llm-coding-agent 0.1a0 released as an experimental coding agent built on the LLM framework

Simon Willison released llm-coding-agent 0.1a0, a new Python library that implements a simple coding agent atop his LLM framework. The project was bootstrapped using Willison’s python-lib-template-repository and guided by two prompts that specified a Claude Code–style agent with file and command tools.

The agent exposes six tools: file reading and writing, exact-string editing, file listing, regex search, and shell command execution with a configurable timeout. These tools are surfaced through the LLM CLI via uvx and can be invoked with commands like llm code --yolo or llm code --allow "pytest*" --allow "git diff*".

A Python API is also provided, centered on a CodingAgent class that accepts a model name, a root directory, and an approval flag, and runs tasks such as fixing failing tests. The README documents usage patterns and examples.

Willison demonstrated the agent by asking it to create a SwiftUI CLI app for ASCII time display in a temporary directory. The agent reasoned about SwiftUI’s limitations for CLI apps and produced a working Swift package that outputs an ASCII clock when run. The full transcript and outputs are available in a gist.

The release includes a project specification (spec.md), a generated README, and a commit history that follows red/green TDD, with each commit including passing tests and updated documentation. The package is published as a pre-alpha on PyPI and can be installed with uvx --prerelease=allow --with llm-coding-agent llm code.