Feature Request: Declarative dependency management with pyproject.toml

[sontianye]

Feature Request: Declarative dependency management with pyproject.toml

First off — thanks for building GenericAgent. The minimalist architecture and self-evolving skill system are really elegant. I've been using it and enjoying the experience.

The issue: Currently there's no pyproject.toml or requirements.txt, so new users need to manually figure out which pip install commands to run. The README lists different pip install lines scattered across sections (streamlit, pywebview, pycryptodome, lark-oapi, etc.), and it's easy to miss one or install unnecessary packages.

For example, a user who just wants the Telegram bot frontend still needs to know to install requests and streamlit as a baseline, which isn't immediately obvious from the Bot section alone. The self-evolving code_run mechanism mitigates this nicely at runtime, but having a declarative manifest upfront would give a smoother first-run experience.

Proposal: Add a pyproject.toml with optional dependency groups so users can install exactly what they need in one command:

uv pip install -e .              # core (requests, streamlit, pywebview, beautifulsoup4)
uv pip install -e ".[tg]"        # + telegram
uv pip install -e ".[browser]"   # + browser automation
uv pip install -e ".[all]"       # everything

This keeps the project's zero-config philosophy — users still just clone and run. The manifest simply makes the implicit dependency graph explicit. The agent's self-install capability via code_run remains untouched.

Optional groups I had in mind:

Group	Packages
core (default)	requests, streamlit, pywebview, beautifulsoup4
qt	PySide6
browser	bottle, simple-websocket-server
tg	python-telegram-bot
qq	qq-botpy
wechat	qrcode, pycryptodome
feishu	lark-oapi
wecom	wecom-aibot-sdk
dingtalk	dingtalk-stream
ocr	rapidocr-onnxruntime, pillow
vision	ultralytics, pillow, numpy
langfuse	langfuse

Happy to submit a PR if this direction aligns. Thoughts?