Knuckles-Team Strategy Lifecycle Workflow
[!IMPORTANT] This is the master pipeline. Every strategy must pass through ALL gates before live activation.
Workflow Execution Steps
Step 1: hypothesis-generation
Agent: quant_research_analyst
Generate trading hypothesis from KG signals, market data, and research.
Tool: emerald_signals(action="alpha", ticker=...)
Step 2: trading-debate
Agent: chief_trading_officer
Run Bull/Bear multi-round debate to vet the hypothesis.
Tool: graph_orchestrate(action="start_debate", task=hypothesis)
Step 3: risk-assessment
Agent: risk_compliance_officer
Pre-trade risk assessment and Kelly criterion position sizing.
Tool: emerald_risk(action="kelly", win_rate=..., win_loss_ratio=...)
Step 4: backtest
Agent: quant_research_analyst
Rigorous backtesting via data-science-mcp on GPU node.
Tool: graph_analyze(action="evaluate_alpha", query=hypothesis)
Step 5: strategy-promotion
Agent: chief_trading_officer
Promote strategy: Draft → Backtesting → Paper → Live.
Tool: emerald_strategy(action="promote", strategy_id=...)
Step 6: human-approval
Agent: chief_trading_officer
REQUIRED: Human approval gate for live trading activation.
Tool: graph_orchestrate(action="request_approval", task="live_trading_activation")
Execution
Run this workflow as a dependency-ordered DAG. Steps with no unmet depends_on run in parallel; dependents run after their prerequisites complete.
- Run first (in parallel): Step 1 — hypothesis-generation; Step 2 — trading-debate; Step 3 — risk-assessment; Step 4 — backtest; Step 5 — strategy-promotion; Step 6 — human-approval
Execution: If graph-os is reachable, offload the whole DAG via graph_orchestrate action=execute_workflow (or the kg-delegate skill) for true parallel/swarm execution. Otherwise execute the steps natively in dependency order: run steps with no unmet depends_on in parallel, then their dependents.