I pulled every Claude Code skill repo I could find on GitHub, sorted them by stars, and picked out the ones that actually help with day-to-day coding work. Stars aren't everything, but a repo with 50k stars has been tested against real codebases by real developers. That counts for something.
Some of these are genuinely useful, some are overhyped. I've tried to be honest about which is which.
This list will change. The skills ecosystem is still young and new repos are appearing weekly. I'll update it as the good ones emerge.
I reviewed each repo's README, commit history, and community activity before including it here. GitHub stars are my primary ranking signal because they reflect sustained community interest, not just initial hype. I also checked whether each skill's actual functionality maps to coding workflows specifically, not just general productivity, and skipped anything that looked abandoned or that duplicated a built-in Claude Code skill without adding value.
What are Claude Code skills?
Claude Code skills are reusable workflows stored as markdown files (SKILL.md) that Claude loads into its context when you invoke a slash command. Type /review and Claude loads a full code review playbook, spawning subagents, running checks, and returning structured feedback. The skill defines the instructions; Claude executes them against your actual codebase.
Skills differ from MCP servers in scope. An MCP server connects Claude to external data or APIs, like a database or a web search service. A skill is a task-focused workflow that tells Claude how to do something specific, like review code, manage context, or follow a spec-driven development process. Skills are simpler to install and don't require running a separate server process.
Developers, DevOps engineers, and solo builders use skills to encode their best practices once and reuse them across every session. A skill can also be shared across a team by committing it to your repo's .claude/skills/ directory.
How to install Claude Code skills
First, make sure Claude Code is installed and you're inside a project directory. Skills live in the .claude/skills/ directory of your project, or in your global Claude config for skills you want available everywhere.
Create the skills directory inside your project:
mkdir -p .claude/skills
Clone the skill repository into that directory, giving it a short name:
Developers debugging failing tests or runtime errors
52.6k
beginner
top pick
1
gstack
23 coding roles as slash commands: reviewer, eng manager, QA, and more.
69kintermediate
best for · Solo developers and founders shipping production code fast
gstack turns Claude Code into a virtual engineering team by giving you 23 opinionated slash commands, each representing a specific role. The /review command runs a production-focused code reviewer that looks for real bugs, not style nits. The /retro command summarises commits and LOC changes across your projects. Designed by a developer shipping 10,000-20,000 lines of code per day, the skills encode specific opinions about what good code review and release management actually look like.
Spec-driven development system that fights context rot in long coding sessions.
50.2kbeginner
best for · Developers working on multi-session features or large codebases
GSD addresses a specific problem: Claude's output quality degrades as the context window fills during long coding sessions. It does this through meta-prompting and context engineering, giving Claude a structured spec before it writes code and refreshing that spec as the session progresses. It works with Claude Code, Cursor, Gemini CLI, and other coding agents using the same workflow files.
install
npx get-shit-done-cc@latest
example usage
/gsd plan
Pros
✓Directly tackles context degradation mid-session
✓npm install is fast; no git cloning needed
✓Works across Claude Code, Cursor, and Gemini CLI
Cons
✗npm package abstracts the underlying prompts; harder to customise
Persistent session memory that carries your coding context across Claude sessions.
47.1kintermediate
best for · Developers running multi-day or multi-week projects with Claude Code
claude-mem captures everything Claude does during a coding session, compresses it using Claude's agent SDK, and injects relevant context back into future sessions via ChromaDB. For coding, this means Claude remembers the architectural decisions you made last week, the bugs you've already fixed, and the conventions specific to your project, without you having to re-explain them every time.
Multi-agent code quality review that finds and fixes issues in your recent changes.
52.6kbeginner
best for · Developers who want automated quality checks before every commit
The /simplify skill from the awesome-claude-skills collection spawns three parallel review agents, each examining your recently changed files from a different angle: code reuse, quality issues, and efficiency. It then applies the fixes automatically. For coding workflows, this replaces the step where you'd manually scan your own diff before committing, catching things like duplicated logic, unnecessary abstraction, and inefficient loops.
Structured PR-style code review as a slash command inside your terminal.
52.6kbeginner
best for · Solo developers and small teams wanting automated pre-commit review
The /review skill from awesome-claude-skills runs a full code review against your changes, following a structured checklist that covers correctness, edge cases, security issues, and readability. For coding teams, it replicates a pull request review without needing a second developer, which is useful during solo development or when you want a pre-review pass before requesting human review.
Build your own coding agent harness from scratch with documented internals.
51.2kadvanced
best for · Developers building custom coding skills or agent harnesses
learn-claude-code is a teaching repository that walks through the complete architecture of a Claude Code-style coding agent: the agent loop, tools, planning, context management, permissions, hooks, and memory. For developers who want to write custom skills or build their own coding agents, understanding these internals is essential. The repo explains each mechanism before relying on it, making it a practical reference rather than just theory.
install
See README for setup instructions
example usage
Reference for writing custom SKILL.md workflows
Pros
✓Explains agent loop and context control with real code
✓Covers hooks, memory, and subagent delegation in depth
✓Directly applicable to writing your own coding skills
Cons
✗Not a drop-in skill; requires reading and adaptation
✗Deliberately omits production edge cases to stay focused on teaching
Long-horizon coding tasks delegated to sandboxed sub-agents with memory.
60.1kadvanced
best for · Developers with long-horizon coding tasks requiring research and iteration
DeerFlow's coding sub-agent handles tasks that take minutes to hours: researching a problem, writing code, running it in a sandbox, and iterating based on output. For complex coding tasks like building a new module from a spec, debugging a hard-to-reproduce issue, or generating test suites, it provides the scaffolding to keep a long-running coding task coherent across many agent steps.
install
See README for setup instructions
example usage
/code implement a rate limiter with Redis backend and full test suite
Pros
✓Sandboxed execution prevents unsafe code from running on your machine
✓Memory system keeps long tasks coherent across many steps
✓Handles research plus implementation in one workflow
Cons
✗Python 3.12+ and Node 22+ required; heavier setup than most skills
✗Overkill for short coding tasks; startup overhead not worth it
Structured debugging workflow that traces errors from stack trace to fix.
52.6kbeginner
best for · Developers debugging failing tests or runtime errors
The /debug skill runs a structured debugging process: it reads the error or failing test, traces the call stack, identifies the root cause, and applies a fix. For coding workflows, it replaces the ad-hoc process of pasting errors into a chat and hoping for a good response. The structured approach means Claude doesn't jump to a fix without first confirming its understanding of the root cause.
The best skills target a specific, recurring pain point: context degradation mid-session, boilerplate code review, missing project context. If a skill's description is vague, check the README for concrete before/after examples. Vague skills produce vague results.
Does it fit your workflow stage?
Some skills help at the start of a feature (spec-driven planning), others during development (context memory), and others at the end (code review, simplification). Pick skills that cover the stages where you lose the most time, not the ones that sound impressive.
How opinionated is it?
Skills like gstack come with strong opinions baked in: specific roles, specific review criteria, specific output formats. That's useful if the opinions match yours. It's friction if they don't. Read the skill's SKILL.md or command files before committing.
Is it actively maintained?
Check the last commit date. A skill that hasn't been updated since early 2025 may not account for changes in Claude Code's skill system, like the merge of commands and skills into a unified format. Stale skills still run, but they may miss newer capabilities like subagent delegation or frontmatter context control.
Can you customise it?
The best skills are starting points, not black boxes. If the repo is just a markdown file, you can fork it and edit the instructions directly. If it's a compiled binary or an opaque npm package, you're dependent on the maintainer. Prefer skills where you can read and modify the underlying prompt.
frequently asked questions
Most do, with caveats. Skills like claude-mem and get-shit-done work at the session level and don't care about your repo structure. Skills that spawn subagents for file review (like the gstack reviewer) may need you to point them at the right subdirectory. Check whether the skill accepts a path argument before assuming it auto-discovers your monorepo layout.
CLAUDE.md provides persistent project context that Claude reads at the start of every session. Skills provide task-specific instructions loaded on demand. They work together: CLAUDE.md tells Claude about your codebase, tech stack, and conventions; a skill tells Claude how to execute a specific task within that context. There's no conflict, but avoid duplicating the same instructions in both places or Claude will weight them redundantly.
An MCP server gives Claude access to external data or tooling, like a linter API, a database, or a web search service. A skill gives Claude a workflow: a set of instructions for how to approach a class of task. For coding, you might use an MCP server to pull in your issue tracker data and a skill to define how Claude should turn that data into implementation steps. They solve different problems and can be used together.
Yes. A minimal skill is just a markdown file named SKILL.md inside a folder under .claude/skills/. Write the steps you'd tell a new developer to follow for a specific task, and Claude will follow them. Start with something small, like a skill that runs your test suite and summarises failures, then iterate. The learn-claude-code repo is a good reference for understanding the underlying mechanics if you want to go deeper.
Skills only load when invoked, so having several installed doesn't add overhead to normal sessions. Skills that spawn multiple subagents (like gstack's parallel review roles) do take longer to run because they're doing more work. Budget a minute or two for multi-agent skills on larger changesets. Single-agent skills like get-shit-done's context refresh are much faster.
tools for humans
toolsforhumans editorial team
Reader ratings and community feedback shape every score. Since 2022, ToolsForHumans has helped 600,000+ people find software that holds up after launch. The picks here come from that.
