Egret Strategy

🐦 EGRET v1.0.0 — Smart-Money Divergence Fader

When the crowd is all-in and price won't follow, fade the crowd. Egret looks for extreme Smart-Money concentration that price is not confirming — the setup that precedes a crowded-side unwind — and takes the other side.

Why this strategy exists

Smart-Money concentration is usually a confirmation signal (the fleet's momentum agents trade with it). But at an extreme — when 70–85% of top traders are piled onto one side — it flips into a contrarian signal, if price refuses to follow. A crowd that's maximally long while price stalls or rolls over has no one left to buy; the unwind is the edge.

Egret is the mirror image of the fleet's momentum agents:

• They trade with SM when price confirms.
• Egret trades against SM when price diverges from an extreme crowd.

CRITICAL RULES

RULE 1: Crowding extreme AND price divergence — both required

SM concentration must be >= crowdingMinPct (default 70%) and price must be failing to confirm the crowd over the lookback window (crowded long but price not rising, or crowded short but price not falling). Extreme crowding with a confirming price is a working trend — Egret does NOT fade that.

RULE 2: This is a fader — tuned opposite the momentum agents

Egret is mean-reversion, so its risk profile is the inverse of Badger/Hawk/Bison (the 2026-05-21 per-strategy-class rule):

• TIGHT Phase 2 ladder (T0 +5% / lock 30) — a fade is a bounded snapback; bank it fast. Holding a fade for a +50% trend would violate the thesis.
• MAKER-ONLY entry (ensure_execution_as_taker: false) — a fade is non-urgent and fee-sensitive; if the maker can't fill, the trade wasn't ripe.
• Time-cuts ENABLED — a fade resolves quickly or the thesis failed.

RULE 3: Producer enters. DSL exits.

No close_position call site. DSL Phase 1 + tight Phase 2 + hardtimeout 48h + weakpeak_cut own all exits.

How Egret scores a trade

Gates (all required):

1. SM concentration >= crowdingMinPct (default 70%)
2. Price diverging from the crowd over the lookback (not confirming)

Score components (max ~9):

| Signal | Points | |---|---| | SM crowded (gate-confirmed) | +2 | | SM ultra-crowded (≥ crowdingExtremePct, default 80%) | +1 | | Price clearly diverging (≥ 0.5% against the crowd) | +3 | | Price merely stalled | +2 | | RSI exhausted (≥ 68 for fade-short / ≤ 32 for fade-long) | +2 | | RSI stretched (≥ 60 / ≤ 40) | +1 |

Floor: minScore: 5.

DSL preset (tight — bank the snapback)

| Phase | Component | Setting | |---|---|---| | Phase 1 | maxlosspct | 15% | | Phase 1 | retracethreshold | 6 | | Time cuts | hardtimeout | 48h (ENABLED) | | Time cuts | weakpeakcut | 120min / 2% (ENABLED) | | Time cuts | deadweightcut | DISABLED | | Phase 2 | T0 | +5% / lock 30% | | Phase 2 | T1 | +10% / lock 50% | | Phase 2 | T2 | +15% / lock 65% | | Phase 2 | T3 | +25% / lock 80% | | Phase 2 | T4 | +40% / lock 90% |

Scanner pattern

This strategy uses the Contrarian fader archetype (same family as Owl / Bald-eagle / Pangolin) with an SM-vs-price divergence trigger — see senpi-trading-runtime/references/producer-patterns.md. Primary MCP calls: leaderboard_get_markets (SM concentration), market_get_asset_data (price momentum + RSI).

Operator install

See [README.md](README.md).

Changelog

v1.0.0 (2026-05-21) — initial release

Adds an SM-divergence trigger to the contrarian-fader archetype. Built with the per-strategy-class DSL rule: tight ladder + maker-only entry + enabled time-cuts (the fader profile), the deliberate inverse of the momentum agents' wide ladder + taker-fallback.

License

MIT — Copyright 2026 Senpi (https://senpi.ai).

Skill Attribution

When creating a strategy, include skill_name and skill_version. See references/skill-attribution.md.

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.