Menu
AI-driven DeFi trading signal generation with Kelly-criterion position sizing and multi-layer risk controls
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
| name | ave-trading |
| description | AI-driven DeFi trading signal generation with Kelly-criterion position sizing and multi-layer risk controls |
| version | 1.0.0 |
| author | Alchemist-X |
| tags | ["defi","trading","ai","signals","risk-management"] |
Generates trading signals from AVE monitoring data, sizes positions using quarter-Kelly criterion, and manages the full position lifecycle through a 7-tier review system. Applies hard service-layer risk guards (drawdown halt, exposure caps, per-trade limits) to every decision before execution.
| Capability | Description |
|---|---|
| Signal Generation | Momentum-based heuristic analysis of on-chain price data producing buy/sell direction with confidence scoring |
| Position Sizing | 1/4 Kelly criterion: quarterKelly = (edge * confidence) / (4 * odds) with edge = predictedReturn - riskPremium |
| Risk Premium Computation | Dynamic premium based on contract risk level, tax burden, and liquidity depth |
| Entry Planning | Rank candidates by monthly return, cap batch exposure at 20% of bankroll, limit to top N entries |
| Position Review | 7-tier review cascade evaluating every open position against current market prices |
| Risk Guards | Portfolio-level hard limits enforced at the service layer before any decision reaches execution |
AVE Market Pulse AVE Entry Planner
================ ==================
AvePulseCandidate[] For each candidate:
| 1. estimatePredictedReturn()
| - direction (buy/sell)
Pulse Filters - confidence (0.5 + |change| * 2, cap 0.9)
(vol > $10k, liq > $5k, - predictedReturn = |change| * confidence
no honeypots, tax < 10%) 2. computeRiskPremium()
| - base: 2%
Composite Score Sort - medium risk: +1%
log10(vol) * log10(liq) * risk - high/critical risk: +5%
| - low liquidity (<$50k): +2%
Top N candidates - tax burden: +totalTax/100
| 3. edge = predictedReturn - riskPremium
v 4. calculateAveKelly({ edge, confidence })
+------------------+ 5. Monthly return = edge / holdingPeriod
| Entry Planner | ------+
+------------------+ | AVE Position Review
| | ====================
+------------------+ |
| Position Review | ------+--- composeAveDecisions() ---+
+------------------+ |
applyRiskGuards()
|
TradeDecisionSet
Every open position is evaluated in priority order. The first matching tier determines the action.
| Tier | Basis | Action | Exit % | Confidence |
|---|---|---|---|---|
| 1 | Stop-loss breached | close | 100% | medium |
| 2 | Profit target hit | close | 100% | high |
| 3 | Edge gone negative (< -5%) | close | 100% | medium |
| 4 | Near stop-loss (within 30% of threshold) | reduce | 50% | low |
| 5 | Edge weakening (0% to -5%) | reduce | 50% | low |
| 6 | Edge positive (> +2%) | hold | 0% | medium |
| 7 | Stable hold (edge near zero) | hold | 0% | low |
Tiers 3, 4, 5, and 7 set humanReviewFlag: true for manual oversight.
| Parameter | Default | Env Variable | Description |
|---|---|---|---|
| Drawdown stop | 30% | DRAWDOWN_STOP_PCT | Halt all trading if portfolio drawdown exceeds threshold |
| Per-position stop-loss | 30% | POSITION_STOP_LOSS_PCT | Close position if unrealized loss exceeds threshold |
| Max total exposure | 80% | MAX_TOTAL_EXPOSURE_PCT | Cap total deployed capital as % of bankroll |
| Max per-trade size | 15% | MAX_TRADE_PCT | Cap any single trade as % of bankroll |
| Max open positions | 22 | MAX_POSITIONS | Hard limit on concurrent positions |
| Min trade size | $5 | MIN_TRADE_USD | Floor for trade notional |
| Liquidity cap | 5% of pair TVL | -- | Max trade size relative to pair liquidity |
| Parameter | Default | Description |
|---|---|---|
| Kelly fraction | 1/4 (quarter Kelly) | Conservative sizing to limit variance |
| Base risk premium | 2% | Minimum hurdle subtracted from predicted return |
| Minimum edge | 0.5% | Entries below this edge are skipped |
| Holding period | 1 month | Assumed holding period for monthly return calculation |
| Max new entries per batch | 4 | Limit on new positions opened per cycle |
| Batch cap | 20% of bankroll | Total capital allocated across all new entries in one cycle |
import { planAveEntries } from "services/orchestrator/src/runtime/ave-entry-planner.js";
const plans = planAveEntries(
filteredCandidates,
{
bankrollUsd: 10_000,
existingPositions: ["0x...abc-ethereum"],
},
{
maxNewEntries: 4,
batchCapPct: 0.2,
stopLossPct: 0.3,
minEdge: 0.005,
}
);
// plans: AveEntryPlan[] sorted by monthly return, batch-capped
import { reviewAvePositions } from "services/orchestrator/src/review/ave-position-review.js";
const reviews = reviewAvePositions(
currentPositions,
currentPriceMap, // Map<tokenId, priceUsd>
{
stopLossPct: 0.3,
targetProfitPct: 0.5, // optional
}
);
for (const review of reviews) {
// review.action: "hold" | "reduce" | "close"
// review.basis: "stop-loss-breached" | "profit-target-hit" | ...
// review.decision: TradeDecision (ready for execution)
}
import { runAveDirectPipeline } from "services/orchestrator/src/runtime/ave-direct-runtime.js";
const result = await runAveDirectPipeline(
{
runId: crypto.randomUUID(),
mode: "full",
bankrollUsd: 10_000,
totalEquityUsd: 10_000,
positions: [],
openPositionCount: 0,
},
orchestratorConfig
);
// result.decisionSet: TradeDecisionSet with all decisions
// result.entryPlans: new positions planned
// result.positionReviews: existing position evaluations
{
"tokenAddress": "0x6982508...a011",
"chain": "ethereum",
"tokenId": "0x6982508...a011-ethereum",
"tokenSymbol": "PEPE",
"direction": "buy",
"sizeUsd": 142.50,
"confidence": 0.67,
"predictedReturn": 0.057,
"riskPremium": 0.02,
"edge": 0.037,
"fullKellyPct": 0.02479,
"quarterKellyPct": 0.00620,
"monthlyReturn": 0.037,
"holdingPeriodMonths": 1,
"currentPriceUsd": 0.00001234,
"liquidityUsd": 18500000,
"riskLevel": "low",
"confidenceBucket": "medium-high"
}
{
"run_id": "a1b2c3d4-...",
"runtime": "ave-direct-runtime",
"generated_at_utc": "2026-04-13T12:00:00.000Z",
"bankroll_usd": 10000,
"mode": "full",
"decisions": [
{
"action": "open",
"event_slug": "ethereum",
"market_slug": "0x698...a011-ethereum",
"token_id": "0x698...a011-ethereum",
"token_symbol": "PEPE",
"side": "BUY",
"notional_usd": 142.50,
"order_type": "FOK",
"ai_prob": 0.67,
"market_prob": 0.50,
"edge": 0.037,
"confidence": "medium-high",
"thesis_md": "**PEPE** on ethereum: Price $0.000012, 24h change +8.50%. Direction: BUY with 67.0% confidence. Edge: 3.70%. Quarter Kelly size: $142.50.",
"sources": [
{
"title": "AVE token data for PEPE",
"url": "https://ave.ai/token/0x698...a011-ethereum",
"retrieved_at_utc": "2026-04-13T12:00:00.000Z"
}
],
"full_kelly_pct": 0.02479,
"quarter_kelly_pct": 0.00620,
"liquidity_cap_usd": 925000,
"stop_loss_pct": 0.3,
"resolution_track_required": false
}
],
"artifacts": []
}
The trading skill consumes every monitoring output (price, volume, liquidity, risk assessment) and transforms it through a quantitative pipeline: momentum estimation, dynamic risk premium, Kelly sizing, 7-tier position review, decision composition with deduplication, and portfolio-level risk guards. Every TradeDecision carries a full audit trail including thesis markdown, source URLs, and Kelly fractions.