Everything Claude Code hits 170K stars: what developers should know

I've watched the Everything Claude Code repo grow from a hackathon project into one of the most-starred repositories on GitHub, and the trajectory makes sense. Everything Claude Code (ECC) solves a problem every team hits once they start taking AI coding agents seriously: the defaults are generic, the configuration is per-project, and the work doesn't carry over.

Created by Affaan Mustafa, ECC packages 48 agents, 184 skills, hooks, rules, and security scanning into a portable, version-controlled system that works across Claude Code, Cursor, Codex, and OpenCode. At 170K stars and 26.4K forks, it's the most visible attempt at a shared configuration layer for AI coding agents I've seen.

What Happened

ECC v2.0.0-rc.1 shipped on April 28, 2026, adding a Rust-based control plane prototype, operator workflows, and a desktop dashboard GUI. The repository now includes 48 agents, 184 skills, and 79 legacy command shims across 12 language ecosystems, with 168 contributors and 12 releases since January 2026.

What I'd highlight about v2.0.0-rc.1 specifically: it's not an incremental release. The Rust control plane in ecc2/ builds locally and exposes dashboard, start, sessions, status, stop, resume, and daemon commands. It's usable as an alpha. That's a sign this project is thinking beyond configs toward a more durable architecture.

The selective install architecture is the other thing worth noting. A Python shop installs only Python rules. A Go team installs only Go agents. No noise, no duplication.

Key Features

• Cross-tool support. Works with Claude Code, Cursor, Codex (macOS app and CLI), OpenCode, Gemini CLI, and Antigravity. Each tool gets adapted configs: .cursor/, .codex/, .opencode/, and .gemini/ directories ship in the repo.
• AgentShield security scanner. Runs three Claude Opus 4.6 agents in a red-team/blue-team/auditor pipeline to scan CLAUDE.md, MCP configs, hooks, and agent definitions. It covers 102 static analysis rules with 1,282 tests. This is the feature I'd evaluate first for any team running MCP configs in production.
• Selective install architecture. Manifest-driven pipeline with install-plan.js and install-apply.js lets teams install only the components they need. A state store tracks what's installed for incremental updates.
• Hook runtime controls. ECC_HOOK_PROFILE=minimal|standard|strict and ECC_DISABLED_HOOKS environment variables let developers tune strictness without editing hook files. Useful for teams with different risk tolerances across environments.
• Continuous Learning v2. An instinct-based system that extracts patterns from coding sessions, scores them by confidence, and clusters related instincts into reusable skills via /evolve. The agent gets better the more your team uses it.
• Token optimization defaults. Ships recommended settings.json values: Sonnet as the default model, 10K thinking tokens, and a 50% autocompact threshold. That combination reduces routine task costs by roughly 60% without affecting output quality.

Why It Matters

Most AI coding agents ship as black boxes. Developers get a prompt interface and limited control over how the agent behaves across sessions, handles security, or manages context windows. ECC treats agent configuration as an engineering problem: versioned, testable, and portable across tools.

For teams, that means shared agent behavior. A code-reviewer agent with specific security rules behaves the same whether a teammate runs it in Cursor or Claude Code. The hook system catches issues like console.log statements, secrets in prompts, and dev servers running outside tmux before code ships.

I'm also seeing a broader shift here. Developers are moving past "try the AI tool" toward "configure the AI tool for production use." ECC's growth to 170K stars signals that structured agent workflows, not raw model capability, are the actual bottleneck. That's a meaningful change in how teams are thinking about this.

Example Use Case

A team building a Next.js and Supabase SaaS app installs ECC via the Claude Code plugin, then copies rules/common and rules/typescript to ~/.claude/rules/. When a developer runs /ecc:plan "Add user authentication with OAuth", the planner agent creates an implementation blueprint. They follow with /tdd to write failing tests first, then /code-review for a quality pass.

Before deploying, npx ecc-agentshield scan --opus --stream runs three Claude Opus 4.6 agents in a red-team/blue-team/auditor pipeline against their Claude Code configuration, flagging injection risks and misconfigured MCP servers. The entire workflow uses Sonnet by default, switching to Opus only for deep architectural reasoning via /model opus.

This is the workflow I'd walk through with a team that's been spending too much time on agent setup. ECC compresses that setup into a single install.

Competitive Context

ECC sits atop multiple tools rather than replacing any of them. Claude Code offers the deepest integration: 48 agents, 184 skills, and 8 hook event types. Cursor receives 15 hook event types via a DRY adapter that reuses Claude Code's hook scripts without duplicating them. Codex lacks hook execution parity entirely, so ECC compensates with instruction-based enforcement via AGENTS.md and sandbox permissions.

No other open-source project currently ships cross-tool agent configurations at this scale. ECC fills the gap between "install an AI coding tool" and "run an AI coding tool with consistent, auditable behavior." AGENTS.md at root is the universal cross-tool file read by all four supported tools, which is the architectural bet worth paying attention to.

The 168-contributor community and the Rust-based v2 architecture in-tree tell me this project is planning for the long term. Worth adopting now rather than after you've already rebuilt the same scaffolding yourself.

My Take

Everything Claude Code is a structured configuration system for AI coding agents that now supports 12 language ecosystems and 184 skills. Teams running Claude Code, Cursor, or Codex in production should evaluate it for standardizing agent behavior, enforcing security policies, and reducing token costs.

At 170K stars and 26.4K forks, the community weight behind it is real. I'd rather adopt a system that's already been stress-tested across this many codebases than build the same scaffolding from scratch.