Use when you have a written implementation plan to execute in a separate session with review checkpoints
npx skills add clawd-aftermath/senpi-skills-af --skill "owl-strategy"
Install specific skill from multi-skill repository
# Description
>-
# SKILL.md
name: owl-strategy
description: >-
OWL v5.2 — Pure contrarian. One scanner, one thesis: the crowd is wrong. Monitors crowding
across top 30 assets (funding extremity, OI concentration, SM tilt). When crowding persists
4+ hours AND exhaustion signals fire (volume declining, price stalling, RSI divergence),
enters AGAINST the crowd to ride the liquidation unwind. 1-2 trades per day max.
Re-crowding exit: if the crowd comes back, thesis is dead, exit immediately.
DSL High Water Mode (mandatory). The patient predator.
v5.2: funding floor lowered from 20% to 12% so the five-factor scoring model actually runs.
Added observability logging (top 3 crowding scores per scan cycle).
license: Apache-2.0
metadata:
author: jason-goldberg
version: "5.2"
platform: senpi
exchange: aftermath
OWL v5.2 — Pure Contrarian
Wait for the crowd to overcommit. Wait for them to exhaust. Then eat their liquidations.
One scanner. One thesis. Every other skill in the zoo enters WITH momentum, WITH the trend, WITH smart money. OWL is the only skill that enters AGAINST the crowd. The edge: crowded trades unwind violently and predictably.
v5 is a complete rebuild. v1-v4 had 3 scanners (contrarian + momentum + correlation). The momentum and correlation scanners caused the agent to enter WITH the crowd while thinking it was contrarian. v5 has one scanner that does one thing: find crowded assets that are exhausting.
v5.2 fix: funding floor lowered from 20% → 12%. At 20%, the funding gate hard-blocked every asset before SM concentration, OI concentration, or any other signal was evaluated. Using the correct Aftermath funding calculation (premiumTwap / indexPrice from all-markets), rates rarely exceed ~11% annualized in normal conditions, so the five-factor scoring model was never running. At 12%, funding must still be meaningfully elevated, but assets with strong SM/OI crowding signals can now accumulate a score. The minCrowdingScore of 8 remains the real quality gate.
v5.2 also adds observability logging. Every scan cycle logs the top 3 crowding scores and active persistence timers to stderr (internal log only, not notifications). This lets us diagnose "is OWL seeing anything?" without changing config or asking the agent.
MANDATORY: DSL High Water Mode
OWL MUST use DSL High Water Mode. This is not optional. Do not substitute standard DSL tiers.
On Aftermath, initial SL is set inline with the entry order via the slTp field. DSL cron handles trailing updates only.
Spec: https://github.com/Senpi-ai/senpi-skills/blob/main/dsl-dynamic-stop-loss/dsl-high-water-spec%201.0.md
When creating DSL state files for any OWL position, you MUST include:
{
"lockMode": "pct_of_high_water",
"phase2TriggerRoe": 10,
"tiers": [
{"triggerPct": 10, "lockHwPct": 20, "consecutiveBreachesRequired": 3},
{"triggerPct": 20, "lockHwPct": 40, "consecutiveBreachesRequired": 3},
{"triggerPct": 35, "lockHwPct": 60, "consecutiveBreachesRequired": 2},
{"triggerPct": 50, "lockHwPct": 75, "consecutiveBreachesRequired": 1},
{"triggerPct": 75, "lockHwPct": 85, "consecutiveBreachesRequired": 1}
]
}
FALLBACK: Use tiersLegacyFallback from config until engine supports pct_of_high_water.
How OWL v5 Works
The Three Phases (all must pass before entry)
Phase 1 — CROWDING (score ≥ 8)
Scan top 30 assets every 15 minutes. Score how one-sided each asset's positioning is:
| Signal | Max Points |
|---|---|
| Funding extremity (annualized rate) | 4 |
| SM concentration (leaderboard tilt) | 4 |
| OI concentration (USD-weighted) | 2 |
| SM confirms funding direction | 1 |
Most assets score 0-3 (not crowded). Only assets scoring 8+ advance.
Phase 2 — PERSISTENCE (4+ hours)
Crowding must persist for at least 4 hours. A brief funding spike that resolves in 30 minutes is noise. True crowding builds over hours — funding stays extreme, OI keeps growing, SM stays tilted. The longer the crowding persists, the more violent the eventual unwind.
Phase 3 — EXHAUSTION (score ≥ 5)
The crowd is positioned, and they've been positioned for hours. Now: are they exhausting? Four signals:
| Signal | Points | What It Means |
|---|---|---|
| Volume declining (recent vs 6h avg) | 3 | Conviction leaving — nobody new is entering |
| Price stalling (crowd long but price flat) | 3 | The trade stopped working — crowd is trapped |
| Volume spike without follow-through | 2 | Capitulation wick — someone tried to push, failed |
| 4h RSI divergence | 2 | Momentum dying despite positioning |
Entry
Total score (crowding + exhaustion) must be ≥ 14. Entry direction is OPPOSITE to the crowd. If the crowd is long, OWL goes short. If the crowd is short, OWL goes long.
This means OWL enters 1-2 times per day at most. Often zero. That's by design.
Pre-Entry Validation (Aftermath)
Before placing any order, preview it via POST /api/perpetuals/account/previews/place-limit-order (or place-market-order).
Check before executing:
- error field present -> abort and log reason
- percentSlippage above threshold (for example 0.5%) -> abort or reduce size
- collateralChange breaches per-position risk cap -> reduce size
- hasPosition: true when expecting a fresh entry -> refresh state and re-evaluate
Preview is free (no gas cost, no state change). Always preview before committing real funds.
Hold
Every 15-minute scan re-evaluates held positions FIRST. The position holds as long as the crowd doesn't come back.
Re-Crowding Exit (unique to OWL)
If the crowd rebuilds in their original direction (re-crowding score ≥ 6), the unwind thesis is dead and the position exits immediately. This is OWL's equivalent of SCORPION's "sting" — an instant, thesis-based exit that overrides DSL.
DSL: Widest in the Zoo
Contrarian entries retrace hard before working. The crowd doesn't unwind smoothly — they fight back first. OWL needs the widest DSL bands of any skill.
| Setting | Value | Compare to FOX |
|---|---|---|
| Phase 1 floor | 4% notional (~40% ROE at 10x) | FOX: 1.5% |
| Phase 2 trigger | +10% ROE | FOX: +7% |
| T1 lock | 20% of HW | FOX: 40% |
| 85% trail at | +75% ROE | FOX: +20% |
| Stagnation TP | 15% ROE, 120min | FOX: 10%, 45min |
| Time exits | All disabled | FOX: 30min hard |
The tradeoff: OWL loses bigger on losers (-35 to -40% ROE) but catches crowding unwinds that produce 50-200%+ ROE when the cascade hits.
Risk Management
| Rule | Value | Notes |
|---|---|---|
| Max positions | 2 | Rare, concentrated bets |
| Max entries/day | 2 base, up to 4 on profitable days | |
| Phase 1 floor | 4% notional | Widest in the zoo |
| G5 per-position cap | 10% of account | Wider than most — contrarian needs room |
| Drawdown halt | 25% from peak | |
| Max consecutive losses | 2 → 180 min cooldown | Long cooldown — if the contrarian thesis failed twice, something changed |
| Re-crowding exit | Immediate | If the crowd comes back, thesis is dead |
| Loss cooldown per asset | 6 hours | Don't re-enter same contrarian trade too soon |
Aftermath Funding Rate Integration
CRITICAL: Do NOT use estimatedFundingRate from ticker/market endpoints for OWL scoring. On Aftermath it is 12.5x inflated versus on-chain settlement behavior.
Use /api/perpetuals/all-markets and compute funding from market state:
const ratePerPeriod = market.marketState.premiumTwap / market.indexPrice;
const annualizedPct = ratePerPeriod * 3 * 365 * 100;
ratePerPeriodis per 8h funding as a decimal fraction- annualized percentage is
ratePerPeriod * 3 * 365 * 100 - example:
0.0001=0.01%per 8h =0.03%per day =10.95%annualized
Use this corrected method for OWL's funding extremity points and persistence checks.
Cron Architecture
| Cron | Interval | Session | Purpose |
|---|---|---|---|
| Scanner | 15 min | isolated | Crowding scan + exhaustion detection + re-crowding check |
| DSL v5 | 3 min | isolated | High Water Mode trailing stops |
15-minute scanner interval is intentional. Crowding builds over hours, not minutes. Scanning every 3 minutes would waste tokens on data that hasn't changed. The DSL cron still runs every 3 minutes for trailing stop protection.
Notification Policy
ONLY alert: Position OPENED (asset, direction, crowding score, exhaustion signals, how long crowded), position CLOSED (DSL or re-crowding exit with reason), risk guardian triggered, critical error.
NEVER alert: Scanner found no crowding, scanner found crowding but no exhaustion, persistence tracking updates, DSL routine check, any reasoning.
All crons isolated. NO_REPLY for idle cycles.
Bootstrap Gate
Check config/bootstrap-complete.json every session. If missing:
1. Verify Senpi MCP
2. Create scanner cron (15 min, isolated) and DSL cron (3 min, isolated)
3. Write config/bootstrap-complete.json
4. Send: "🦉 OWL v5 is online. Pure contrarian. Scanning for crowded exhaustion. Silence = no opportunity."
Expected Behavior
| Metric | Expected |
|---|---|
| Trades/day | 0-2 (crowding unwinds are rare) |
| Avg hold time | 4-24 hours |
| Win rate | ~45-55% (wider stops, contrarian timing is hard) |
| Avg winner | 40-150%+ ROE (crowding unwinds are violent) |
| Avg loser | -25 to -40% ROE (wide floors, structural invalidation) |
| Fee drag/day | $2-8 (very few trades, all maker entries) |
| Profit factor | Target 1.3-2.0 (big winners compensate for wider losers) |
Files
| File | Purpose |
|---|---|
scripts/owl-scanner.py |
Crowding + exhaustion + re-crowding — the only scanner |
scripts/owl_config.py |
Shared config, MCP helpers |
config/owl-config.json |
All variables with DSL High Water tiers + legacy fallback |
License
Apache-2.0 — Built by Senpi (https://senpi.ai). Attribution required for derivative works.
Source: https://github.com/Senpi-ai/senpi-skills
# Supported AI Coding Agents
This skill is compatible with the SKILL.md standard and works with all major AI coding agents:
Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.