Five Skill Sets for AI Agents on GitHub: Superpowers Has 285.5K Stars
Coding agents like Claude Code and Codex can write code, but they don't know your workflow: planning, tests, review. To avoid repeating that in every prompt, developers create "skills", ready-made instructions for the agent. The leader among the five sets KDnuggets covered, Superpowers, has collected 285.5K stars (GitHub's count of repository bookmarks). Pick one set and test it on a real task.
AI-processed from KDnuggets; edited by Hamidun News
Claude Code, Codex, Cursor, and OpenCode already write code well. But development is more than code: you need to understand requirements, plan a feature, write tests, find the cause of a bug, verify changes, and decide what is ready to ship. If you explain that sequence to the agent again in every prompt, you waste a lot of time, and the results still drift. That is why "skills" have appeared: reusable instructions and scripts that teach the agent to carry out a specific task the way you do. KDnuggets reviewed five popular open collections of such skills. Popularity is measured by GitHub "stars", which are bookmarks users give repositories they like. The leader, Superpowers, has 285.5 thousand.
Full Methodology: Superpowers and ECC
Superpowers by Jesse Vincent and the Prime Radiant team is one of the most popular projects of this kind on GitHub. It is not a set of isolated tricks but a complete development process for the agent. Imagine asking it to add a new feature to an app. The agent first works out what you actually need (the brainstorming skill). Once you approve the design, it writes a plan and implements it through subagents. Code is written with TDD, meaning tests first, then implementation. Review and a final check follow before the branch is finished.
Debugging works the same way: the systematic-debugging skill looks for the root cause instead of changing code at random until something works. In Claude Code, the set is installed with /plugin install superpowers@claude-plugins-official. It supports Codex, Cursor, Gemini CLI, OpenCode, GitHub Copilot CLI, and other environments.
ECC is the largest collection on the list. Its author, Affaan, describes the cycle as: plan → test → implement → review → verify → remember → improve. It includes not only skills but also specialized agents, rules, hooks, memory, continuous learning, security tools, and integrations. The last steps of the cycle are especially interesting: ECC preserves useful context between sessions and turns recurring techniques into new skills. A separate tool, AgentShield, checks agent configurations for secrets, risky permissions, hooks, and MCP settings. The article does not give ECC's star count.
Installation: npx [email protected] setup, or a step-by-step wizard via install --guided. Claude Code has the strongest support, and other environments vary in how well they are covered. If you only need a good debugging prompt, this is clearly overkill. If you want a full engineering platform with memory and security, it is one of the most ambitious options.
Compact and Applied Sets
Skills for Real Engineers by Matt Pocock (260.1K stars) takes a different approach. The author uses it every day for "real engineering, not vibe coding". The skills are small, flexible, and designed to combine. For example, improve-codebase-architecture does not let the agent guess: it asks you about the change, builds a shared model of the domain, clarifies terminology, and updates the project documents. For larger tasks you can chain skills: grill-with-docs → to-spec → to-tickets → implement → code-review. Installation: npx skills@latest add mattpocock/skills. The installer lets you choose skills and agents and copies the files into your project, so you can edit them to fit your needs.
Awesome LLM Apps by Shubham Saboo (137.3K stars) is more a showcase than a skill set: it contains over 100 open AI agents, skills, and RAG applications built on GPT, Claude, Gemini, DeepSeek, Llama, and Qwen. It is useful for seeing how agents, tools, search, and multiple models come together into working applications.
Among its skills is Project Graveyard, which scans your local Git repositories, finds abandoned projects, and suggests which one is worth finishing. Another skill checks whether a diff has grown beyond the original goal, and another uses Git history to explain why a piece of code exists. Self-Improving Agent Skills automatically tests and improves skills, while Advisor Orchestrator Worker splits a large task among an orchestrator, an advisor, and workers.
Agent Skills by Addy Osmani (93.6K stars) is aimed at production. The process is divided into six stages: DEFINE → PLAN → BUILD → VERIFY → REVIEW → SHIP. It covers requirements, TDD, frontend, API design, debugging, security, performance, review, CI/CD, observability, and documentation. Lifecycle commands load the right playbooks for each stage. The reviewer especially liked the constraint-driven-development skill. It sets requirements for tests, linting, security, coverage, and performance, and moves expensive checks toward the end instead of running them after every edit. Installation: npx skills add addyosmani/agent-skills. You can first list the options with --list or install a single skill with --skill test-driven-development. Claude Code, Cursor, Codex, Copilot, and Cline are supported.
What to Do With This
The approaches differ considerably: Superpowers offers a methodology, Pocock offers small building blocks, ECC offers a platform with memory, Awesome LLM Apps offers a testing ground, and Osmani offers production discipline. The reviewer advises against installing everything. Choose workflows that fit your style, test them on real projects, and adjust them to your needs.
Keep one limitation in mind: GitHub stars indicate popularity, not quality. The material contains no measurements of how much these skills actually speed up work or reduce errors. So test them yourself: take one recurring task, such as review or debugging, and compare the results with and without the skill. For a team, this is a chance to write your development rules down once instead of repeating them in every prompt.
- Nov 21, 2026Parcha выложила AgentRun: ревью кода дешевеет с $1 до 18 центов за проверку
- Nov 21, 2026Ai2 tested five top models: they focus on the real problem in only 35–41% of answers
- Nov 21, 2026Google AdMob Teaches Claude Code, Codex and Cursor to Add Ads to Apps
- Nov 21, 2026Anthropic to Sell Shares at $83 and Raise More Than $94 Billion in Record IPO
Key Questions Answered
What are the key details on Full Methodology: Superpowers and ECC?
Superpowers by Jesse Vincent and the Prime Radiant team is one of the most popular projects of this kind on GitHub. It is not a set of isolated tricks but a complete development process for the agent. Imagine asking it to add a new feature to an app. The agent first works out what you actually need (the brainstorming skill). Once you approve the design, it writes a plan and implements it through subagents. Code is written with TDD, meaning tests first, then implementation.
What are the key details on Compact and Applied Sets?
Skills for Real Engineers by Matt Pocock (260.1K stars) takes a different approach. The author uses it every day for "real engineering, not vibe coding". The skills are small, flexible, and designed to combine. For example, improve-codebase-architecture does not let the agent guess: it asks you about the change, builds a shared model of the domain, clarifies terminology, and updates the project documents. For larger tasks you can chain skills: grill-with-docs → to-spec → to-tickets → implement → code-review.
What to Do With This?
The approaches differ considerably: Superpowers offers a methodology, Pocock offers small building blocks, ECC offers a platform with memory, Awesome LLM Apps offers a testing ground, and Osmani offers production discipline. The reviewer advises against installing everything. Choose workflows that fit your style, test them on real projects, and adjust them to your needs. Keep one limitation in mind: GitHub stars indicate popularity, not quality.
Build & Ship Production AI Agents Without Boilerplate Code
Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.