Mini-Agent

Mini-Agent is an extremely lightweight open-source single-agent framework centered on the ReAct (Reasoning and Acting) loop, consisting of an agent main loop, an LLM client adapter layer, and a tool system with built-in file and shell tools as well as custom extensions. Through an adapter pattern, it natively supports Anthropic and OpenAI protocols, and can integrate MCP tools and Claude Skills, with conversation memory and automatic context compaction. The entire project is only about 3,300 lines of Python, making it easy to learn and develop rapidly.

Strengths: Clean, concise code and clear architecture allow developers to get started quickly and perform secondary development based on its CLI demo or framework for tasks such as automated weekly report merging, data analysis, copywriting, etc. Its lightweight design makes customization and embedding straightforward.

Limitations: Production stability is moderate, not suitable for enterprise environments requiring high reliability; extensibility is medium, mainly focused on single-agent scenarios and lacks built-in multi-agent collaboration support.