean-reverts.
GridRunner
Dynamic ATR-based grid trading around a VWAP anchor. Places virtual buy/sell levels above and below VWAP, spaced by ATR fractions. When price crosses a grid level, it triggers a trade. Stateful: tracks which grid level was most recently crossed.
| Parameter | Default | Description |
|---|
gridLevels | 5 | Number of levels above/below center |
atrPeriod | 14 | ATR period for spacing |
gridSpacingAtrFrac | 0.5 | Grid spacing as fraction of ATR |
Indicators
All indicator functions are pure (no side effects, no IO). Each takes numeric arrays or Bar[] and returns computed values.
| Function | Input | Output | Description |
|---|
computeVWAP | Bar[] | number[] | Running volume-weighted average price |
computeATR | Bar[], period | number[] | Wilder-smoothed average true range |
computeMACD | closes[], fast, slow, signal | MacdResult | MACD line, signal, histogram |
computeStochastic | Bar[], kPeriod, dPeriod | StochResult | %K and %D oscillator lines |
computeKeltner | Bar[], period, atrMult | KeltnerEnvelope | EMA basis + ATR-based channels |
Risk Management
The RiskHarness validates every verdict before execution:
const guardrails: Guardrails = {
ceilingLeverage: 5, // max leverage
maxExposureFrac: 0.25, // max bankroll fraction per trade
cutLossPct: 0.03, // 3% stop-loss
lockGainPct: 0.06, // 6% take-profit
pauseMs: 30_000, // 30s cooldown between trades
};
- Cooldown: rejects if less than
pauseMs since last trade
- Duplicate exposure: rejects buy when already long (or sell when already short)
- Exposure fraction: rejects if allocation exceeds
maxExposureFrac
- Leverage: rejects if implied leverage exceeds
ceilingLeverage
- Stop-loss / take-profit: force-exits when thresholds are breached
FlightController
The FlightController runs the autonomous trading loop:
Each tick:
1. Fetch bars from feed (OracleTap or ReplayTap)
2. Check guardrails (stop-loss / take-profit)
3. If guardrail exit → execute and return
4. Get playbook verdict via assess()
5. Validate through RiskHarness
6. If buy/sell/exit → execute trade
7. Wait cadenceMs, repeat
On-Chain Integration
Autopilot agents are registered on-chain via the Shoot Anchor program:
import { ShootProgram, deriveAgentPda } from "@shoot/autopilot";
// Derive agent PDA
const [agentPda] = deriveAgentPda(ownerPublicKey);
// Build register instruction
const ix = await program.buildRegisterAgentIx(
"MyTrendSurfer",
strategyHash // SHA-256 of strategy config
);
- Owner public key
- Strategy hash (SHA-256 of config, updated on strategy change)
- ELO rating (starts at 1000)
- Win/loss record and total P&L
- Competition count and last trade timestamp
Testing
cd sdk
npm install
npm test # 144 tests across 14 suites
| Suite | Tests | Coverage |
|---|
| Indicators (VWAP, ATR, MACD, Stochastic, Keltner) | 42 | Pure math, edge cases, known values |
| Playbooks (TrendSurfer, FadeTrader, RangeSniper, FundingArb, GridRunner) | 66 | Signals, exits, conviction, exposure |
| Cockpit (FlightController, RiskHarness) | 23 | Risk validation, stop-loss, tick lifecycle |
| Feed + On-chain (ReplayTap, ShootProgram, PerpBuilder) | 13 | Cursor replay, PDA derivation, IX builders |
