gaaiyun/joinquant-skill

Jukuan Strategy Assistant · JoinQuant Skill

Let the Coding Agent first check the Jukuan API before writing the policy.

Quick Start · Ability Description · Validation and Boundaries · Related projects

JoinQuant API knowledge base, policy skeleton, and code checking tools for Coding Agents are used.

Coding agents like Cursor, Claude Code, and Codex are familiar with Python but may not understand the runtime environment and API constraints of JuBroad. They may invent functions, mix research and backtesting interfaces, ignore rebalancing and transaction costs, and ultimately provide a syntactically correct strategy that cannot run on JuBroad.

This repository brings together Jukuan API documentation, policy frameworks, static checks, factor examples, research report parsing processes, and MCP tools in one skill. Agents can load required data by task, and people can also directly use the CLI within it.

[! IMPORTANT] This repository can only check code structure and some JuBroad usage; it cannot verify strategy benefits, nor can it replace JuBroad Cloud’s compilation, backtesting, or simulated trading. lint By only stating that the code does not hit the current rule; You still need to check data timing, transaction costs, risk constraints, and execution conditions.

Navigation

• Quick Start
• What problems does it solve?
• Capability description
• Common workflows
• Installation
• MCP server
• Validate and use boundaries
• Project structure
• Related projects
• Star History

Quick Start

1. Clone warehouse

The following API search, policy generation, and lint commands do not require third-party dependencies. Testing dependencies for factors, factor_lab, PDF parsing, and MCP are available in the Optional Dependencies.

2. Generate a strategy framework

3. Check the strategy

After fixing errors and warnings, paste the code into Jukuan Web Editor and continue platform compilation, backtesting.

4. Let the Agent use this skill

After registration, submit the requirements directly to the Agent:

The agent reads the corresponding template and reference according to the task, checks the API, and then generates code.

For detailed installation steps, see INSTALL_CN.md, and for the complete development process, see WORKFLOW.md.

What problems does it solve?

API Illusion

jqdata.get_stock_data(), pandas.DataFrame.diff_in_diff() Functions like jqdata.get_stock_data(), pandas.DataFrame.diff_in_diff() do not exist. The repository provides references split by topic, raw API data search, and lint rules, allowing agents to verify before generating code.

Mixing Operating Environments

The same function may have different parameters and return structures in the research environment, backtest environment, and local jqdatasdk. Each reference is organized around specific use cases, avoiding sticking entire local research code into the aggregation policy.