Camel Strategy

🐫 CAMEL v1.0 — Carry Hedge Fund

Camel runs two concurrent strategy wallets that harvest funding carry — it takes the side that collects the funding payment, on names where the crowded trade is exhausting so price doesn't fight the carry. One producer script (camel-producer.py) serves both; the CAMEL_LEG env var selects which book a given daemon is.

> The edge is carry, not direction. On Hyperliquid, positive funding > means longs pay shorts; negative means shorts pay longs. Camel > systematically takes the collecting side of the most extreme funding, > gated to crowds that are rolling over (shorts) or capitulating (longs). > Funding is a recurring, structural inefficiency — income that compounds.

| Book | Style | Wallet env | Runtime | Scanner | |---|---|---|---|---| | harvest | Short the most-positive-funding names (short collects) | CAMEL_HARVEST_WALLET | runtime-harvest.yaml | camel_harvest_signals | | payout | Long the most-negative-funding names (paid to hold) | CAMEL_PAYOUT_WALLET | runtime-payout.yaml | camel_payout_signals |

Taking some shorts (harvest) and some longs (payout) also skews the fund slightly net-neutral as a by-product — but the return driver is the funding, not the net direction.

The funding rank (shared by both books)

Each tick the producer pulls the live instrument board once and reads the per-asset funding field (hourly decimal) for every liquid main-DEX perp (top-universeMaxNames by 24h volume, then a relative-to-market floor: dayNtlVlm ≥ volFloorPctOfMedian × the cohort median — no hardcoded $ floor). Annualized ≈ funding × 24 × 365. The harvest book ranks funding descending (most positive); the payout book ranks ascending (most negative). Only the top rankPoolSize names (default 12) then get 1h+4h candles for confirmation.

> Data robustness: funding comes from the always-available instrument > board — NOT the ClickHouse-backed funding_history endpoint (which can > 503 or require elevated scope). funding_history is optional enrichment only.

HARVEST book — short the expensive longs (carry + fade)

Scoring (raw integer; minScore 4)

| Component | Pts | Source | |---|---|---| | Funding (annualized) | +3 (≥~88%/yr) / +2 (≥~53%) / +1 (≥~26%) / disqualify ( −floor) | instrument funding | | 4h trend | +2 BULLISH / +1 NEUTRAL / disqualify BEARISH | 4h candles | | RSI oversold | +1 (RSI ≤ rsiOversold 30) | 1h RSI | | 24h bounce | +1 (24h ≥ 0) / −1 (24h ≤ −5%, still crashing) | instrument ctx |

Disqualifies a fresh 4h downtrend — never longs a knife. Longs only crowded shorts that are capitulating.

Execution & exit (both books)

• slots 4, margin_pct 18%, tick 300s
• strict 5x leverage clamp, then each asset's Hyperliquid venue max
• DSL carry management (tighter than a momentum leg): carry P&L per period

is small, so a price loss must be cut before it dwarfs the funding — phase1 max_loss 10% / retrace 6 / 1 breach; weak_peak_cut ON (4h @ 1.5), dead_weight_cut ON (8h), hard_timeout 3d (funding regimes decay); phase2 ladder 6%→lock0 / 14%→45 / 28%→65 / 50%→80 / 90%→90

Leverage clamping (both books)

Desired leverage = the book cap (maxLeverage 5). The producer clamps to each asset's Hyperliquid venue max; the runtime gate rejects any leverage above 5.

XYZ handling

Camel ranks the main-DEX crypto cross-section only — XYZ funding is sparse. The main/xyz clearinghouse sections are two VIEWS of ONE cross-margined wallet, so get_positions() takes accountValue ONCE via max() — never sums.

Risk gates (risk.guard_rails)

| Gate | harvest | payout | |---|---|---| | dailylosslimitpct | 10 | 10 | | maxentriesperday | 6 | 6 | | maxconsecutivelosses | 4 | 4 | | cooldownminutes | 60 | 60 | | drawdownhaltpct | 18 | 18 | | perassetcooldownminutes | 180 | 180 | | dataretentionhours | 96 | 96 | | drawdownresetondayrollover | true | true |

Entries and exits both use FEE_OPTIMIZED_LIMIT (ensure_execution_as_taker true; 45s maker-first window). Fee note: carry P&L is small per period — keep turnover modest so fees don't eat the funding collected.

Files

| File | Purpose | |---|---| | runtime-harvest.yaml | Harvest-book runtime spec (wallet, DSL, risk, LLM gate) | | runtime-payout.yaml | Payout-book runtime spec | | scripts/camel-producer.py | Book-aware producer daemon (one script, both books) | | scripts/camel_config.py | Leg resolution + SenpiClient wrapper + helpers | | config/camel-harvest-config.json | Harvest-book tunables (funding floor/tiers) | | config/camel-payout-config.json | Payout-book tunables |

Operator install

See [README.md](README.md) — the two books are two daemons (CAMEL_LEG=harvest and CAMEL_LEG=payout) on two equally-funded wallets, each with its own runtime YAML.

Hard rule for user-conversation Claude sessions

User-conversation Claude sessions MUST NOT call any of: create_position, close_position, edit_position, ratchet_stop_add, ratchet_stop_edit, ratchet_stop_delete, cancel_order, strategy_close, strategy_close_positions.

These tools are reserved for the producer daemon (entry path) and the DSL ratchet engine (exit path). User-conversation sessions are read-only. Each producer daemon handles real signals on its next tick.

License

Apache-2.0 — Copyright 2026 Senpi (https://senpi.ai)

Source & license

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

• Author: Senpi-ai
• Source: Senpi-ai/senpi-skills
• License: MIT

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