🐺 JACKAL v1.0 — The Smart Stalker

RULE 5: Never act on a signal without independent TA confirmation.

RULE 6: Per-source exposure cap 40% — no single trader dominates.

RULE 7: HARD_STOP circuit breaker triggers at -25% drawdown.

Why This Agent Exists

Every other Senpi predator generates signals from MARKET data — price, volume, SM positioning, funding. None use the actions of other top-performing Senpi users as a signal source. That's a missing layer.

The new any-user-lookup MCP capability (strategy_list with userIds filter, plus strategy_get_clearinghouse_state on any wallet) makes it possible to observe what other Senpi users are doing in real time. Jackal turns that observation into a trading signal.

Naive "copy the winner" fails because:

• Winners regress (pr0br000 went +$1,275 week 3 → -$500 week 4)
• Signals lag (by the time you copy, trade is half-over)
• Style-mismatches amplify risk (scalpers' signals aren't copyable)
• No filter layer means you copy losers too

Jackal solves these with a 4-layer funnel.

Architecture: 4-layer funnel

LAYER 1 — ELIGIBILITY  (every 6h full refresh, 1h rescore)
  Trajectory-based scoring (not rank-based):
    pnl_trajectory  35%   — improving trend beats absolute level
    rank_velocity   25%   — rising fast catches emerging movers
    recent_outcomes 15%   — last 5 trades W/L pattern
    win_loss_ratio  10%   — classic asymmetry
    fee_efficiency  10%   — not overtrading
    consistency      5%   — positive days in last 14
  
  Hard disqualifiers:
    - fee_drag > 40% (overtrader)
    - avg_winner_hold < 2h (scalper — can't copy at our latency)
    - drawdown_24h > 10% (active blowup)
    - ratio < 1.5 (no edge)
    - <10 closed trades or <7 days old (insufficient history)

LAYER 2 — POSITION DISCOVERY  (every 3 min)
  For each Active Pool member: strategy_get_clearinghouse_state
  (with forceFetch=true for <60s latency)
  Diff against last-known → detect NEW positions
  Track first_seen_ts for age filtering

LAYER 3 — SIGNAL SCORING  (per detected position)
  score = 
    0.40 × source_quality      (avg quality_score of source traders)
  + 0.30 × consensus           (1.0x for 1 source, 1.8x for 2, 3.0x for 3+)
  + 0.30 × position_freshness  (peaks at 1h, declines past 4h)
  + GOLD_BONUS (+15)           (newly-promoted source + 2+ pool consensus)

LAYER 4 — INDEPENDENT CONFIRMATION + EXECUTION
  Senpi TA gate: 4h trend, 1h momentum, SM alignment
  If any fails: SKIP even at high score
  If passes: size by score tier (20/35/55% margin × 3/5/7x leverage)
  Own DSL manages exit — never wait for source to close

Two-tier pool mechanics

Watchlist (~100–200 traders):

• Pulled from Arena leaderboard + top Senpi points leaderboard
• Polled for re-eligibility every 6h (full refresh) / 1h (rescore)
• Jackal does NOT act on their trades
• Entry/exit automatic — crosses score thresholds to promote

Active Pool (~20–30 traders):

• Quality score ≥ 70 for 6+ consecutive hours
• Polled every 3 min (scan) with forceFetch=true
• Jackal acts on their trades subject to consensus + TA filters
• Demoted if score < 50 for 12h OR 24h drawdown > 10%
• 48h re-promotion cooldown after demote

Catching pr0br000 counterfactual (Arena Weeks 2-3):

Jackal would have acted on pr0br000's trades from April 3 to April 16 — catching WLD, HEMI, MON winners. Auto-demoted BEFORE the April 17 ZEC disaster because the 24h drawdown trigger fires at -10%.

The GOLD SIGNAL

The highest-conviction pattern: a trader who was just promoted from Watchlist to Active Pool (within last 24h) opens a position that 2+ already-Active Pool members are also in. This means:

• A new rising star is agreeing with proven smart money
• Signal is BOTH new (emerging mover) AND confirmed (consensus)
• Historical win pattern aligns with existing pool

When Jackal detects a GOLD SIGNAL:

• Max sizing: 55% margin × 7x leverage
• Bypass score tiers (gold_signal_flag = true)
• Still requires TA confirmation — no TA override

This is where Jackal's asymmetric upside lives.

Sizing tiers

Never exceeds 7x. Patient hold profile — multi-day consensus signals play out slowly. Leverage + fees would kill the edge.

DSL — Consensus-aware patience

Jackal inherits Python's patience DSL profile but with adjustments:

• Wider Phase-1 floor (22% vs Python's 20%) — consensus signals need breathing room
• Looser early Phase-2 (+6%→20% lock vs Python's +5%→15%) — even consensus trades need runway

Risk controls

What Jackal Does NOT Do

1. Does NOT use Senpi's native MIRROR strategy. Those are passive single-trader copies. Jackal is a multi-source custom strategy that uses MCP read capability + own execution.

2. Does NOT copy scalpers. Traders whose avg winning hold is <2h get filtered out. Latency makes their edge uncopyable.

3. Does NOT copy a trader's size or exit. Own sizing (conviction score) and own DSL. A source trader holding through drawdown doesn't mean we hold.

4. Does NOT require perfect source stability. Sources can lose trades — they just can't dominate the pool. Per-source 40% cap

   • auto-demotion on drawdown protects against one-source reliance.

5. Does NOT chase rank-1 traders who are peaking. Trajectory scoring means cooling-off stars get demoted naturally.

Install

mkdir -p /data/workspace/skills/jackal-strategy/{config,scripts,state}

curl -s https://raw.githubusercontent.com/Senpi-ai/senpi-skills/main/jackal/runtime.yaml -o /data/workspace/skills/jackal-strategy/runtime.yaml
curl -s https://raw.githubusercontent.com/Senpi-ai/senpi-skills/main/jackal/SKILL.md -o /data/workspace/skills/jackal-strategy/SKILL.md
curl -s https://raw.githubusercontent.com/Senpi-ai/senpi-skills/main/jackal/config/jackal-config.json -o /data/workspace/skills/jackal-strategy/config/jackal-config.json
curl -s https://raw.githubusercontent.com/Senpi-ai/senpi-skills/main/jackal/scripts/jackal-scanner.py -o /data/workspace/skills/jackal-strategy/scripts/jackal-scanner.py
curl -s https://raw.githubusercontent.com/Senpi-ai/senpi-skills/main/jackal/scripts/jackal_config.py -o /data/workspace/skills/jackal-strategy/scripts/jackal_config.py
curl -s https://raw.githubusercontent.com/Senpi-ai/senpi-skills/main/jackal/scripts/jackal_pool.py -o /data/workspace/skills/jackal-strategy/scripts/jackal_pool.py

Configure

sed -i 's/${WALLET_ADDRESS}/<YOUR_STRATEGY_WALLET>/' /data/workspace/skills/jackal-strategy/runtime.yaml
sed -i 's/${TELEGRAM_CHAT_ID}/<YOUR_TELEGRAM_CHAT_ID>/' /data/workspace/skills/jackal-strategy/runtime.yaml

Install runtime + create scanner cron

openclaw senpi runtime create --path /data/workspace/skills/jackal-strategy/runtime.yaml
openclaw senpi runtime list
# Create 3-minute cron: python3 /data/workspace/skills/jackal-strategy/scripts/jackal-scanner.py

First-run behavior

On the first run, Jackal has no pool yet. The scanner will:

1. Fetch Arena leaderboard + Senpi points leaderboard (~100-200 candidates)
2. Score each candidate's recent trade history + PnL trajectory
3. Build initial Watchlist
4. No candidate will be promoted to Active Pool yet (requires 6h consistency)
5. First real trades typically start 6-24 hours after first run

This is expected and desired. Patience from setup.

Operational cost

• Watchlist refresh: 6h cadence, ~100-200 MCP calls per refresh (batched over several minutes)
• Score refresh: 1h cadence, ~30-100 MCP calls per refresh
• Position scan: 3min cadence, ~20-30 MCP calls per scan
• Total: ~1.2 MCP calls/second sustained

Changelog

v1.0 (2026-04-17) — FIRST SHIP

• First secondary-signal agent in the fleet
• Two-tier pool architecture (Watchlist + Active Pool)
• Trajectory-based quality scoring (6 components, weighted composite)
• Emerging-mover detection via rank velocity
• GOLD SIGNAL pattern (new promotion + existing consensus)
• Consensus multiplier (3+ traders = 3.0x score boost)
• Independent Senpi TA confirmation gate
• Own sizing + own DSL (not native MIRROR)
• Hard disqualifiers: scalper filter, fee-drag cap, drawdown trigger
• Per-source 40% exposure cap
• 48h re-promotion cooldown
• Patient DSL profile (72h timeout, wide Phase 1)

License

MIT — Built by Senpi (https://senpi.ai). Source: https://github.com/Senpi-ai/senpi-skills