Protobuf-py: Protobuf for Python, without compromises
Python Hub
Interesting articles and projects for Python developers
You shouldn't trust Trusted Publishing
The post explains that PyPI Trusted Publishing is an authentication mechanism for machine-to-machine trust between CI/CD workflows and package registries, not a signal that a package is safe or high quality. It shows how Trusted Publishing reduces long-lived credential risk, while warning that users should not treat it like a “green checkmark” because malicious or low-quality pac…
https://blog.yossarian.net/2026/07/07/You-shouldnt-trust-trusted-publishing
microsoft / SkillOpt
SkillOpt is a text-space optimizer that trains reusable natural-language skills for frozen LLM agents through trajectory-driven edits, validation-gated updates, and deployable best_skill.md artifacts.
antigravity-sdk-python
A Python library for building AI agents that leverage the full power of Google Antigravity.
https://github.com/google-antigravity/antigravity-sdk-python
PEP 814: Add frozendict built-in type
The post explains the path from rejected PEP 416 and stalled PEP 603 to accepted PEP 814, which adds a built-in frozendict type to Python 3.15. It covers how frozendict works, why it is hashable only when values are hashable, where the standard library now supports it, and what C/Python code may need to update.
https://vstinner.github.io/pep-814-add-frozendict-builtin-type.html
fortress
Stealth Chromium engine that stops scrapers and browser agents from getting blocked, with one line of code change.
Full Python GUI apps in the browser – no JavaScript, no server
What’s the simplest way to distribute a Python app to normal users?
Automate Excel with Python: From manual grind to one-click workflow
Sandboxing an AI Agent
The post explains why autonomous AI agents should run in isolated sandboxes instead of directly on a developer’s laptop, especially as agents gain auto-approval, long-running tasks, file access, and network access. It compares sandbox architectures, from tool-backend setups to agents living fully inside the sandbox, and explains the tradeoffs across security, secrets, streaming, performa...
Miles: A PyTorch-Native Stack for Large-Scale LLM RL Post-Training
Miles is an open source PyTorch-native framework for large-scale LLM reinforcement-learning post-training, combining SGLang for rollouts, Megatron-LM for training, Ray for orchestration, and PyTorch for extensibility. The post explains how Miles handles the hard systems problems behind frontier RL, including async rollout/training, fast…
https://pytorch.org/blog/miles-a-pytorch-native-stack-for-large-scale-llm-rl-post-training/
Pure-Python symbolic regression that rediscovered Kepler's law from 8 data point
Hamiltonian Neural Networks from a Differential Geometry Perspective
The post explains Hamiltonian Neural Networks through differential geometry, showing why a normal MLP can fit motion data but still invent or lose energy over long rollouts. It shows that by learning a scalar Hamiltonian and deriving motion through the symplectic gradient, HNNs make energy conservation structurally unavoidable instead of hoping the network lear…
https://abscondita.com/blog/symplectic-sledgehammer-for-a-spring
ProtoMotions3
A GPU-Accelerated Framework for Simulated Humanoids.
Write a coding agent from first principles: better tools
This tutorial builds on the coding agent you implemented in the tutorial “Write a coding agent from first principles”. In this tutorial, you'll take your agent and improve its capabilities by implementing the text edit and bash command tools that Anthropic provides.
https://mathspp.com/blog/write-a-coding-agent-from-first-principles-better-tools
Runloom – Go-style coroutines for Python free-threaded
Otary – Image and Geometry Python Library Now Has Tutorials
bradautomates / claude-video
Give Claude the ability to watch any video. /watch downloads, extracts frames, transcribes, hands it all to Claude.