GitHub Topics: gym, agent-environment, registry, domain-profile, pydantic, mcp

为 knowlyr-agent 生态提供统一的 Gym 风格环境协议、环境注册表和领域配置，是其余 5 个子包的共享基础。

核心能力 / Core Capabilities

AgentEnv 协议 → TimeStep 返回 → Registry 注册/查找 → DomainProfile 领域配置

设计特点 / Design Highlights

安装 / Installation

pip install knowlyr-core

开发模式：

pip install knowlyr-core[dev]

快速开始 / Quick Start

环境协议 / Environment Protocol

from knowlyrcore import AgentEnv, TimeStep, make, register

# 注册自定义环境
class MyEnv(AgentEnv):
    domain = "coding"

    def reset(self, *, task=None, seed=None) -> TimeStep:
        return TimeStep(observation="ready")

    def step(self, action: dict) -> TimeStep:
        return TimeStep(observation="done", terminated=True)

    def close(self): ...

    @property
    def available_tools(self) -> list[str]:
        return ["bash", "submit"]

register("my/env", MyEnv, domain="coding")
env = make("my/env")
ts = env.reset()

领域配置 / Domain Profiles

from knowlyrcore import get_domain_profile, list_domain_profiles

# 查看所有领域
print(list_domain_profiles())
# ['coding', 'browser', 'conversation', 'engineering', 'advisory', 'discussion', 'generic']

# 获取领域工具列表
profile = get_domain_profile("coding")
for tool in profile.tools:
    print(f"{tool.name}: {tool.description}")

包装器组合 / Composable Wrappers

from knowlyrcore.wrappers import MaxStepsWrapper, TimeoutWrapper

env = make("my/env")
env = MaxStepsWrapper(env, max_steps=30)
env = TimeoutWrapper(env, timeout=300)

API 概览

License

MIT

AI Data Pipeline 生态

10 个工具覆盖 AI 数据工程全流程，均支持 CLI + MCP，可独立使用也可组合成流水线。

graph LR
    A[Radar] --> B[Recipe] --> C[Synth] --> E[Check] --> F[Audit] --> G[Hub]
    B --> D[Label] --> E
    G --> H[Sandbox] --> I[Recorder] --> J[Reward]
    style H fill:#e1f5fe