mini-coding-agent

Why we included this project

Anyone who has wondered what a coding agent actually does under the hood will find this a good place to look. It is a deliberately small, single-file Python harness that runs a real agent loop against a local Ollama model, built around six practical pieces: gathering repo context up front, keeping a stable prompt prefix so repeated model calls can reuse cached parts, exposing structured tools with path validation and approval gates, trimming and compressing context, persisting transcripts and memory for resumable sessions, and delegating scoped subtasks to bounded subagents. Since it has no runtime dependency beyond the standard library, you can read the whole implementation in one sitting and run it against a workspace to watch each piece work. For engineers who want to understand agent internals before building their own, it is a solid teaching reference; for anyone who wants a local, privacy-friendly coding assistant, it is a minimal starting point that inspects and edits a repository on your machine.