The author announced llm-coding-agent 0.1a0, an alpha Python library that implements a Claude Code–style coding agent on top of their evolving LLM library / agent framework. The package has been published as a prerelease on PyPI.
What it is and how it was developed
The project started as a new Python library using the python-lib-template-repository GitHub template. During development the author ran two prompts (a Claude Code web transcript is available):
- Write a spec.md for this project — it will depend on the latest “llm” alpha from PyPI and implement a Claude code style coding agent complete with tools for reading and editing files and executing commands.
- Commit the spec, then build it using red/green TDD in a series of sensible commits (each with passing tests and updated docs) — occasionally manually test it using the OpenAI API key in your environment.
The resulting README and the sequence of commits are provided by the author. After publishing a "slop-alpha" the package can be run with uvx as follows:
uvx --prerelease=allow --with llm-coding-agent llm code
The author describes it as a pretty good first attempt.
User interface and API
The README lists example commands such as:
- llm code --yolo
- llm code --allow "pytest*" --allow "git diff*"
It also exposes a Python API. One showcased class is:
CodingAgent(model="gpt-5.5", root="/path", approve=True).run("Fix the failing test in tests/test_parser.py")
The author notes they did not explicitly request this class but are pleased it was implemented.
Implemented tools
The package implements a set of tools the agent can use; the author listed them from the uvx ... llm tools output. Key tools and their behavior:
-
CodingTools_edit_file(path: str, old_string: str, new_string: str, replace_all: bool = False) -> str
- Replace an exact string in a file. old_string must match the file contents exactly (including whitespace) and must identify a unique location unless replace_all is true. Returns a diff of the change for verification.
-
CodingTools_execute_command(command: str, timeout: int = 120) -> str
- Run a shell command in the session root directory. Returns combined stdout and stderr followed by an Exit code line. timeout is in seconds (maximum 600); on timeout the whole process tree is killed.
-
CodingTools_list_files(pattern: str = '**/*', path: str = '.') -> str
- List files matching a glob pattern, newest first. Skips hidden directories, node_modules, pycache and (in a git repository) anything covered by .gitignore. Returns at most 200 paths relative to the searched directory.
-
CodingTools_read_file(path: str, offset: int = 0, limit: int = 2000) -> str
- Read a text file, returning numbered lines like cat -n. Paths are relative to the session root. Use offset (0-based first line) and limit (max lines) to page through large files.
-
CodingTools_search_files(pattern: str, path: str = '.', glob: str = None, max_results: int = 100) -> str
- Search file contents for a regular expression. Returns matches as path:line_number:line, capped at max_results. Use glob (e.g. "*.py") to restrict which files are searched.
-
CodingTools_write_file(path: str, content: str) -> str
- Create or overwrite a file with the given content. Parent directories are created as needed. Prefer edit_file for modifying existing files.
Examples and behavior
The author tested the tool by running llm code --yolo and then prompting it to create a /tmp/demo folder and, within it, a simple SwiftUI CLI app that prints the time in ASCII art.
According to the transcript, GPT-5.5 observed that "SwiftUI isn't suitable for a true CLI" and then produced an application which, when run with swift run AsciiTime, outputs an ASCII-art time display (excerpt in the original material):
█ █████ ████ █ █ ███
██ █ █ █ ██ █ ██ █ █
█ ████ ███ █ █ █
█ █ █ █ █ █ █ █
███ ████ ████ ███ ███ █████
The author shared the transcript and other project artifacts.
Why this matters
The release demonstrates a straightforward way to turn an LLM library into an agent framework and to build an automated coding agent that can read, search, edit files and execute commands. The provided toolset covers common tasks and can serve as a practical starting point for further development and integrations.
Access
The package is available on PyPI as a prerelease. The README, commit history and transcript are available among the author’s shared materials.
(Tags: projects, ai, generative-ai, llm, llm-tool-use, coding-agents, claude-code, claude-mythos-fable)
