open claw的数据标注
Document
Data CollectionProject Description (Directly Fillable Version) The OpenClaw Agent Data Collection Project is a human-machine dialogue data labeling initiative designed for training AI agents. Project Scope: Simulate real users engaging in multi-turn English natural conversations with the OpenClaw Agent, collect interaction data that meets training standards, and support model capability evaluation and the implementation of labeling specifications. Core Data Labeling Tasks: Responsible for dialogue flow control, guiding tool invocation behaviors, and verifying the compliance of dialogue data, ensuring conversations fully cover task objectives while recording and annotating key interaction nodes and tool invocation actions. Project Scale: Each batch requires completing ≥5 rounds of effective dialogue interactions, triggering ≥5 total tool invocations across ≥3 different types of tools to form a standardized human-machine interaction dataset. Quality Metrics: 100% task relevance rate, ≥80% tool invocation success rate, interaction logic consistent with real human behavior, and final deliverables that fully meet all requirements in the task description without omissions or deviations.