Claude Code’s plugin and skills ecosystem has exploded. There are now dozens of framework repos promising to turn your terminal agent into a disciplined engineering team. I reviewed nine of the most popular ones so you don’t have to.
How I Ranked Them
Five weighted criteria:
- Immediate usefulness (30%) — can you install it and get value today?
- Methodology depth (20%) — is there a real workflow behind it, or just prompts?
- Accessibility (20%) — how steep is the learning curve?
- Cross-platform support (15%) — does it work beyond Claude Code?
- Community & maintenance (15%) — is it actively developed with real users?
The Rankings
1. Superpowers (~156k stars)
A composable skills library enforcing a spec-first, TDD/YAGNI/DRY workflow. Skills like systematic-debugging, brainstorming, subagent-driven-development, and code-review trigger automatically before each task. Works with six major agents out of the box via official plugin installs. The mandatory brainstorming → plan → subagent → TDD → review pipeline is opinionated but genuinely productive. The strongest all-rounder — install this one first if you install nothing else.
2. gstack (~74k stars)
Garry Tan’s personal setup: 23 opinionated tools that turn Claude Code into a virtual engineering team with distinct roles (architect, reviewer, debugger, etc.). Covers the full pipeline from planning through deployment. Well-documented CLAUDE.md. Requires Bun/Node, and the learning curve is steep because there are 23 tools to understand. But if you want a comprehensive, role-based workflow, this is the most thoughtful implementation.
3. get-shit-done (GSD) (~54k stars)
Built around avoiding context rot by spawning fresh sub-agents for each task. Supports a remarkable 15 runtimes. The planning directory can get messy on larger projects, and the repo includes a $GSD token mention which feels off-putting. But the core architecture — isolated agents that don’t accumulate stale context — solves a real problem that most frameworks ignore.
4. everything-claude-code (~158k stars)
The biggest collection by raw numbers: 48 agents and 183 skills, polyglot support. Ambitious scope but confusing packaging — it’s hard to know where to start, and quality varies across the massive skill library. Best treated as a buffet you pick from rather than a coherent system.
5. Spec Kit (~89k stars)
A GitHub-native, spec-driven development framework. Enterprise-grade rigour with formal specification documents driving all code generation. Supports 30+ agents — the broadest compatibility of any framework. The trade-off is heavy upfront work writing specs before any code gets generated. Best suited for teams that already practice spec-driven development.
6. oh-my-claudecode (~28k stars)
A unique approach: multi-model tmux orchestration that routes tasks to cheaper models (Gemini, Codex) for routine work while reserving Claude for complex reasoning. Claims 30–50% token savings. Clever architecture but tightly coupled to Claude Code internals, and breaking changes are common. For power users comfortable with tmux-based workflows.
7. BMAD Method
Agile methodology AI-ified — product owner, scrum master, architect roles mapped to agent personas. Thorough if you’re running a structured team process. Overkill for solo developers. The methodology is sound but the overhead only pays off on larger, multi-person projects.
8. shanraisshan/claude-code-best-practice (~43k stars)
A comprehensive knowledge base and guide rather than an installable tool. Excellent for learning Claude Code deeply — prompt patterns, workflow tips, best practices. Not something you “install” to get immediate productivity, but the best educational resource in this list.
9. davila7/claude-code-templates
A component marketplace of portable markdown templates. Quality varies and tooling assumes Claude Code specifically. Useful as a starting point for building your own custom templates, less useful as a turnkey system.
They’re Not Claude Code Exclusive
A common misconception: most of these frameworks work across multiple coding agents. The ecosystem is clearly moving toward universal compatibility.
| Framework | Claude Code | Codex (OpenAI) | Cursor | Gemini CLI | OpenCode | GitHub Copilot | Others |
|---|---|---|---|---|---|---|---|
| Superpowers | ✅ | ✅ (CLI + App) | ✅ | ✅ | ✅ | ✅ | — |
| gstack | ✅ | ✅ | ✅ | — | ✅ | — | Factory Droid, Slate, Kiro, Hermes, GBrain, OpenClaw |
| GSD | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | Windsurf, Antigravity, Augment, Trae, Qwen Code, Cline, CodeBuddy, Kilo |
| everything-claude-code | ✅ | ✅ (CLI + App) | ✅ | ✅ | ✅ | — | Antigravity |
| Spec Kit | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | 30+ agents listed |
| oh-my-claudecode | ✅ | ✅ (via tmux) | — | ✅ (via tmux) | — | — | — |
| BMAD Method | ✅ | ✅ | ✅ | — | ✅* | — | Antigravity (community) |
| shanraisshan | ✅ | Separate repo | — | — | — | — | Claude Code-focused |
| davila7 | ✅ | — | — | — | — | — | Portable markdown, but tooling targets Claude Code |
(* = community/experimental support)
The standouts for cross-platform breadth are GSD (15 runtimes), Spec Kit (30+), and Superpowers (6 major agents with official plugin installs). The most tightly Claude Code-coupled are shanraisshan (educational content specific to Claude Code features), davila7 (CLI tooling assumes Claude Code), and oh-my-claudecode (deeply integrated with Claude Code internals, though it shells out to other models for multi-model orchestration).
My Recommendation
If you only install one, make it Superpowers — it’s the most immediately productive with the broadest agent support. If you want a comprehensive role-based workflow and don’t mind setup time, pair it with gstack. If context rot is your main pain point, GSD solves that specific problem well.
For learning, read through shanraisshan regardless of what you install — it’ll make you better at working with any of these frameworks.
Generally, don’t install more than two. Pick the ones that match how you work, go deep, and iterate. The best framework is the one you actually use.
Leave a Comment