claude-mem is a persistent memory system for AI coding assistants, primarily targeting Claude Code but also supporting Codex, Cursor, Windsurf, OpenClaw, and other platforms. It captures observations, summaries, and user prompts from coding sessions via lifecycle hooks (SessionStart, PostToolUse, etc.), stores them in SQLite (local) or Postgres (server), and injects relevant context into future sessions to maintain continuity across agent interactions. The system operates as a background worker daemon (Bun) or a multi-user server (Express + BullMQ + Valkey) with API key authentication, team/project scoping, and rate limiting. Architecturally, claude-mem is a hybrid of a CLI plugin and a standalone server. The CLI component integrates into IDEs via hook scripts and MCP servers, while the server component provides REST APIs for memory management, search, and cloud sync. It features a dual-storage backend (SQLite for local, Postgres for server) with vector search via ChromaDB and full-text search via SQLite FTS5. The system includes a React-based viewer for browsing memory feeds, telemetry (opt-out PostHog), and a Cloudflare Workers-based sync hub for cross-device synchronization. The codebase demonstrates strong engineering practices with well-abstracted components (search strategies, provider abstraction, atomic file operations, telemetry scrubbing) but carries some technical debt from its dual-runtime evolution (legacy worker vs. server-beta) and extensive Windows-specific workarounds. It is actively maintained with frequent releases and comprehensive test coverage.
CowAgent is a production-grade multi-agent AI assistant platform that routes messages from 12+ IM channels (WeChat, Feishu, Telegram, DingTalk, Slack, Discord, QQ, WeCom, web console, desktop Electron app) through a modular agent core. The core implements planning, tool use (bash, file editing, browser, web search, MCP, sub-agent delegation, scheduler, memory, vision), three-tier memory with vector embeddings and markdown-aware chunking, knowledge base management, skill system with SKILL.md manifests and a skill hub, self-evolution (idle-conversation review loop), and multi-agent team orchestration with peer transport. The architecture is a modular monolith: channel adapters normalize messages into a unified ChatMessage, the Bridge dispatches to per-agent/session model resolution, the Agent protocol (Task/AgentResult/AgentAction) drives the execution loop with streaming, cancellation, and steering, and all state persists to SQLite with incremental migration. Key entry points are app.py (startup orchestration), cli/cli.py (daemon lifecycle), channel/web/web_channel.py (HTTP API and SSE streaming), and the desktop Electron client (React/TypeScript with Zustand stores). External dependencies include LLM APIs (OpenAI, Claude, Gemini, DeepSeek, etc.), IM platform webhooks/SDKs, optional voice TTS/ASR providers, and MCP servers. The codebase demonstrates strong engineering practices with extensive regression tests (200+ test files), atomic file writes, thread-safe context propagation via ContextVar, and security guards for credential paths and SSRF.
Onlook is a browser-based visual editor for building and modifying Next.js + Tailwind CSS websites. It provides a Figma-like canvas where users can directly manipulate React components, with changes written back to source code via AST-based transformation. The application integrates an AI chat assistant that can generate code, edit files, and run terminal commands through a set of well-defined tools (e.g., search-replace, file edit, bash). The monorepo architecture separates concerns into packages for AI agents, code parsing, font management, UI components, database access (Drizzle + Supabase), git operations, and sandbox management (CodeSandbox). It also includes a complete billing system (Stripe), GitHub integration for project import, custom domain management, and a publish/deployment pipeline. The codebase demonstrates strong engineering practices with type-safe tRPC APIs, MobX state management in the editor, extensive use of Radix UI primitives, and a consistent monorepo toolchain (Bun workspaces, ESLint, Prettier). While the editor engine is somewhat monolithic, individual packages like the AST parser, font helpers, and utility library are well-isolated and testable.
This repository is a curated collection of tips, scripts, and a Claude Code plugin (dx) that enhances developer workflows with Claude Code. It is primarily a knowledge base and tooling repository, not a standalone application. The core deliverable is the dx plugin, which bundles skills for GitHub Actions debugging, conversation half/quarter-cloning, context handoffs, Reddit/HN research, private GitHub search, and version checks. The repository also includes a custom status line script (context-bar.sh), a setup script for recommended configurations, and extensive markdown content documenting 49 tips for using Claude Code effectively. The architecture is a mix of bash scripts, skill definitions (markdown with YAML frontmatter), and a plugin manifest. The standout component is the half-clone-conversation.sh script, a 400+ line bash utility that clones Claude Code conversations with careful edge-case handling, cross-platform support (macOS, Linux, Windows Git Bash), and a comprehensive 19-test suite. The repository demonstrates high craftsmanship in its bash scripting, thorough testing, and detailed documentation of workarounds for platform limitations (e.g., Reddit's blocking, GitHub's private repo search).
This monorepo (Nx) implements a complete ecosystem for managing and distributing 'skills'—curated, structured instructions for AI coding agents. The codebase reveals a production CLI (`agent-skills`) built with Ink/React for interactive terminal UI, an MCP server (`agent-skills-mcp`) for agent-in-context skill lookup via FastMCP, a Next.js marketplace website for browsing, and a skills catalog with over 150 skill definitions across architecture, cloud, development, quality, security, and other domains. The core library (`@tech-leads-club/core`) uses a clean hexagonal architecture with port/adapter separation for file system, HTTP, environment, logging, and shell operations, enabling testability and future portability. The CLI supports interactive and non-interactive install/remove/update workflows with atomic lockfile writes, append-only audit logging, and content-hash integrity verification. The MCP server implements a progressive disclosure pattern (search → read → fetch/prepare) to minimize token usage, with registry caching and ETag revalidation. The marketplace is highly SEO-optimized with JSON-LD structured data, breadcrumbs, and static generation. Property-based tests (fast-check) for UI components and cache boundaries demonstrate strong engineering discipline. The skills catalog itself is the core product—each skill is a markdown file with YAML frontmatter, optional scripts/references, and trigger phrases, all validated by a custom validation tool and security-scanned in CI. While the architecture is solid and the codebase is actively maintained with frequent releases, some areas show drift risk: duplicate validation scripts across skills, hardcoded benchmark data in the marketplace, and tight coupling to external CDN availability for registry downloads. The cloud-deploy skill references an enormous set of Cloudflare product docs that will inevitably become outdated. Overall, this is a well-crafted, modern TypeScript monorepo that effectively solves the problem of giving AI agents reusable, safe, and discoverable instructions.
This repository provides a two-skill workflow system for academic paper creation designed to work with Claude Code. The 'academic-paper-strategist' skill handles research planning through platform analysis, literature search, gap identification, and outline optimization with quality gates. The 'academic-paper-composer' skill executes systematic writing from outlines with chapter-level quality checks and final evaluation. The system emphasizes quality control through structured rubrics and iterative validation, transforming research ideas into submission-ready manuscripts with comprehensive quality reports. While primarily documentation-heavy, it contains several self-contained, reusable evaluation systems that demonstrate sophisticated assessment frameworks.
Optio is a self-hosted, Kubernetes-native orchestration platform for AI coding agents (Claude Code, OpenAI Codex, GitHub Copilot, Gemini, Cursor, OpenCode). In production it runs three task tiers — repo tasks (issue/request → agent pod → PR), standalone jobs (parameterized agent runs without repo checkout), and persistent agents (long-lived, message-driven agents with turn-based execution) — plus an autonomous PR review loop. The core loop is driven by ~15 BullMQ workers (task, PR watcher, review, webhook, ticket sync, reconcile, workflow trigger) against a Postgres database, with Redis pub/sub powering a real-time event bus and WebSocket log/terminal/chat streaming to the Next.js dashboard and CLI. Agents execute in ephemeral pods built from per-language Docker images; a K8s-style reconciliation control plane (world snapshot + pure decision functions + CAS-gated executor) converges four run kinds and catches drift. The architecture is a pnpm/turbo monorepo: a Fastify 5 API with Drizzle ORM and Zod-validated OpenAPI schemas; a Next.js 15 admin dashboard (tasks, jobs, reviews, sessions, agents, cluster, analytics); a Commander-based CLI (login, task/repo/session/secret commands over REST + WebSocket); and a static docs site. Shared packages provide a clean AgentAdapter interface (`validateSecrets` → `buildContainerConfig` → `parseResult`), a Docker/Kubernetes container runtime abstraction, ticket provider adapters (GitHub, GitLab, Jira, Linear, Notion), and shared types/utilities (task state machines, repo URL parsing, SSRF guards, cost arithmetic). Workspace-scoped multi-tenancy with admin/member/viewer RBAC, AES-256-GCM secret encryption with AAD binding, WebSocket upgrade tokens, HMAC-signed internal calls, and a security-focused audit trail (optio_actions) are enforced throughout. Engineering strengths: the test suite is unusually thorough for a project this size — nearly every route, service, worker, and shared utility has a unit test, with per-file Postgres/Redis isolation for integration tests, real-server E2E tests using a deterministic fake container runtime (with mock directives like [[mock:pr]]), and Playwright smoke tests. The reconcile pattern, generic state machine utilities, and pure decision functions in `packages/shared/src/reconcile/` are the highest-craftsmanship code. Known debt: five near-duplicate NDJSON event parsers (one per agent, with drift risk), hardcoded model pricing tables in the Codex/Gemini adapters, scattered duplicated workspace-scoping checks, a 60+ file migration history with historical prefix collisions (now script-enforced), and raw SQL in the analytics service. The project moves fast (v0.1 → v0.4.1 within months, last commit 2026-08) and some surface area (OpenCode/OpenClaw adapters, skills marketplace) is explicitly experimental, but the core architecture is coherent and production-oriented.
The Agentic Startup repository is a spec-driven development (SDD) framework for Claude Code that enables multi-agent AI collaboration for software development. It provides a structured workflow: specify → validate → implement → test → review → document, with optional skills for analysis, refactoring, debugging, and constitution enforcement. The framework uses three implementation tiers (Direct, Incremental, Factory) selected by an automatic complexity classifier, and enforces quality gates through validation checks, drift detection, and constitution-based governance rules (L1/L2/L3 levels). Key features include progressive disclosure for context efficiency, information barriers to prevent bias, and automatic documentation generation. The system is designed as a plugin-based architecture with 'start' (core workflow) and 'team' (specialized agents) components, storing specifications in .start/specs/ and generating documentation in docs/.
This is a Rust CLI tool called `facts` that implements a fact-driven development workflow. It manages a `.facts` file containing atomic, verifiable claims about a project, each with lifecycle tags (`@draft` → `@spec` → `@implemented`) and optional shell commands for automated verification. The tool provides a rich set of subcommands for listing, adding, editing, removing, linting, and checking facts, along with a terminal UI (TUI) built with Ratatui and Crossterm. It integrates with AI coding agents (Claude, Codex) to verify manual facts via an XML-based verdict protocol, and ships with bundled agent skills for discovery, refinement, and implementation. The project is self-hosting—it uses its own `.facts` file to describe itself—and is distributed via npm, PyPI, and direct install. The architecture is a monolith with clearly separated modules for parsing, writing, ID generation, tag filtering, file locking, and agent communication. The codebase demonstrates strong engineering practices: deterministic round-trip serialization, atomic file writes, advisory locking with stale detection, and a boolean expression evaluator for tag filters. It is actively maintained with a comprehensive test suite and CI/CD pipeline.
This repository is a comprehensive tutorial on building AI agent harnesses, structured as 17 progressive lessons. Each lesson introduces a single mechanism (tool use, permissions, hooks, planning, subagents, skill loading, context compaction, memory, task system, background tasks, cron scheduler, agent teams, MCP integration, workflow runtime, and goal loop) that extends a core agent loop without modifying it. The harness enables a language model (Claude) to perceive, reason, and act through tool interactions while managing context, permissions, and collaboration. The repository includes runnable Python implementations for each lesson and an interactive Next.js web application with visualizations for learning and experimentation.
This repository is a Claude Code/Pi skills framework for academic research and writing. It packages markdown skills, agent definitions, and Python tooling that together support the full research pipeline: topic discovery, deep research, paper writing, review, citation verification, and meta-analysis. The dominant character is application/tooling: it has CLI entry points, a Pi wrapper, dozens of lint/check scripts, and a large test suite. It is not a data pipeline, though it includes network-backed citation lookup scripts and research-oriented data-integrity checks.
This repository is a full-stack, multi-tenant AI API gateway platform (named 'sub2api') that proxies requests to a wide range of AI model providers including OpenAI, Anthropic, Gemini, Grok, DeepSeek, and many others. It provides authentication (JWT, OAuth, API keys), rate limiting, token-level billing, subscription management, user and admin management, payment processing (Alipay, WeChat Pay, Stripe, Airwallex, EasyPay), content moderation, channel monitoring, and operational dashboards. The backend is written in Go using the Gin framework and Ent ORM, with PostgreSQL and Redis for data storage and caching. The frontend is built with Vue 3, TypeScript, and Pinia stores, offering both admin and user interfaces. The system is designed for production deployment with Docker, includes extensive unit and integration tests, and supports a plugin architecture for extensible OAuth transport. Despite its large codebase and many features, the architecture follows a clear service-repository pattern with well-isolated packages, making it a solid example of a complex API gateway implementation.
Serena is a production-grade MCP (Model Context Protocol) server that provides AI coding agents with deep code intelligence across 40+ programming languages. It acts as an 'IDE for your agent', offering symbol-level retrieval, editing, refactoring, and debugging capabilities by orchestrating Language Server Protocol (LSP) backends. The system is designed to integrate with AI clients like Claude Code, Codex, Copilot, and Grok, exposing a rich set of tools for file manipulation, symbol navigation, memory persistence, and project configuration. Architecturally, Serena is a Python application with a CLI entry point that starts an MCP server, a Flask-based web dashboard for monitoring and configuration, and a hooks system for client-side pre-tool-use events. Its core is the `solidlsp` package, a custom synchronous LSP abstraction layer that manages the lifecycle of 40+ language servers (Python, TypeScript, Rust, Go, Java, etc.), handling initialization, capability negotiation, and request/response framing. The `interprompt` library provides a sandboxed Jinja templating engine for prompt generation, and the memory manager persists project-specific context across sessions. Key features include: multi-language symbol search, references, definitions, hover, rename, and diagnostics; a code editor abstraction supporting both LSP and JetBrains backends; a configurable tool registry with automatic discovery; and a comprehensive test suite with per-language integration tests. The project emphasizes reproducibility through pinned dependency versions and SHA256 verification for downloaded language server binaries, though this creates a maintenance burden as versions must be manually updated.
Xonsh is a cross-platform, Python-powered interactive shell that combines Python syntax with shell subprocess commands. The codebase implements a full shell environment with a custom parser (PLY-based with version-specific extensions), an execution engine that compiles xonsh code to Python AST, a rich tab-completion system with multiple backends (bash, git, pip, python, custom completers), a pluggable history backend (JSON, SQLite), a prompt system with git status and virtualenv information, and a subprocess management layer with job control. It also includes CLI tools for formatting, linting, and interactive configuration (xonfig wizard and web UI). The architecture is monolithic with a clear module structure: parsers, executers, completers, prompts, history, procs, and shells. External dependencies include prompt_toolkit, pygments, and PLY (vendored). The repository is actively maintained with frequent releases, supports Python 3.10+, and has an extensive test suite. Some technical drift exists due to deprecated API shims, multiple parser versions, and LLM-generated tests, but the core is well-engineered.
CodeStable is a framework for AI coding assistants (Claude Code, Codex CLI) that provides a set of 'skills' — thin harnesses that guide AI agents through software development tasks such as feature implementation, bug fixing, refactoring, epic decomposition, code review, and lesson management. The system emphasizes risk-based assurance, evidence-driven decisions, and minimal project memory, with a core philosophy of 'thin harness, thick context'. It is distributed as plugins for both Claude Code and Codex CLI, and includes a meta-skill (eval-cs-skill) for authoring, evaluating, optimizing, and releasing skills. The repository contains extensive experiment infrastructure with 15+ structured experiments covering routing accuracy, end-to-end outcomes, planted-defect recall, and learning transfer across multiple models. It features a sophisticated evaluation pipeline with deterministic oracles, hidden tests, regression tests, and checkpointed resume. The codebase demonstrates high craftsmanship with clean abstractions, strict validation, and cognitive honesty tagging (measured/soft/underpowered). Key architectural patterns include the thin skill pattern (compact SKILL.md with progressive disclosure via references), adapter pattern for AI model execution (Claude Code, Codex CLI, API, mock), deterministic seed building for reproducible experiments, and a risk-to-assurance mapping used generically by all task skills. The project has 8 ADRs documenting design decisions and a v1-to-v2 migration path that reduced 32 skills to 8. While the codebase is well-maintained and actively developed, some experiments are marked underpowered and the learning transfer epic was cancelled without completing primary hypotheses. The evaluation system is complex and tightly coupled to specific model names and macOS sandbox profiles, which may limit portability.
Claudoscope is a native macOS menu-bar application that serves as an observability and configuration management dashboard for Claude Code, Anthropic's AI coding assistant. It reads local Claude Code session transcripts (JSONL), caches parsed summaries in a GRDB-backed SQLite store, and presents analytics on token usage, cost, cache effectiveness, model distribution, and attribution per skill, MCP tool, or agent. The app also includes a comprehensive configuration linter (100+ rules for security, skills, hooks, routing, hardening, and more), an installer for Claude Code hardening baselines and agent routing stacks, real-time file monitoring via FSEvents, and an optional MCP server that exposes read-only tools for AI-driven querying. It operates as a background agent with a menu bar popover and a full dashboard window. The application is built as a monolithic SwiftUI macOS app with modular service actors, extensive unit tests (60+ test files), and a clean separation of concerns. Key entry points include the session parser, analytics engine, linter service, and installer file operations. It depends on Claude Code's local file structure and specific JSONL schemas, which drives a medium technical drift risk as those formats evolve. The codebase demonstrates high craftsmanship with reusable patterns such as pure analytic functions, atomic configuration writers, and a design token system, making it a strong example of a well-engineered developer tool.
free-claude-code is a local proxy and developer tool that connects coding agents (Claude Code and similar harnesses) to OpenAI-compatible AI providers. It is a Python package (`free_claude_code`) with a CLI, a local API server, an admin web UI, and a substantial configuration subsystem. The system abstracts over a dozen AI providers (including NVIDIA NIM, Cloudflare, HuggingFace, and others) with provider-specific workarounds for tool parameter aliasing, thinking templates, and reasoning replay. It persists sessions, messages, and model metadata in SQLite, manages a tree-based conversation model with Telegram platform support, and renders transcripts with segment-based truncation and subagent state management. The repo is a shipped developer tool with installer/update scripts, GitHub issue templates, dependabot config, and a diagnostic `fcc-doctor` command; it is not a research or data project.
Chatbox is a cross-platform desktop AI chat client (Windows/macOS/Linux) built with Electron, React, and TypeScript. It supports 30+ AI providers (OpenAI, Claude, Gemini, DeepSeek, Ollama, etc.) via a unified provider registry and model abstraction layer. The application features agent mode with tool use, sandboxed code execution, MCP server integration, knowledge base RAG, web search, image generation, and a skills marketplace. It uses a hexagonal architecture with ports/adapters in a core package (`@chatbox/core`) that is host-agnostic, and React bindings in `@chatbox/react`. The renderer uses TanStack Router, Zustand/Jotai for state, and Mantine/MUI for UI. The main process handles IPC, OAuth flows, file system sandbox, and analytics. The codebase is large and well-tested (Vitest unit tests, integration tests, contract tests) with a focus on offline-first, local storage, and cross-platform compatibility (desktop, web, mobile via Capacitor). The application is actively maintained with a commercial 'Pro' version that shares code with this open-source Community Edition. Notable engineering patterns include a clean port/adapter separation, in-memory test doubles for all repositories, a session write coordinator that serializes updates and cancels stale queries, a draft tokenization system with web workers, and a provider registry that allows dynamic model discovery. The codebase has some technical debt: multiple UI libraries, legacy migration paths, and commented-out code, but overall demonstrates strong craftsmanship and reusability of its core abstractions.
A Neovim plugin that integrates with AI CLI tools (opencode, claude, cursor-agent, gemini) to provide agentic coding workflows. It allows users to perform semantic searches, replace visual selections, run 'vibe' sessions for code generation, and get tutorials. The plugin manages state, request tracking, logging, and window management. It supports multiple providers, custom rules via SKILL.md files, and file completions. The architecture is modular with clear separation of concerns: providers, operations, extensions, and UI layers.
ccNexus is a desktop application (built with Wails v2, Go backend + web frontend) and a headless server that acts as a smart proxy for multiple AI API endpoints, including Claude, OpenAI Chat, OpenAI Responses, and Gemini. It manages endpoint rotation with automatic failover, credential token pools (API keys, access tokens, refresh tokens), and real-time format conversion between the different API protocols. The system provides a full-featured GUI for managing endpoints, viewing usage statistics, configuring backup/restore (WebDAV, local, S3), launching Claude Code or Codex CLI terminals, and applying visual themes. A separate web UI is available for headless deployments via Docker. The architecture follows a layered design: a proxy layer handles request routing, authentication, and streaming response transformation; a transformer layer converts between API formats using a registry of converters; a storage layer persists endpoints, credentials, and stats in SQLite; and a service layer orchestrates higher-level operations like backup, updates, and terminal launching. The codebase demonstrates strong engineering practices including thread-safe configuration, adapter patterns for storage abstraction, SSE for real-time stats, and comprehensive contract tests. Notable features include zero-cost endpoint testing, incremental stats updates, and a credential lifecycle management system with rate limit tracking and automatic cooldown. While primarily an application, ccNexus includes some data-oriented components such as token counting, usage statistics aggregation, and credential rate limit parsing. However, these serve the operational needs of the proxy rather than constituting a data research pipeline. The project is actively maintained with a responsive issue tracker and regular releases.