Zodios is a TypeScript library that provides a fully typed and validated REST API client built on top of axios and zod, along with a companion server adapter for Express. The core library (`@zodios/core`) offers a centralized API definition format using zod schemas, auto-complete for paths and parameters on the client side, and runtime validation of both request parameters and responses. It includes a sophisticated plugin system for intercepting requests and responses with support for named plugin replacement and per-endpoint registration. The library also provides helper functions for building API definitions (`makeApi`, `apiBuilder`, `parametersBuilder`, `makeCrudApi`, `mergeApis`) and internal plugins for form-data, form-url, and text request formats. The system is designed for end-to-end type safety, allowing developers to share API definitions between client and server code. It has been well-architected with a clean separation of concerns: the client class (`Zodios`) handles HTTP communication through axios and delegates schema validation to the validation plugin, while the API definition helpers provide a builder pattern for composing endpoints. The plugin system is particularly notable for its flexible registration (global, by alias, or by method+path) and its correct reverse-order execution for response interceptors. Overall, the codebase demonstrates strong TypeScript craftsmanship, disciplined use of type inference and utility types (e.g., path parameter extraction at the type level), and a modular design that allows features like CRUD generation and API merging. While not revolutionary, it is a solid, production-quality library that effectively solves the problem of type-safe REST communication.
This is a Spring Boot starter library that deeply integrates Square's Retrofit HTTP client with Spring Boot's auto-configuration, allowing developers to declare typed HTTP APIs via Java interfaces annotated with `@RetrofitClient`. The library provides a comprehensive set of features: automatic OkHttpClient management, a pluggable interceptor chain (global, network, path-match), configurable retry with exponential backoff and jitter, circuit breaking (via Sentinel or Resilience4j), logging strategies, Micrometer metrics, custom error decoding, and connection pooling. It also supports Spring Boot Actuator for runtime introspection, GraalVM native image/AOT compilation, and reactive call adapters (RxJava 2/3, Reactor). The architecture is modular with clear SPIs for extensibility (e.g., `SourceOkHttpClientRegistrar`, `GlobalInterceptor`, `RetrofitDegrade`, `ErrorDecoder`), and the configuration is annotation-driven with global defaults overridable per-client. The codebase is well-tested with integration tests using MockWebServer, and documentation is provided in multiple languages. However, the tight coupling to specific versions of Retrofit, OkHttp, and Spring Boot introduces moderate technical drift, as some optional dependencies (e.g., Sentinel 1.8.6, Resilience4j 1.7.1) are older. The library is primarily a framework integration rather than a standalone application, but its design patterns serve as a reference for building Spring Boot auto-configuration with complex SPI chains.
This repository is a starter template for Nuxt 4 applications, created by Anthony Fu. It provides a pre-configured setup with modern tooling including Vue 3, Vite, UnoCSS for atomic CSS, Pinia for state management, PWA support via @vite-pwa/nuxt, and auto-imports for composables and components. The architecture follows Nuxt's file-based routing and SSR model, with a Nitro server layer for API endpoints (e.g., a simple pageview counter). The codebase demonstrates clean, idiomatic usage of Nuxt 4 features like `useState`, `useFetch`, layout system, and the Composition API. It is not a production application but a foundation for building Nuxt projects, with several reusable configuration modules and composables.
atmoz/sftp is a Docker image that provides a production-ready SFTP server built on OpenSSH. At runtime, the container starts sshd with a chroot jail, forcing every user into their own home directory and restricting access to SFTP only. User accounts are provisioned from three sources: command-line arguments, the SFTP_USERS environment variable, and a mounted users.conf file. The entrypoint merges these definitions on first run, generates unique SSH host keys if none are mounted, and optionally executes custom scripts from /etc/sftp.d before launching sshd or passing through arbitrary commands. The core logic lives in two well-crafted Bash scripts. files/create-sftp-user parses compact user specifications (user:pass[:e][:uid[:gid[:dir1,dir2]]]), validates them with POSIX regexes, creates system users with optional UID/GID, sets plain or encrypted passwords, aggregates SSH public keys from .ssh/keys into a deduplicated authorized_keys file, and creates writable subdirectories. files/entrypoint handles configuration merging, first-run initialization, host key generation, and command passthrough. The sshd_config enforces secure defaults: no root login, no TCP forwarding, no X11, and internal-sftp with ChrootDirectory %h. The project includes a solid integration test suite (tests/run) using shunit2 that works with both Docker and Podman, exercising real SFTP sessions, key-based login, directory creation, custom UID/GID, and legacy config paths. CI workflows rebuild images daily to pick up OpenSSH security updates and run ShellCheck plus signature verification. The codebase is small, focused, and well-maintained, with no significant technical debt or security debt. It is a strong example of a purpose-built infrastructure container with clean shell scripting and meaningful test coverage.
The `pre-commit` npm package automatically installs and manages git pre-commit hooks for Node.js projects. When installed as a devDependency, its `install.js` script locates the project's `.git` directory (including submodules and linked worktrees) and writes a shell wrapper into `.git/hooks/pre-commit`. Upon commit, that wrapper executes a crafted bash script (`hook`) that resolves the `pre-commit` package path via Node's `require.resolve` (supporting Yarn Plug'n'Play), then runs `index.js`. The `Hook` class in `index.js` reads the project's `package.json`, normalizes the `pre-commit`/`precommit` configuration into an ordered list of npm scripts, and executes them sequentially using `cross-spawn`. If any script fails with a non-zero exit code, the commit is aborted with a clear error message. The tool gracefully handles missing packages (exits 0 to avoid blocking commits), environment quirks (emacs/magit's `GIT_LITERAL_PATHSPECS`, SourceTree's path issues), and provides backup/restore of existing hooks. This is a single-purpose utility library with a clean Node.js core and a well-tested bash wrapper.
EvoSkill is a CLI tool and Python library for automatically discovering and improving skills for coding agents. It implements a self-improvement loop: a base agent is evaluated on a task, failures are analyzed by a proposer agent which suggests skill/prompt mutations, a generator creates new skills/prompts, and the evolved agent is re-evaluated. The best programs are tracked via git branches, enabling frontier management and rollback. The system supports multiple agent SDKs (Claude Code, Codex CLI, OpenCode, OpenHands, Goose, Harbor) through a unified harness abstraction, allowing the same evolution loop to work across different agent platforms. The architecture is modular: CLI commands (init, run, eval, skills, diff, logs, reset, remote) provide the user interface; an API layer (EvoSkill, EvalRunner) offers programmatic access; the loop engine (SelfImprovingLoop) orchestrates the evolution; the registry (ProgramManager) handles git-based program versioning; the harness layer abstracts SDK-specific execution with retry, timeout, and structured output parsing; agent profiles define role-specific configurations (base agent, proposer, generator, evaluator); and evaluation modules support multiple scoring methods (exact match, tolerance, LLM-as-judge, script-based, Harbor). External dependencies include various agent SDKs (claude-agent-sdk, opencode-ai, openhands-ai, etc.), LLM providers (OpenRouter, Fireworks), and infrastructure tools (Docker, Daytona for remote execution). The repository also includes example benchmarks (OfficeQA, SEAL-QA, Dabstep, LiveCodeBench) with datasets and evaluation scripts, demonstrating the system's application to real-world tasks. The codebase exhibits strong engineering practices with clean abstractions, reusable components, and comprehensive testing, though there are some areas of duplication (e.g., skill-creator scripts) and global mutable state in the SDK configuration.
This repository is a collection of Claude Code skills and Python scripts that automate daily paper recommendation and note-taking for researchers. It fetches papers from HuggingFace Daily Papers, HuggingFace Trending, and arXiv API, scores them against user-defined keywords, enriches metadata (authors, affiliations, figures, method names), and generates a tiered daily recommendation markdown file (must-read, worth reading, skip). For must-read papers, it invokes a paper-reader skill that produces detailed structured notes with figures, formulas, tables, and concept wiki entries, all saved into an Obsidian vault. It also fetches GitHub Trending repositories and writes notes. The system includes a web viewer (FastAPI + vanilla JS) for browsing the vault content, with full-text search, wikilink resolution, and a Claude chat interface. The architecture is a three-step pipeline (fetch, review, notes) orchestrated by Claude Code skills, with Python scripts handling data fetching, scoring, enrichment, and file management. The codebase demonstrates solid engineering: async HTTP with fallback chains, robust HTML parsing, heuristic text extraction, and clean modular design. It is primarily an application (a set of tools) rather than a data research project, though it involves significant data processing.
Episteme is a modern, cross-platform e-book and document reader built with Kotlin Multiplatform and Compose Multiplatform. It supports a wide array of formats including EPUB, PDF, MOBI, audiobooks (M4B), FB2, DOCX, ODT, PPTX, and comic archives (CBZ/CBR). The app offers a rich reading experience with paginated and vertical scroll modes, text-to-speech, ink and text annotations, OPDS catalog browsing, cloud sync via Google Drive and Firestore, AI-powered summarization, and extensive customization (themes, fonts, brightness). The codebase is organized into a shared module containing core business logic, models, and policies, with platform-specific implementations for Android (native Compose, Room, WorkManager), iOS (Swift), and Desktop (Compose Desktop, Swing interop). The architecture follows a clean separation of concerns with expect/actual abstractions for platform services, making it a strong example of production-ready KMP application design.
This repository contains Pound, a Nintendo Switch emulator focused on emulating Switch 1 and 2 hardware. The emulator features a modular architecture with CPU emulation handled by the Ballistic ARM recompiler for performance, GPU emulation translating Switch graphics commands to host APIs (Vulkan, OpenGL, DirectX, Metal), and SDL-based cross-platform windowing, input, and audio systems. Dear ImGui integration provides debugging and configuration interfaces. The codebase demonstrates careful separation of concerns, with distinct subsystems for memory management, CPU state, platform abstraction, and hardware emulation. Notable engineering practices include the clean interface boundary with the Ballistic recompiler, memory allocation abstraction via a host allocator interface, and platform detection utilities. While the project shows active development and solid engineering, it faces typical emulator challenges including platform-specific maintenance burdens and dependency on external shader translation tools.
ACGTI (ACG Type Indicator) is a production-grade, open-source web application that delivers a personality quiz matching users to anime characters based on MBTI dimensions. The frontend is a Vue 3 SPA with TypeScript, Vite, and vue-router, featuring 39 Likert-scale questions, 8 archetypes, and over 100 curated characters with multilingual support (zh-CN, zh-TW, en, ja). The backend runs on Cloudflare Pages Functions with D1 database for anonymous statistics collection, AI-generated insights via Workers AI, and a cron worker for periodic stats snapshots. The codebase includes extensive offline analysis scripts (Python and Node.js) for weight calibration, probability simulation, and data validation, demonstrating a hybrid character with strong data engineering practices. The architecture is clean: frontend scoring engine is decoupled from backend, i18n is well-structured, and CI/CD pipelines enforce data integrity. While the project is actively maintained and uses modern tooling, some manual synchronization points (e.g., count constants, duplicate data sources) introduce minor fragility.
DeepTide is a cross-platform terminal AI coding agent. The Rust workspace (deeptide-core, deeptide-host, deeptide-cli, deeptide-gui) implements the production agent engine: an interactive REPL session with an agent loop, tool dispatch (bash, edit, web, git, memory, MCP, LSP, vision, cron, planning), cross-provider model routing with local-first/cloud-escalation policies, cost tracking, tokens-per-second telemetry, tool usage observability, durable memory with BM25 retrieval ranking, session import/checkpoint/rewind, sensitive-file read protection, and a theme system for syntax highlighting and diffs. The CLI and GUI share a common host layer, and the npm package downloads prebuilt Rust binaries. Architecturally it is a multi-crate Rust workspace plus supporting native components: native/ds4 is a hard-forked DeepSeek V4 Flash Metal inference engine for local LLM execution, and native/dsgo is a Swift gateway that translates between Anthropic and OpenAI protocols with SSE streaming. Primary entry points are crates/deeptide-cli/src/main.rs (CLI), crates/deeptide-gui/src/main.rs (eframe/egui GUI), and bin.js (Bun launcher that delegates to @paean-ai/zero-cli). External dependencies include DeepSeek/Anthropic/OpenAI-compatible APIs, local Metal GPU inference, and GitHub Releases for binary distribution. The repo also contains 50+ self-contained HTML5 Canvas pixel-art games and Three.js showcases under samples/, which appear to be AI-generated demo content demonstrating agent output quality rather than core product code. These share heavily duplicated boilerplate (seeded RNG, localStorage save/load, bilingual i18n, screen state machines). The core Rust code is well-tested with integration tests covering routing, cost tracking, memory ranking, tool result summarization, and slash commands; live API tests are marked #[ignore]. Hybrid note: while primarily an application, it contains research-oriented elements such as memory capture benchmarks (benchmarks/auto_capture_bench.py), prompt cache layout benchmarks (benchmarks/cache_memory_bench.py), and deterministic puzzle generation algorithms with uniqueness verification.
This repository is the Microsoft 365 Agents SDK, a multi-language framework for building conversational agents that integrate with Microsoft 365 Copilot, Microsoft Teams, Azure Bot Service, and custom channels. It provides SDKs in .NET, JavaScript/TypeScript, and Python, with the actual SDK libraries hosted in separate language-specific repositories. The root repo serves as a central hub containing sample applications, formal protocol specifications, automation tooling, and AI assistant skill definitions. The sample applications demonstrate a wide range of capabilities including quickstart templates, adaptive cards, streaming responses, proactive messaging, OAuth/auto-signin flows, on-behalf-of authorization, retrieval-augmented generation (RAG) via Copilot Retrieval API, OpenTelemetry observability, Copilot Studio integration, Teams-specific features (message extensions, meetings, task modules, attachments), Slack integration, Semantic Kernel/LangChain multi-turn agents, and named-pipe transport for non-HTTP deployments. The automation pipeline in `automation/teams-sample-sync/` is a sophisticated CI/CD system for migrating Teams SDK samples into the Agents SDK format. It implements deterministic planning, versioned state contracts, integrity verification via digests, and structured validation gates. The `agent-plugins/` directory contains executable skill definitions for AI coding assistants to assist with SDK usage, migration, provisioning, and debugging. The `specs/` directory contains formal protocol specifications including the Activity Protocol, card formats, transcript schema, and bot manifest definitions.
This repository is a multi-platform decompilation of **Castlevania: Symphony of the Night**, targeting the original PlayStation (PSX), PSP, Sega Saturn, and a modern PC port. It reconstructs the game's source code from original binaries to produce byte-identical builds, while also providing a full reimplementation with a cross-platform port. The codebase is organized into per-stage, per-boss, and per-weapon overlay modules, each defining entity update tables, initialization data, and asset references. A central entity system with step-based state machines drives all game objects, and a shared `g_api` function table enables cross-overlay communication. The project includes extensive tooling for asset extraction/rebuilding (Go), automated code linting and transformation (Rust), binary diffing (Python), and duplicate function detection (Rust). While the decompilation is largely complete for the US PSX version, many PSP and Saturn functions remain as `INCLUDE_ASM` stubs, and the PC port contains several `UnsupportedEntity` placeholders. The architecture demonstrates high craftsmanship in its overlay system, entity dispatch patterns, and modular toolchain, though platform-specific `#ifdef` branching and code duplication across overlays introduce maintenance complexity.
Outpost is a production-grade, multi-tenant event delivery platform by Hookdeck. It ingests events via an HTTP API or message queue adapters (RabbitMQ, AWS SQS, GCP PubSub, Azure Service Bus), routes them to configurable destinations (webhooks, AWS S3/SQS/Kinesis, Azure Service Bus, GCP PubSub, Kafka, RabbitMQ, Hookdeck), and provides end-to-end observability including delivery logs, metrics, alerting, and a self-service user portal. The system is implemented as a modular Go monolith with pluggable storage drivers (PostgreSQL, ClickHouse, Redis) and a clean provider pattern for destination types. It includes generated SDKs for Go, Python, and TypeScript (via Speakeasy), plus a React-based embedded portal. The codebase demonstrates strong engineering practices: abstract backends with conformance test suites, cursor-based pagination, configurable backoff strategies, distributed locking, and comprehensive integration tests covering all destination types. Notable reusable components include a generic LRU cache with TTL, a JSON filter matcher, a cursor pagination library, and an ID generation utility.
This repository contains steamguard-cli, a command-line Steam Mobile Authenticator written in Rust. It allows users to generate two-factor authentication codes, manage Steam mobile authenticator accounts, confirm or deny trade and login confirmations, and perform account setup, import, encryption/decryption, and QR code login from the terminal. The codebase is organized as a workspace with two crates: a library `steamguard` that implements core Steam API interactions (authentication, two-factor, phone linking, confirmation parsing) and a CLI binary that provides a comprehensive set of subcommands using clap. The library features an extensible encryption system supporting both modern Argon2id-AES and legacy SDA-compatible encryption, a transport abstraction for HTTP clients, and protobuf-based API definitions reverse-engineered from Steam's services. The project demonstrates solid Rust practices including use of secrecy for secrets, zeroize for memory clearing, and a clean trait-based architecture.
This is a security-first Android PDF viewer built by GrapheneOS. It uses pdf.js inside a heavily sandboxed WebView to render PDFs without requiring any app permissions. The app reads PDFs exclusively through Android content providers, never granting the WebView direct file or network access. The architecture is a single-activity Jetpack Compose app with a ViewModel managing state, a JavaScript bridge (PdfJsChannel) for communication with pdf.js, and a custom WebViewClient that intercepts all resource requests to serve only bundled assets. The codebase demonstrates strong security practices: strict Content-Security-Policy, disabled network and file access, no cookies, and no dynamic code evaluation. It includes comprehensive Android instrumentation tests and JavaScript unit tests. The repository is actively maintained with daily Dependabot updates and CI pipelines. While the app is functionally focused and not architecturally complex, the security hardening and clean separation between native and web layers are noteworthy.
bees (Best-Effort Extent-Same) is a specialized C++ application designed for userspace deduplication on the Btrfs filesystem. It operates as both a CLI tool and a systemd daemon (beesd), scanning filesystem extents to identify duplicate blocks and reclaim space using Btrfs-specific ioctls like BTRFS_IOC_FILE_EXTENT_SAME. The architecture is modular, separating low-level kernel interaction (lib/fs.cc), hash table management (src/bees-hash.cc), and task scheduling (include/crucible/task.h). It prioritizes memory efficiency and incremental scanning to minimize downtime on large filesystems. The system relies heavily on Linux kernel internals, specifically Btrfs tree structures and extent metadata. It implements a custom memory-mapped hash table (BeesHashTable) to persist fingerprint data across restarts, allowing incremental deduplication without rescanning the entire filesystem. Concurrency is managed via a custom TaskMaster that adjusts worker thread counts based on system load, preventing I/O saturation during deduplication operations. While the code demonstrates high craftsmanship in system programming, it carries significant technical debt risks due to its tight coupling with volatile kernel APIs. The project maintains a robust test suite for its utility libraries but relies on integration testing for the core deduplication logic. It is a mature tool for Btrfs administrators but requires careful maintenance as kernel internals evolve.
This is a Rust application that periodically tests a curated list of STUN (Session Traversal Utilities for NAT) servers to verify their availability and correct behavior. The system sends STUN Binding Requests over UDP and TCP to each candidate server, compares the returned mapped address against a reference server's result to detect NAT-induced mapping changes, and optionally checks for the OTHER-ADDRESS attribute to identify servers that support NAT behavior testing (RFC 5780). Healthy servers are written to several text files (hostnames, IPv4 addresses, IPv6 addresses) with and without the NAT-testing capability filter. The output lists are shuffled hourly to distribute load across servers. The architecture is a single binary built with Tokio async runtime, modularized into server loading, STUN query logic, output formatting, and a GeoIP caching layer. It is not a library; it is a self-contained service intended to run as a GitHub Actions cron job. The code is clean, idiomatic Rust with proper error handling and logging.
This repository provides a Python library (`espargos`) and a suite of Qt6/QML demo applications for controlling ESPARGOS digital phased array hardware (ESP32-based). The system captures WiFi channel state information (CSI) from 802.11g/n/ax packets, raw I/Q samples, and performs radar-like FTM timestamp measurements. It supports calibration of phase and timing offsets across multiple boards, enabling phase-coherent combined arrays. Demos include real-time spatial spectrum (MUSIC/MVDR), azimuth-delay, polarization analysis, IQ signal analysis, transmit beamforming, and more. The architecture is modular with capability registration, sensor clustering, ring buffers, and a calibration object model. The library is well-structured but tightly coupled to the specific hardware, and demos depend on the common framework. It is actively maintained with no significant technical drift.
FileRise is a self-hosted web file manager and storage hub with WebDAV, sharing, and per-folder ACLs. The backend is PHP 8.3+ with Apache, using flat-file storage (no database) and a REST API. The frontend is a modular vanilla JS SPA with i18n, dark mode, and drag-and-drop. It supports multi-source storage (local, S3, SFTP, WebDAV, etc.), encryption at rest (libsodium), TOTP 2FA, OIDC SSO, and background workers for transfers and ZIP creation. A commercial Pro add-on provides AI chat, client portals, automation, and advanced storage sources. The codebase is actively maintained with frequent releases and a comprehensive security regression test suite. While the architecture is clean with well-defined interfaces (storage adapter, ACL, encryption), there is some legacy debt from a namespace migration and monolithic domain models.