Mini-Coding-Agent
A small, standalone, readable implementation of a coding agent loop in Python, written to explain the core components that make up a coding-agent harness.
Mini-Coding-Agent is a small, standalone coding agent, provided as a single Python script and CLI, built to explain the core components of a coding-agent harness rather than to be a full product. It runs a minimal local agent loop around six building blocks: collecting a stable snapshot of live repository context, a stable prompt prefix kept separate from changing turn state so repeated model calls can reuse a cache, structured tools with validated inputs, workspace-path checks and approval gates for risky actions, clipping and deduplicating long or repeated output to keep the prompt size under control, a durable transcript plus a smaller working memory so sessions can resume, and bounded delegation of scoped subtasks to helper agents.
Its model backend is Ollama, and it requires Python 3.10 or newer with Ollama installed and at least one model pulled locally; it has no other runtime dependency beyond the standard library.