caribou-strategy
CARIBOU v1.0 — Cross-Asset Trend Fund (managed futures / CTA). Trend-follows a maximally diversified universe spanning EVERY asset class on Hyperliquid — crypto, xyz stocks, indices, metals, energy — long the uptrends and short the downtrends. Each asset is judged by its OWN trend (time-series, 4h hard gate + daily align + momentum + RSI guard), positions are sized to EQUAL RISK (volatility parity — margin scales inversely with the asset's ATR%), and the book is capped per asset CLASS so it can never collapse into one class. Two INDEPENDENT sleeves on SEPARATE wallets (long / short), so the fund may hold the same asset long in one and short in the other. Long uptrends + short downtrends across uncorrelated classes = net beta ~0 and crisis-positive (it shorts the fallers when markets break). The managed-futures hedge. NOT a copy-trader; runtime owns the LLM gate (pass-through), the asymmetric trend DSL, and all risk.guard_rails.
Resources
6Install
npx skillscat add senpi-ai/senpi-skills/caribou-strategy Install via the SkillsCat registry.
🦌 CARIBOU v1.0 — Cross-Asset Trend Fund (Managed Futures / CTA)
The great migration: follow the trend across all terrain. Caribou trend-follows
a maximally diversified universe spanning every asset class on Hyperliquid —
crypto, xyz stocks, indices, metals, energy — long the uptrends and short the
downtrends, each position sized to equal risk.
| Sleeve | What it does | Direction | Wallet |
|---|---|---|---|
| long | opens LONGS on assets in a confirmed uptrend | LONG only | one |
| short | opens SHORTS on assets in a confirmed downtrend (the crisis-alpha engine) | SHORT only | one |
Two sleeves, two wallets, fully independent — so the fund is not restricted
from holding the same asset in opposite directions (e.g. the long sleeve
trails out a stale ETH long while the short sleeve opens a fresh ETH short on a
trend flip). Each wallet nets per-asset on its own.
Why a cross-asset trend fund
Managed futures (CTA) is the most-proven hedge-fund category, and its entire edge
is cross-asset diversification: trends appear in different markets at different
times. When crypto chops, gold may be trending; when stocks sell off, oil or the
dollar may be trending. The more uncorrelated markets you trend-follow, the
smoother the curve — which is exactly why "use every asset class" is the thesis,
not a bolt-on. Long uptrends + short downtrends across uncorrelated classes →
net beta near zero and crisis-positive (short the fallers when everything
bleeds). That is the hedge.
How it works (producer-side, per sleeve)
- Universe = every class. From the live instrument board, bucket all liquid
names into crypto / equity / index / metal / energy, top-N + relative-median
liquidity gate within each class (so a thin stock isn't compared to BTC volume). - Time-series trend, per asset. Each asset judged vs its own history:
4h structure is the hard gate (bullish→long candidate / bearish→short),
the daily aligns (+/−), 24h momentum confirms, and an RSI guard
blocks blow-off longs / capitulation shorts. - Volatility-parity sizing.
margin = equity × baseRiskPct × (referenceVol / asset_ATR%), clamped to [3%, 15%] of equity. A calm index (~1.5% ATR) gets ~2×
base margin; a wild memecoin (~6%) gets ~0.5× — each position contributes ~equal
risk. This is the managed-futures formula and what keeps the curve from being
crypto-dominated. - Per-class caps. No asset class may exceed 40% of equity in margin — the
book stays diversified across classes (the entire CTA edge). Up to 8 slots. - Conviction leverage.
baseLeverage(3x), bumped tomaxLeverage(5x) on an
apex-score trend, then clamped to each asset's HL venue max.
Asymmetric trend DSL — cut losers fast, let winners run
Trend-following lives and dies on "cut losers short, let winners run." So the
DSL is asymmetric: a tight Phase 1 (cut a losing trend quickly) + a wide
Phase 2 ladder that lets a winner run to +150% before locking 84%, and
time-cuts OFF (a real trend can run for weeks). Taker-fallback entries — a
trend-follower must catch the move (the opposite of a fader).
Deploy — two wallets
CARIBOU_LEG=long CARIBOU_LONG_WALLET=<wallet A> # the long sleeve
CARIBOU_LEG=short CARIBOU_SHORT_WALLET=<wallet B> # the short sleeveFund both for the full hedge (net-beta-neutral). Funding split is your dial —
default ~50/50 long/short; tilt toward long in a broad bull, toward short if you
want more crisis protection.
Fleet-standard rules (enforced)
- Max leverage 5x (strict clamp, then venue max). Vol-parity margin (no
hardcoded $); per-class cap 40%; up to 8 slots. - Drawdown halt 22%, daily loss limit 12%,
max_consecutive_losses 6(trend
takes many small losses before the big trend — don't halt on normal chop). - Mandatory DSL; entries + exits
FEE_OPTIMIZED_LIMITwith taker fallback. - Sizes off
max(main, xyz)account value — never the sum (cross-margin). - Signature-adaptive daemon launch; per-(leg,wallet) fcntl lock.
Hard rule — user-conversation sessions are READ-ONLY
A Claude session conversing with a user MUST NOT call create_position,close_position, edit_position, ratchet_stop_*, cancel_order, or anystrategy_close* tool against Caribou's wallets. Entries are emitted only by the
producer daemons; exits are owned only by the runtime DSL.
v1.0 — initial build
The first cross-asset trend fund — managed futures across every Hyperliquid
asset class, vol-parity sized, class-capped, two independent sleeves. v1.0 gates on
price-action trend (4h/daily structure + momentum + RSI + ATR vol). Cross-asset
correlation-aware netting and a vol-target overlay are planned v1.1 enhancements.