owl-strategy

🦉 OWL v8.0 — Pure Contrarian. Helpers-Native. Maker Exits.

Wait for the crowd to overcommit. Wait for them to exhaust. Then eat their liquidations.

One thesis: the crowd is wrong. Every other skill in the fleet 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.

---

What changed in v8.0.0

Plumbing-only migration. NO thesis change. v7.1's crowding/exhaustion scoring, persistence gates, MACRO_TREND_GATE, conviction leverage, DSL preset are all preserved verbatim.

Six-layer plumbing flip on @senpi-ai/runtime 1.1.0 (same pattern as Wolverine/Lemon/Cheetah v3.0+ migrations):

Layer	v7.1	v8.0
MCP calls	mcporter subprocess	SenpiClient.mcp_call() direct HTTPS
Signal emit	openclaw senpi external-scanner ingest subprocess	SenpiClient.push_signal() direct HTTP POST
Reentrancy	hand-rolled fcntl flock on producer.lock	producer_daemon.scanner_lock (stale-PID auto-recovery)
Tick scheduling	openclaw cron + agentTurn	long-lived producer_daemon process
/state alive_check	none	wallet=/scanner= kwargs → daemon self-terminates if runtime/scanner removed
Signal type tagging	inferred from payload	signal_type="OWL_CONTRARIAN_FADE" passed explicitly

Why v8.0: keeps Owl on the same plumbing as the rest of the fleet. Eliminates per-tick subprocess overhead (~30-60s → ~1-3s ticks). No thesis change — v7.1 was the most recent thesis-level work.

---

Thesis (preserved from v7.1 / v6.2)

• Crowding score: funding extremity (0-4) + SM tilt (0-3, +1 if confirms funding) + OI concentration (0-2). Floor 6.
• Persistence: 1+ hour above floor, with 2-tick tolerance for noise (v5.3)
• Exhaustion score: volume declining + price stalling + volume spike no follow-through + RSI divergence. ≥ 2 distinct signals, score ≥ 5.
• Combined score ≥ 12 to fire.
• Macro trend gate: block crypto fades when |BTC 4h move| > 3% (v7.1).
• 6h post-loss per-asset cooldown.
• Universe: ALL crypto perps with OI > $3M (v6.1 expansion, no top-30 truncation).
• XYZ banned (different unwind dynamics; needs separate calibration).

---

Files

File	Purpose
runtime.yaml	Runtime spec (scanners, actions, exit DSL, guard_rails) — schema version 1.8.0
scripts/owl-producer.py	Long-lived helpers-native producer daemon (15-min cadence)
scripts/owl_config.py	Shared MCP helper + _wrapper_client proxy + atomic state I/O
config/owl-config.json	Operator-tunable defaults (informational; producer constants WIN)

---

Producer behavior

Runs every 15 minutes (long-lived daemon, NOT openclaw cron). On each tick:

1. Resolve wallet (OWL_WALLET env var, falls back to config.wallet); fail loud if both empty.
2. Read account value via strategy_get_clearinghouse_state for sizing + dynamic cap.
3. Apply dynamic daily cap (defense-in-depth alongside runtime guard_rails).
4. Fetch universe: market_list_instruments (~80-100 crypto perps with OI > $3M).
5. Fetch SM positioning map (one MCP call shared across all assets).
6. MACRO_TREND_GATE: if |BTC 4h move| > 3%, skip tick.
7. Score crowding per asset + update persistence history.
8. Filter to persisted candidates (≥ 1h above crowding floor).
9. Detect exhaustion for persisted candidates only (saves MCP calls).
10. Filter combined score ≥ 12 + per-asset cooldown.
11. Emit top contrarian candidate via cfg._wrapper_client.push_signal(..., signal_type="OWL_CONTRARIAN_FADE", ...) with conviction-scaled leverage (7/8/10 by score).
12. Persist crowding history + cooldown state under state/<wallet-hash>/.

NO execution code. NO position-tracking. NO DSL state. The runtime owns all of that.

Direction is OPPOSITE of crowd direction. This is the entire edge — Owl is the only fleet agent that fades crowding.

---

Entry flow

Producer daemon tick (every 900s)
  ↓ Crowding score >= 6, persisted >= 1h
  ↓ Exhaustion score >= 5 with >= 2 signals
  ↓ Combined score >= 12
  ↓ SenpiClient.push_signal(scanner="owl_signals", signal_type="OWL_CONTRARIAN_FADE", ...)
Runtime (127.0.0.1:8787)
  ↓ Schema-validates fields against runtime.yaml
  ↓ LLM gate (decision_model = ${OWL_DECISION_MODEL})
  ↓ Pass-through unless malformed; rejects if direction == crowdDirection
  ↓ OPEN_POSITION via FEE_OPTIMIZED_LIMIT (maker-only, 60s, NO taker fallback)
DSL (runtime-managed)
  ↓ Phase 1 max_loss 35% / single-breach (wide for contrarian retrace)
  ↓ Phase 2 6-tier ladder starting at 5% trigger (Lemon-pattern first leg)
  ↓ hard_timeout 480m, weak_peak 120m@2%, dead_weight 30m
  ↓ Exit via FEE_OPTIMIZED_LIMIT (maker-first, 60s, taker fallback)

---

Required env vars

The runtime YAML uses these substitutions:

Var	Purpose
${WALLET_ADDRESS}	Strategy wallet address
${TELEGRAM_CHAT_ID}	Telegram chat ID for notifications
${OWL_DECISION_MODEL}	Bare model name for LLM gate. NO provider prefix.

The producer reads:

Var	Purpose	Default
SENPI_AUTH_TOKEN	Required for MCP calls + signal POST	— (required; producer fails loud)
OWL_WALLET	Wallet (must match runtime YAML's wallet). Agent-specific by design. Per Turbine v2.0.9 contamination fix.	falls back to config.wallet
OWL_MARGIN_PCT	Fraction of account value per slot	0.25
OWL_MIN_OI_USD	Override OI floor	3000000

---

Risk envelope (declarative, runtime-enforced)

Setting	Value
Slots	2
Margin per slot	25% of account value (default ~$250 on $1k)
Default leverage	8x (conviction tiers: 7x at 12-13, 8x at 14-15, 10x at 16+)
Daily loss halt	10%
Drawdown halt	25% (also producer-side circuit breaker)
Max entries per day	4 (runtime ceiling); producer dynamic cap by PnL
Max consecutive losses	4
Per-asset cooldown	360 min (6h post-loss)
Asset universe	All crypto perps with OI > $3M (XYZ banned)

Dynamic daily cap (producer-side, v6 carryover):

Account state	Max entries / day
PnL < -25%	0 (HARD STOP — runtime also halts)
PnL < -15%	1
PnL < -5%	2
Day realized PnL < +$150	2 (base)
Day realized PnL ≥ +$150	3
Day realized PnL ≥ +$400	4

---

Expected behavior

Metric	Expected
Crowded assets per scan	5-15 above floor
Persisted ≥ 1h	1-3 typically
Exhaustion confluence	0-1 per scan
Trades per day	0-2
Avg hold	1-8h
Win rate target	60-70% (contrarian unwinds, when timed right, are reliable)

Silence is correct. Owl waits for both the crowd state AND the exhaustion trigger. Most days, one or both is missing. Don't improvise.

---

v8.0.0 changelog

• Plumbing-only. Six-layer flip per Wolverine v5.0 / Lemon v2.0 template. Thesis preserved verbatim from v7.1.
• owl_config.py: mcporter_call() now routes through SenpiClient.mcp_call() (direct HTTPS). Exposes _wrapper_client proxy for push_signal().
• owl-producer.py: drops fcntl lockfile + subprocess.run([...ingest]). Uses producer_daemon(fn=main, interval_seconds=900, ...) and cfg._wrapper_client.push_signal(signal_type="OWL_CONTRARIAN_FADE", ...).
• runtime.yaml: schema version bump 1.7.1 → 1.8.0. Description rewritten. DSL preset, scanner owl_signals, action owl_entry, consecutive_breaches_required: 1 unchanged.
• State files unchanged. crowding-history.json / asset-cooldowns.json / trade-counter.json continue to live under state/<wallet-hash>/.

---

License

Apache-2.0 — Built by Senpi (https://senpi.ai). Attribution required for derivative works. Source: https://github.com/Senpi-ai/senpi-skills

---

Skill Attribution

When creating a strategy, include skill_name and skill_version in the call. See references/skill-attribution.md for details.