Extension Development Guide

Welcome to participate in the secondary development of CatInCup. The design philosophy of this framework is modularity and decoupling.

1. Skill Development (No-Code/Low-Code Extension)

Follow Anthropic's official specifications. A Skill is a directory under data/skill, with the core being the SKILL.md file.

• Requirements: SKILL.md must include name and description metadata at the top.

• Content: Use natural language to detailedly describe the workflow and constraints of the skill, and you can put scripts together in the folder.

2. Native Plugin Development

Native plugins are used to expand the framework's hard capability boundaries (Python library level). They are stored in src/plugins/plugin_collection and must contain three files:

__init__.py       # Plugin's exposed function interface
plugin_name.py    # Plugin's specific business logic implementation
plugin_name.yaml  # Detailed description of parameters and Schema (refer to existing plugins for format)

Note: Be sure to handle third-party dependency isolation and introduction in the code.

3. Harness Engineering Development

For certain specific industries, you can customize your own Harness Engineering according to the industry's work characteristics.

In this framework, we provide developers with a BaseTask class, defined as a general task, providing many basic built-in functions and class attributes.

Developers can define their own Harness Engineering in the src/models/expert/ folder, that is, give a class inheriting BaseTask, modify or use the basic built-in functions (refer to the trading task), and implement their own workflow.

The expert interface provides developers with out-of-the-box native tools and plugin ecosystem, and enjoys the framework's API polling, concurrency compatibility and other features

4. Sensor Development (Environmental Sensing)

Sensors are the tentacles through which the Agent connects to the physical world and external applications (such as RSS subscriptions, external chat software).

• Implementation: Write Python scripts, start independent background threads to continuously monitor external events.

• Data Push: After monitoring events, call the add_message function to asynchronously push external information into the Agent's message queue for processing. You can combine with Cron tools to develop time-based定时 Sensor triggers.

5. Contribution Guide

• When submitting a PR, please ensure one PR solves one specific problem to avoid giant mixed submissions.

• When encountering bugs or having new Feature ideas, please first submit an Issue on GitHub for discussion.