GitHub Trending AI Developer Tools: 2026-05-13 Ranking and Engineering Signal

Three repositories deserve immediate inspection: rohitg00/agentmemory (persistent memory for coding agents, Apache-2.0, actively shipping with over six thousand five hundred stars), obra/superpowers (composable agent workflow skills with over one hundred eighty-eight thousand total stars and the largest community in this ranking), and mattpocock/skills (opinionated engineering skills for Claude and Codex agents, MIT license). tinyhumansai/openhuman is an ambitious Rust-based personal AI harness with token compression and over one hundred integrations, but its GPL-3.0 copyleft license and early-stage status warrant caution before adoption. millionco/react-doctor brings AI-aware code quality scanning to React projects with agent integration. rasbt/LLMs-from-scratch is a well-established educational resource that remains relevant as LLM architecture evolves. github/spec-kit is a new specification tool from GitHub that appeared with nearly thirteen hundred stars gained in a single day and is worth monitoring.

All star counts and push timestamps reflect GitHub API state as of the enrichment window. The "Stars Gained" column represents the daily delta reported by GitHub Trending; it is a short-window attention metric, not a durability signal (GitHub Trending). Total star counts are provided for enriched repositories only. Repositories without enrichment data show N/A in the Total Stars and License columns.

Cluster 1: Agent Memory, Skills, and Workflow Infrastructure

The dominant signal in today's ranking is the maturation of the agent infrastructure layer. rohitg00/agentmemory continues its multi-day trending run with over six thousand five hundred total stars and a release cadence that now includes Codex plugin platform support and OpenClaw compatibility fixes (commit 25dddc43798c). The repository exposes a four-tier memory consolidation pipeline (Working to Episodic to Semantic to Procedural), triple-stream hybrid search via BM25, vector, and knowledge graph, and over fifty MCP tools. The Apache-2.0 license and npm/Docker install paths make it the most adoptable memory layer in the current ecosystem.

obra/superpowers carries the largest existing community in this ranking at over one hundred eighty-eight thousand stars. It provides composable skill definitions for AI agent workflows — brainstorming, dispatching parallel agents, systematic debugging, test-driven development, and writing plans. The philosophy is model-agnostic and agent-agnostic: skills are markdown-based instruction sets that any compatible coding agent can consume. The MIT license and established maintenance history make this a low-risk inspection target for teams building agent-assisted engineering processes.

mattpocock/skills gained the highest star delta among enriched repositories at three thousand eight hundred sixty-seven, reaching over seventy-five thousand total stars. The repository focuses on what Matt Pocock terms "real engineering, not vibe coding" — structured skill definitions that prevent agents from losing intent, being verbose, or creating architectural decay. Skills include engineering workflows (issue triage, domain doc layout, TDD loops) and productivity patterns (compressed communication, skill authoring). The MIT license and the author's established reputation in the TypeScript community make this a low-risk, high-value reference.

Cluster 2: Personal AI Agent Harnesses

tinyhumansai/openhuman holds the top trending position with over one thousand stars gained today and a total approaching three thousand five hundred. Written in Rust with a Tauri desktop shell, it positions itself as a "Personal AI super intelligence" — a local-first agent that integrates with over one hundred third-party services via OAuth, runs a Memory Tree knowledge base backed by SQLite, and compresses token usage through a feature called TokenJuice. The README includes a feature comparison table against other agent harnesses. The project is in early beta with active daily commits (v0.53.35 as of today). However, the GPL-3.0 copyleft license, ninety open issues, and beta status mean this is inspection territory, not production-ready infrastructure.

Cluster 3: AI-Assisted Code Quality and Education

millionco/react-doctor is a one-command React codebase health scanner from the Million.js team (YC W24). It runs a diagnostic pass using Oxlint and Knip, produces a zero-to-one hundred health score, and outputs actionable findings across state management, performance, architecture, security, and accessibility. Notably, it includes an install command that teaches coding agents React best practices via a SKILL.md integration. This positions it at the intersection of static analysis tooling and AI agent workflow — the scanner finds problems, and the agent skill prevents them from recurring. The MIT license and documented API make it straightforward to evaluate.

rasbt/LLMs-from-scratch continues to trend as a foundational educational resource. Authored by Sebastian Raschka and published by Manning, it covers coding GPT-like models from the ground up, with implementations of Llama, Qwen, Gemma, and Olmo architectures. The companion seventeen-hour video course and exercise solutions make it a comprehensive reference for understanding the model architectures that AI developer tools increasingly depend on. This repository trends because the developer community values foundational literacy as the agent ecosystem accelerates.

Inspect first: rohitg00/agentmemory. The problem it solves (context loss between coding agent sessions) is concrete and well-scoped. The Apache-2.0 license, npm/Docker install paths, and MCP integration make it adoptable without licensing or infrastructure lock-in. It has been trending for multiple consecutive days, indicating sustained developer interest beyond a single social media spike. Check the iii engine dependency and local storage model before committing to production use.

Inspect second: obra/superpowers. With the largest community in this ranking, this repository represents a battle-tested approach to structuring agent workflows. The skill definitions cover the full software development lifecycle from planning through debugging. Even teams that do not adopt the specific skills directly can use them as reference patterns for their own agent instruction sets.

Inspect third: mattpocock/skills. The README articulates agent failure modes with unusual clarity. Even if you do not adopt the specific skill definitions, the diagnostic framework for understanding why agents produce bad output is valuable. The install path is non-destructive — skills can be evaluated in isolation.

Evaluate cautiously: tinyhumansai/openhuman. The Rust-based agent harness, token compression, and integration breadth are interesting, but GPL-3.0 copyleft, ninety open issues, and beta status mean this is early-adopter territory. Monitor the release cadence and issue resolution rate before investing integration effort.

Worth watching: github/spec-kit. A new specification tool from GitHub with nearly thirteen hundred stars gained in one day is a strong attention signal. However, without enriched metadata, the AI-developer-tool fit cannot be confirmed. Check the repository description and README before allocating inspection time.

Do not inspect for AI dev tool use: yikart/AiToEarn (AI-powered earning platform), CloakHQ/CloakBrowser (anti-detect Chromium with dual-use risk), and apernet/hysteria (network proxy). These repositories trended but do not solve AI-developer-tool problems.