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 "mantis-strategy"
Install specific skill from multi-skill repository
# Description
>-
# SKILL.md
name: mantis-strategy
description: >-
MANTIS v2.0 — Momentum Event Consensus. Complete rewrite from v1.0. Uses
leaderboard_get_momentum_events (real-time threshold crossings) instead of
discovery_get_top_traders (stale positions). When 2+ quality SM traders
cross momentum thresholds on the same asset/direction within 60 minutes,
confirmed by market concentration + volume, MANTIS enters with the momentum.
Replaces both MANTIS v1.0 and SCORPION v1.1.
license: Apache-2.0
metadata:
author: jason-goldberg
version: "2.0"
platform: senpi
exchange: aftermath
MANTIS v2.0 — Momentum Event Consensus
Follow smart money ACTIONS, not stale positions.
Why v2.0 Is a Complete Rewrite
v1.0 was broken at the data source level. It used discovery_get_top_traders to find historically good whales, then read their current open positions via market_get_positions. This returned ALL positions including legacy ones — BTC shorts opened months ago at $89-108k showed up as "fresh 8-whale consensus" when they were actually stale hedges. The 30-minute aging filter only checked "how long have we been seeing this position in our scans," not when the whale actually opened it.
v2.0 uses leaderboard_get_momentum_events as the primary data source. These are real-time threshold-crossing events ($2M+/$5.5M+/$10M+) within a 4-hour rolling window. They capture when a trader's delta PnL crosses a significance level — this is an ACTION, not a stale position. Combined with leaderboard_get_markets for aggregate SM concentration and market_get_asset_data for volume confirmation.
SCORPION v1.1 is deprecated. It had the same stale-position problem with even looser filters (2 whales, 10min). Result: 406 trades, -24% ROI. MANTIS v2.0 replaces both skills.
MANDATORY: DSL High Water Mode
MANTIS MUST use DSL High Water Mode. This is not optional.
Spec: https://github.com/Senpi-ai/senpi-skills/blob/main/dsl-dynamic-stop-loss/dsl-high-water-spec%201.0.md
DSL tiers in mantis-config.json. Arm DSL immediately after every entry fill. Zero naked positions.
Five-Gate Entry Model
Gate 1 — Momentum Events (primary signal)
Poll leaderboard_get_momentum_events for recent tier 1+ crossings within the last 60 minutes. These are traders whose delta PnL just crossed $2M+ (T1), $5.5M+ (T2), or $10M+ (T3).
Each event includes:
- top_positions: snapshot of which markets drove the momentum, with direction and delta PnL
- concentration: how focused the trader's gains are (0-1)
- trader_tags: TCS (Elite/Reliable/Streaky/Choppy) and TAS (Patient/Tactical/Active/Degen)
- tier: 1/2/3 significance level
Group events by asset+direction. 2+ unique traders on the same asset/direction = consensus.
Gate 2 — Trader Quality Filter
Not all momentum events are equal. Filter by:
- TCS (Trading Consistency Score): Only Elite and Reliable. Streaky/Choppy traders filtered out.
- TAS (Trading Activity Style): Block Degen. Allow Patient, Tactical, Active.
- Concentration ≥ 0.4: Trader's gains are focused, not spread thin across 20 positions.
Gate 3 — Market Confirmation
Call leaderboard_get_markets to check aggregate SM concentration on the asset. Requires 5+ top traders with significant positions in this market. This confirms the momentum events aren't isolated — the broader SM community is active here.
Gate 4 — Volume Confirmation
Call market_get_asset_data to check 1h volume. Current volume must be ≥ 50% of 6h average. Don't enter into dead markets where you'll get chopped up on spread.
Gate 5 — Regime Filter (penalty, not block)
Check BTC 4h regime. If entering counter-trend (long in bearish, short in bullish), apply -3 score penalty. SM momentum may override regime, so this is a penalty not a block. Regime-aligned entries get +1 bonus.
Scoring
| Component | Points | Notes |
|---|---|---|
| Trader count | 2 per trader | Core signal strength |
| Avg tier | 1-3 | Higher tiers = bigger moves |
| Avg concentration | 1-2 | High conviction traders |
| Market confirmation | 1-2 | Hot market bonus |
| Volume strength | 0-1 | Strong volume bonus |
| Regime alignment | -3 to +1 | Penalty/bonus |
Minimum score: 10 to trigger entry.
Entry Direction
MANTIS enters WITH the smart money momentum. If SM is long, MANTIS goes long. This is follow-the-leader, not contrarian (that's OWL's job).
Risk Management
| Rule | Value |
|---|---|
| Max positions | 3 |
| Max entries/day | 3 base, up to 6 on profitable days |
| Absolute floor | 3% notional |
| Drawdown halt | 20% from peak |
| Daily loss limit | 8% |
| Cooldown after 3 consecutive losses | 90 min |
| Stagnation TP | 10% ROE stale for 1 hour |
Cron Architecture
| Cron | Interval | Session | Purpose |
|---|---|---|---|
| Scanner | 5 min | isolated | Momentum events + consensus + all gates |
| DSL v5 | 3 min | isolated | High Water Mode trailing stops |
Both crons MUST be isolated sessions with agentTurn payload. Use NO_REPLY for idle cycles.
Notification Policy
ONLY alert: Position OPENED (asset, direction, trader count, avg tier, score breakdown), position CLOSED (DSL or thesis exit with reason), risk guardian triggered, critical error.
NEVER alert: Scanner found no events, quality filter removed events, no consensus, gates failed, DSL routine check, any reasoning.
What "Working Correctly" Looks Like
- Zero trades for hours: Normal. Momentum events that pass ALL five gates are rare by design.
- Events seen but no consensus: The quality filter and 2+ trader requirement are doing their job.
- Consensus found but score too low: Volume or regime gates keeping us out of bad setups.
- 1-2 trades per day: Ideal. This is a patience skill.
Bootstrap Gate
On EVERY session, check if config/bootstrap-complete.json exists. If not:
1. Verify Senpi MCP
2. Create scanner cron (5 min, isolated) and DSL cron (3 min, isolated)
3. Write config/bootstrap-complete.json
4. Send: "🦗 MANTIS v2.0 is online. Monitoring momentum events for SM consensus. Silence = no conviction."
Files
| File | Purpose |
|---|---|
scripts/mantis-scanner.py |
Five-gate scanner (momentum events → quality → markets → volume → regime) |
scripts/mantis_config.py |
Shared config, MCP helpers, state I/O |
config/mantis-config.json |
All configurable variables with DSL High Water tiers |
# 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.