Quant Strategy Research Lab

Problem context

Quant strategy research is rarely a one-shot prompt. Strategy design, factor construction, backtesting, risk control, execution, and compliance all depend on shared context. Manually chaining prompts can cause parameter loss, weak traceability, and difficulty resuming interrupted work.

How it works

The skill organizes 15 research roles into a directed acyclic graph and exposes two usage modes:
- Toolbox mode: trigger a single role such as backtesting, risk analysis, or alpha research for a quick check.
- Pipeline mode: select a track such as core, factor, execution, macro, statistical arbitrage, ML, or market-making, then run a predefined role sequence, for example the core track 01→02→03→04→10.

Key steps include reading the role registry; initializing state/research-context.md as the state source; collecting market, capital scale, and main objective; loading the role template and injecting upstream parameters; dispatching a subagent; parsing [QAGATE]...[/QAGATE] for required fields, numerical sanity, and signatures; appending outputs to the state file; and suggesting the next role. It also supports resuming from a role, rerunning from a role, checking progress, and resetting the workflow.

Boundaries

It is better viewed as a quant research orchestrator than a trading system. Role outputs still need human review, especially backtest metrics, risk assumptions, and execution parameters. By default, quality warnings may be logged and the pipeline continues; strict review can be requested. Production use still requires real data sources, access control, audit logs, and compliance processes.