Lewis Cheng

A Claude Code and Codex skill that interrupts agentic coding with short exercises built on retrieval practice, spacing and metacognition, plus an optional repo orientation generator.

Three editions, five subsystems, and performance tables that show what the protection costs: pure forwarding at 67,849 QPS drops to 10,574 with every module on.

One INI file per aircraft, an ADS-B data source you choose, and output to Discord, Telegram or Mastodon. The repo is honest about being unstructured and about which feed works better.

A Flutter library that bundles reactive state, routing and dependency injection behind a single static Get class, and is published on pub.dev as the get package.

A weekly automated pipeline scrapes and ranks GitHub accounts and repos, then renders EJS templates into a GitHub Pages site. The pipeline scripts in package.json are the whole project.

A pure Python library that parses your code, infers types, and answers the questions an editor plugin needs to ask, plus a note about the Rust successor its author now points people to.

The headless branch of the Aimeos shop system arrives not as a library but as a full Laravel skeleton with its own routes, config, migrations and Vite setup, and the repository is thin on installation detail.

The Android team's camera sample app rebuilt as a Compose catalog where each feature is a Gradle library module, with the camera plumbing factored out into shared modules.

A self-hosted FastAPI service that keeps track of 3D printer filament spools, updates remaining weight as jobs run, and ships its own web client.

Anime2Sketch is a PyTorch inference repository that turns anime art, manga panels and illustrations into clean line drawings using pretrained weights from an AODA research paper. It is small, MIT licensed, and inference only, which is both its main convenience and its main constraint.

The containers/kubernetes-mcp-server project puts a Kubernetes API client behind the Model Context Protocol so an LLM tool can list pods, read logs, apply resources and run Helm releases without shelling out to kubectl. The design is a single Go binary, and the trade-off is that it ships no built-in authorization of its own.

Blueprint turns a single human-readable YAML definition into models, migrations, controllers, routes, factories, form requests, jobs, events and tests. It is a scaffolding tool for Laravel 11 or higher, and its grammar is the contract you have to maintain.

Lieflat Charts is a SKILL.md-format data visualization skill for AI agents, built around three visual systems (Lupi, Glance, Basics) plus interactive graph templates and twelve report layouts. The judgement: it is a styling and layout layer, not a charting engine, and its licence status is unresolved.

CAMEL is a Python framework for studying how AI agents communicate, cooperate, and scale, with support for up to one million agents. This review looks at its design principles, practical setup, and where it fits in the research landscape.

Taze is a TypeScript CLI that checks and updates dependencies across a repo or monorepo. It is safe by default, but its defaults hide major updates and locked versions unless you ask for them.

LibreDesk bundles live chat, email, a help center and an AI assistant into a single Go binary backed by Postgres and Redis. It is aimed at teams that want Zendesk-style ticketing without a hosted vendor, and the AGPL-3.0 licence is the main thing to weigh before adopting it.

Hydra Download Manager splits a file across connections and mirrors and rebalances work as peers slow down. The engine is credible, the install scripts are the easy path, and the GPL-3.0 binaries are not the permissive part of the workspace.

A small MIT licensed CLI from dylang that audits a package.json against the registry and your own source, then offers to fix what it finds through an interactive picker.

This is a PyTorch implementation of two published segmentation architectures, with six pretrained models and four deployment paths. What makes it worth reading is the fourth note under the results tables, where the author states that repeated training runs of the same configuration vary by about two points of the headline metric, which means most of the differences in his own table are inside the noise.

CellDock is a native macOS menu bar app for the QDC507 cellular module. It brings SMS, voice calls, eSIM profile management and a per-module SOCKS5 proxy to the desktop, at the cost of hardware compatibility and a licence file that needs reading.

SUMO-RL wraps the SUMO traffic simulator in Gymnasium and PettingZoo APIs so you can train single-agent or multi-agent signal controllers. The wrapper is thin, the reward and observation functions are replaceable, and the real cost is SUMO itself.

Nightfox ships seven palettes, a compile step for startup speed and a daltonization mode for colorblind users. Here is what the README and repository actually document, and where the theme stops being the right choice.

The project ships two things: a FastAPI proxy that speaks /v1/chat/completions and /v1/messages, and a reusable Python library for API key rotation and failover. It is aimed at people who already hold keys for several providers and want one base URL in front of all of them.

ComfyScript turns ComfyUI graphs into Python you can read, run and generate. It is a good fit for people who want loops and library calls around diffusion nodes, and a poor fit for anyone who only wants a GUI.

BaleVPN relays internet access from one Bale account to another over the messenger's LiveKit voice-call infrastructure. The README is unusually honest about what that costs you in privacy.

A TypeScript CLI that turns the clock times of your commits into a workload report, a trend line across months and a local web page you can read without uploading anything.

Boost.Beast is a header-only C++11 library that supplies HTTP/1 and WebSocket protocol types and algorithms on top of Boost.Asio. It suits engineers who already write asynchronous Asio code and want protocol building blocks rather than an opinionated networking framework.

Playwright for .NET is Microsoft's official C# port of Playwright, automating Chromium, Firefox and WebKit behind one API. It fits .NET teams that want browser tests in the same language as their application, and it costs anyone who needs a frozen, versioned toolchain.

Defold is a free, multi-platform game engine written in C++ with Lua scripting. This review covers what the repository actually contains, how to build it from source, and where the editor-first workflow stops being the right choice.

Scattertext turns a two-category text comparison into an interactive HTML scatter plot where each point is a term, and its coordinates are the term's relative frequency in each category. The library is aimed at people who already have a corpus in a pandas DataFrame and want to see, not just rank, what distinguishes one side from the other.

Forem is the AGPL-3.0 Ruby on Rails codebase that runs dev.to, packaged for self-hosting with Kamal 2 and a Docker development stack. It suits teams who want a full community product, not a weekend project.

graphql-anycable is a drop-in adapter that lets the graphql gem run subscriptions over AnyCable, storing state in Redis and re-executing queries at event time. It works, but you must plan Redis cleanup and know where the broadcast option changes behavior.

Nexus Agent is an Apache-2.0, Java 25 and Spring AI Alibaba agent platform that puts deterministic orchestration in front of both RAG and ReAct execution. The interesting part is not the feature list but the ordering: ambiguity detection first, knowledge Q&A second, open-ended agent last.

Harbor is an MIT-licensed desktop media player for the open Stremio addon protocol, built on Tauri with a Rust core. It hosts, indexes and bundles no media: you bring your own addons and sources, and the maintainers suggest building it yourself.

git-ai is a Rust Git extension that records which agent, model and prompt produced each line of AI code. It installs as a single shell script, stores attribution outside the repository, and adds a git ai blame and git ai stats command.

ESP32 Bit Pirate is MIT-licensed ESP32 firmware that puts Bus Pirate-style protocol work behind a serial or browser CLI. It is broad by design, and the breadth is also where its limits start.

OwnMem puts agent memory in Markdown inside your repository and delivers it through deterministic local recall with evidence gates. It is a governance tool for teams that already treat memory as a liability, not a feature.

Hunk is a review-first terminal diff viewer for agent-authored changesets, built on OpenTUI and Pierre diffs. It offers a multi-file sidebar, inline AI annotations, and Git, Jujutsu, and Sapling support, but its focus on interactive review means it is not a drop-in replacement for plain-text diff tools.

Magic Resume is a free, browser-based resume editor that stores data locally and can call AI providers for writing help. The Apache 2.0 licence carries a non-commercial restriction, which decides most adoption questions before the code does.

One Composer package that teaches a Laravel application to answer a single route with either a Blade view or a JSON payload, and to switch between them without either side noticing.

Hidden Bar is a small Swift menu bar utility from Dwarves Foundation that tucks icons behind an arrow instead of hiding them at the system level. It is free, MIT licensed, and requires macOS 13 Ventura or newer.

chnroutes2 is a data project that aggregates China-related BGP routes from several providers into a single, frequently updated route set. It is aimed at network operators and developers who need current CN route information without running their own collectors.

Anthropic's Claude Cookbooks repo collects Jupyter notebooks and guides for building with Claude, from tool use to RAG to cost tuning. The material is strong for copy-paste starters, but it is not a maintained library and carries no versioning or support guarantees.

A Rust TUI that signs its own AWS requests with SigV4 and talks to the Query and JSON protocols directly, avoiding the SDK entirely. Broad service coverage and a seven-step credential chain, still shipping under a release candidate tag.

A record attempt on Keller Jordan's modded-nanogpt speedrun that reports 74.19 to 74.6 seconds against record #86's 75.96 seconds. The gain comes from a tail EMA of the readout weights, which lets the run finish in 1288 steps instead of 1300.

Makina Maps is a full-stack Docker Compose setup that builds, serves, and updates OpenStreetMap-derived vector and raster tiles on request, using OpenMapTiles and TileServer-GL. It trades pre-generated MBTiles archives for a live database and an NGINX cache, which makes updates fast but adds operational complexity.

SecureDrop is a self-hosted whistleblower submission system that media organizations run on their own infrastructure to accept documents from anonymous sources. This review covers what the repository actually ships, how to bring up a local development instance, and where the project's focus has moved.

LyleMi/Learn-Web-Hacking is a Chinese-language study notebook covering web protocols, common vulnerabilities, intranet penetration and defence, published as a Sphinx site under CC0-1.0. It is a reference to read and fork, not a guided training platform.

This repository is one developer's versioned pi-setup: extensions, themes, skills and config for the Pi coding agent, plus sync tooling to back the live ~/.pi/agent up to GitHub and restore it on any machine. It wraps an existing Pi install rather than replacing it, and ships no license.

Fara1.5 is Microsoft's family of three native computer use agent models (4B, 9B, 27B) built on Qwen3.5, shipped with an MIT-licensed Python repo, a fara-cli entry point and a WebTailBench harness. The repo is a client and an evaluation rig, not the model, and every run needs an endpoint you host or rent.

httptap wraps a command, intercepts its HTTP and HTTPS traffic in a private network namespace, and prints each request and response. It is a Linux-only Go binary for debugging what a CLI tool or script actually sends.

AmpliGraph is an Apache-2.0 Python library that turns a knowledge graph of triples into vector embeddings and scores candidate links with model-specific scoring functions. Its 2.x line moved to TensorFlow 2 and a Keras-style API, dropped several older models, and changed the input/output pipeline.

chromedp talks to Chrome through the DevTools Protocol using pure Go, so browser automation stays inside your existing Go build and test pipeline. The trade-off is that you manage the browser process yourself, and the documentation lives in the Go reference rather than the README.

A Ruby gem that parses your code for I18n key usage so that missing translations fail a health check instead of a production request, and dead keys can be pruned on purpose.

ai-usagebar tracks Claude, Codex, Copilot, GLM, OpenRouter and other AI coding plan quotas from a Waybar module, an Omarchy panel, or a terminal TUI. It is a Rust port of claudebar, and it is honest about what it cannot fetch.

slot-text is an MIT-licensed TypeScript library that animates short UI labels by rolling characters or words in and out. It ships vanilla, React, Vue, Solid and Svelte entry points, and it has one interesting trick: a flash() call that reverts itself.

Arm's reference secure world firmware for A-profile processors, mirrored read-only, implementing five Arm interface standards across a five-stage boot, with the project explicitly telling users to penetration-test their own derivative and a version scheme where the patch component is reserved for long-term support releases.

LMMS is a cross-platform DAW written in C++ with a piano roll, beat sequencer, song editor and mixer. It installs from lmms.io or builds with CMake, and it is the right tool for MIDI and SoundFont work, not for a VST3 plugin collection.

JD Open Source ships an 8B vision-language interaction model with weights, data, a training recipe and a streaming system in one Apache-2.0 repository. The install path is a three-script shell workflow, and the interesting design choice is proactive speech rather than turn-based question answering.

gh-dash is a Go terminal UI that renders your GitHub pull requests and issues inside a gh CLI extension. It suits engineers who live in the shell, and it is not a Git client.