In this section, we primarily analyze the specific types of user requirements, namely Requirement.
In our study, we further classified the 809 feature requests into seven categories of Requirement to describe the concrete manifestations of user needs during usage.
R1. Integration Requirement
The kind of requirements aims to integrate external tools, services, model providers, or third-party libraries with the agent framework to extend its capabilities.
For example:
AutoGen4135 requests the integration of a local LLM service (Ollama) as a new model provider into Magentic-One.
R2. Feature Proposal
The kind of requirement calls for introducing entirely new features, tools, integrations, or capabilities, thereby extending the functional boundaries of the agent framework.
For example:
In CrewAI113, the Assistant does not support any command prefixes (e.g., `/d`, `/e`). The user proposes adding a command system to trigger specific behaviors, such as retrieving documentation, providing examples, performing searches, etc.
R3. Feature Enhancement
The kind of requirement focuses on optimizing, strengthening, or adjusting existing functionality to improve performance, usability, flexibility, or overall user experience.
For example:
LangChain33157 proposes an enhancement to the existing `retriever.with_listeners(on_end=...)` callback by additionally returning the embedding vector of the input query, thereby eliminating the need for users to invoke the embedding interface a second time.
R4. Documentation Improvement
The kind of requirement requests updates, additions, or refinements to documentation.
For example:
LangChain35360 proposes the inclusion of an official RAG troubleshooting checklist page in the documentation, supplemented by example code or optional lightweight utilities, to guide users in debugging RAG systems using the WFGY 16-problem map.
R5. Infrastructure Optimization
The kind of requirement concerns improvements to foundational support systems, including build processes, dependency management, packaging, deployment, and environment configuration.
For example:
CrewAI3026 proposes refactoring the framework into a minimal core library, with additional functionality provided as optional extension packages, thereby reducing the installation footprint and mitigating the risks associated with dependency vulnerabilities.
R6. Others
The requirement in agent frameworks encompasses issues that do not fit into any of the previous five categories.