Skip to content
/ baton Public

Python wrapper for codex/claude CLI, enabling arbitrarily complex agent automation/workflows through programmable prompts.

License

Notifications You must be signed in to change notification settings

banbox/baton

Repository files navigation

AIBaton

中文文档

Warning

This project has migrated! To avoid a PyPI package name conflict, this project has been officially renamed to aibaton. This repository (baton) is no longer maintained. Please migrate to the new repository as soon as possible.

New Repository: github.com/banbox/aibaton

Headless Agent Automation — Python wrapper for codex/claude CLI, enabling arbitrarily complex agent automation/workflows through programmable prompts.

Features

  • Prompt-Oriented Programming — Natural language prompts as Python code, seamlessly interact with agents
  • Skill Support — Use skills to convert intent into aibaton code, making complex tasks more controllable
  • Loop Detection — Built-in iterative optimization (loop_max)
  • Response Parsing — Parse agent results with semantic options using parse(options)

Use Cases

  • Repetitive Tasks — Batch tasks that reuse the same prompt workflow for agent interaction
  • Context Explosion Tasks — Tasks with extensive context where agents miss too many details, requiring finer-grained control
  • Automated Workflows — Build workflows directly on agents, implement arbitrarily complex logic with simple Python syntax

Not Suitable For

  • Non-Repetitive Tasks — No need to script for repeated use
  • Complex Tasks Requiring Step-by-Step Confirmation — Difficult to pre-define all scenarios via script, requires human confirmation at each agent stage

Example

Install the aibaton SKILL using codex, then enter the following prompt:

$aibaton Find Go code files over 300 lines in the project and optimize them, extract redundant code into sub-functions following DRY principles, and update references in other files.

Output executable script:

from aibaton import run, set_default
import os

set_default(provider="codex", dangerous_permissions=True)

prompt = """Review the current file code and optimize, extract redundant code into sub-functions following DRY principles, and update references in other files."""

def get_go_files(root_dir: str, min_lines: int = 0) -> list[str]:
    go_files = []
    for dirpath, _, filenames in os.walk(root_dir):
        for f in filenames:
            if f.endswith('.go'):
                fpath = os.path.join(dirpath, f)
                if min_lines <= 0 or sum(1 for _ in open(fpath)) > min_lines:
                    go_files.append(fpath)
    return go_files

if __name__ == '__main__':
    par_dir = '/app/banbot'
    for fpath in get_go_files(par_dir, min_lines=300):
        run(f"File path: {fpath}\n\n{prompt}", cwd=par_dir, add_dirs=[par_dir], stream=True)

Run the script as prompted.

More Complex Task Example

A complex workflow prompt

After codex uses aibaton skill, output executable script: work_flow.py

Installation

pip install aibaton
# or
uv pip install aibaton

License

Apache 2.0

About

Python wrapper for codex/claude CLI, enabling arbitrarily complex agent automation/workflows through programmable prompts.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages