15 most-starred Claude Code GitHub repositories in 2026.

The Claude Code ecosystem now includes hundreds of repositories promising better memory, lower token use, stronger workflows, reusable skills, agent orchestration, and easier setup. GitHub stars help identify attention, but they do not establish security, maintainability, or fit for a production engineering team.

This guide ranks directly relevant repositories by GitHub stars, then adds the information a team needs before installation: category, license signal, operational purpose, and the primary review question.

Methodology: Datrick queried the GitHub repository search API for repositories tagged or clearly described as supporting Claude Code, sorted by stars, on July 11, 2026. We manually removed generic AI applications with only incidental Claude support and added Anthropic's official Claude Code repository and gstack as directly relevant reference projects. Stars and metadata are a point-in-time snapshot.

Top Claude Code repositories by GitHub stars

Important: “No SPDX license reported” does not prove that no terms exist. It means the repository metadata returned by GitHub did not identify a standard SPDX license. Review the repository's actual license and usage terms before copying or distributing anything.

1. ECC: a broad agent harness

ECC combines skills, memory, security controls, research-first development, and agent-harness optimization across Claude Code and other coding agents. Its breadth is the attraction and the main review burden.

Review before adoption: identify every installed command, hook, permission, memory location, and external integration. A broad harness can improve consistency, but it also changes more of the development environment than a single-purpose skill.

2. Anthropic Claude Code: the official baseline

anthropics/claude-code is the official repository for Anthropic's terminal-based agentic coding tool. Use its documentation and release information as the baseline before adding community abstractions.

Review before adoption: confirm Anthropic's current installation, commercial, data, permission, and platform documentation rather than inferring rights from GitHub metadata alone.

3. Gstack: opinionated delivery roles

gstack packages structured roles and workflows for product, design, engineering, QA, release, and documentation work. It is relevant when a team wants repeatable operating patterns around Claude Code rather than isolated prompts.

Review before adoption: decide which opinions match the team's software lifecycle, approval boundaries, and deployment process. More workflow is not automatically better for a small or regulated change.

4. CC Switch: a multi-agent desktop client

CC Switch presents a cross-platform client for switching among Claude Code and other coding-agent environments. It can reduce setup friction for developers using multiple providers.

Review before adoption: inspect credential storage, provider configuration, update distribution, local data, and how the application launches or proxies each agent.

5. UI UX Pro Max Skill: design guidance

UI UX Pro Max Skill supplies design-oriented instructions for agent-assisted interface work. It is useful when teams need stronger visual and interaction constraints than a general coding prompt provides.

Review before adoption: test output against the existing design system, accessibility requirements, browser support, and product audience. A popular design skill should not override local standards.

6. Claude Mem: persistent session context

Claude Mem captures agent activity, compresses it, and brings relevant context into later sessions. That addresses a real continuity problem, but persistent memory changes the data boundary.

Review before adoption: document what is captured, where it is stored, how it is summarized, who can read it, how secrets are excluded, and how data is deleted.

7. Graphify: code and schema knowledge graphs

Graphify turns code, database schemas, infrastructure, documents, and other artifacts into a queryable knowledge graph. It can support codebase understanding and cross-system investigation.

Review before adoption: test indexing scope, data residency, stale-graph behavior, secret exclusion, and whether the graph improves retrieval accuracy on the team's actual repositories.

8. Agent Skills: reusable engineering guidance

Agent Skills provides production-oriented engineering skills for coding agents. Skills are easier to pilot than a full harness because the team can select a bounded capability.

Review before adoption: inspect each skill independently. Verify commands, assumptions, generated artifacts, framework versions, and whether instructions conflict with repository rules.

9. Learn Claude Code: understand the harness

Learn Claude Code is a learning-oriented, minimal agent-harness project. Its value is architectural understanding rather than installing a large productivity layer.

Review before adoption: treat it as educational code unless the repository explicitly supports the production requirements you need. Learning projects and operational dependencies have different maintenance expectations.

10. RTK: reduce command-output tokens

RTK proxies common development commands and compresses their output before it reaches the model. This can reduce context consumption and cost when verbose tool output is a measurable problem.

Review before adoption: verify that compression never removes failure details, warnings, file paths, line numbers, or evidence required for debugging and audit.

Other notable repositories

• Awesome Claude Skills: useful for discovery, but every linked skill requires its own license and security review.
• Get Shit Done: a focused spec-driven workflow for teams that want context engineering and staged implementation.
• Ruflo: broad multi-agent orchestration that should be evaluated as a platform, not a prompt collection.
• Claude Code Best Practice: a learning and practice resource; validate recommendations against current official behavior.
• Awesome Claude Code: a large ecosystem directory, useful for research but not an endorsement of every linked project.

Security checklist before installing a Claude Code repository

1. Confirm ownership and source. Watch for forks, typo-squatted packages, renamed organizations, and unofficial installers.
2. Read the license and terms. Stars do not grant commercial use or redistribution rights.
3. Inspect installation scripts. Review shell commands, package lifecycle scripts, downloaded binaries, and elevated privileges.
4. Review hooks and commands. Identify every automatic action before, during, and after Claude Code tool calls.
5. Map data access. Check repository files, home-directory data, conversation logs, memory stores, network destinations, and telemetry.
6. Protect credentials. Do not expose production tokens, cloud credentials, SSH keys, customer data, or unrestricted MCP tools during a pilot.
7. Test in isolation. Use a disposable repository and constrained credentials before a company codebase.
8. Pin and monitor versions. Record the approved version, review updates, and retain a removal path.

How teams should choose

Start from the operating problem, not the leaderboard. If sessions lose context, evaluate a memory or graph tool. If delivery lacks structure, evaluate one workflow framework. If tool output consumes excessive context, test a token optimizer. If the team needs reusable expertise, pilot one skill.

Measure task completion, review defects, time saved, token use, developer adoption, and security exceptions. Avoid installing several overlapping frameworks at once; attribution becomes impossible and conflicting instructions degrade agent behavior.

Frequently asked questions