{"content_id":"meiwm0e86m","slug":"top-5-ai-agent-skills-comparison","locale":"en","schema_type":"TechArticle","category":"ai_data","category_name":"AI Data","title":"Top 5 Agent Skills Compared: The Pitfalls of Popularity Rankings and Selection Criteria","summary":"This article compares five agent skill repositories included in the snapshot provided on August 10, 2026, focusing on their features and design philosophies. It also explains why repository stars and install counts do not indicate actual skill quality, along with criteria for verifying licenses, tokens, and security.","sponsorship_disclosure":null,"author":{"name":"Injoys Editorial Team","url":"https://injoys.com/ko/about"},"key_points":["An agent skill is a work package that bundles the instructions, scripts, examples, and reference materials needed for a specific task and loads them when required.","The provided Top 5 order is merely a repository-level popularity snapshot, not a globally recognized ranking that proves the usage or quality of individual skills.","The leading candidates share a common approach: rather than coding immediately, they first require clarification of requirements, planning, testing, minimal changes, and verification.","Activating a large bundle of skills all at once can increase context costs and instruction conflicts, so only the necessary items should be selected.","Before installing an external skill, check its license, execution scripts, network access, permission scope, and maintenance status."],"content_markdown":"Agent skills are reusable work packages that enable AI agents to perform specific tasks consistently. They are broader in scope than simple prompts and can include not only instructions but also scripts, templates, examples, and reference materials.\n\nThis article analyzes five popular candidates presented as of August 10, 2026. However, because there is no raw data or point-in-time archive that would allow the exact figures in the provided materials to be independently reproduced, they should not be interpreted as a definitive “global usage ranking.”\n\n## What Are Agent Skills?\n\nThe core concept of Agent Skills, as described by Anthropic, is **progressive disclosure**. Instead of reading every work instruction from the outset, an agent first reviews the names and descriptions of skills, then identifies only those relevant to the current task and loads their main content and supporting materials.\n\nA typical skill folder may contain the following elements:\n\n- `SKILL.md`: The skill’s purpose, conditions for use, and work procedures\n- Scripts: Tasks to be executed deterministically, such as validation, conversion, and generation\n- Reference materials: API specifications, organizational policies, data structures, and domain knowledge\n- Templates and examples: The desired output format and quality standards\n- Evaluation materials: Test cases for comparing results before and after using the skill\n\nSkills are not model training that teaches an agent new foundational capabilities. They are closer to context assets that provide task knowledge and procedures at execution time.\n\n### Differences from Prompts, Rules, and MCP\n\n| Component | Primary role | When it is usually loaded | Points to note |\n|---|---|---|---|\n| General prompt | Specifies a single request and its desired result | When the user makes a request | May offer low reusability and consistency |\n| Persistent rules | Policies and behavioral constraints applied to every session | At session start or at all times | Continuously consume context and may conflict with one another |\n| Agent Skills | Provide procedures, materials, and scripts for specific tasks | When a relevant task is identified | Depend on routing accuracy and skill quality |\n| MCP | A standard interface for connecting to external tools and data | When a tool call is needed | Authentication, permissions, and external system security are important |\n\nWhile MCP primarily addresses “what can be connected to,” skills explain “which procedures and standards should be used” with connected tools and materials. Rather than competing, the two technologies are complementary and can be used together.\n\n## How Should the Top 5 Ranking Be Interpreted?\n\nThe provided materials ranked the following five candidates by the number of stars on their GitHub repositories. However, GitHub stars are closer to indicators of interest and bookmarking; they do not directly measure downloads, active users, or task success rates.\n\n| Provided rank | Repository | Type | Representative strength | Risk to verify |\n|---:|---|---|---|---|\n| 1 | `obra/superpowers` | A workflow collection that governs development procedures | Prioritizes requirement confirmation, planning, testing, and validation | Its procedures may be excessive for simple tasks |\n| 2 | `affaan-m/everything-claude-code` | A collection of configurations, commands, and agents for Claude Code | Broadly covers multiple stages of the development lifecycle | Installing everything may cause instruction conflicts and increase context usage |\n| 3 | `mattpocock/skills` | A skill collection focused on questions and design reviews | Provides an entry point for clarifying ideas before implementation | File-specific licenses and commercial-use conditions must be checked |\n| 4 | `multica-ai/andrej-karpathy-skills` | A third-party project that reformulates publicly shared development principles as skills | Emphasizes simplicity, minimal changes, and verifiable goals | Must distinguish whether the person named in the project officially created or endorsed it |\n| 5 | `anthropics/skills` | Anthropic’s official examples and document-production skills | Useful for examining skill structures and examples of output generation | Licenses for individual directories must be checked separately, even within the official repository |\n\nThe provided materials state that each repository has approximately 160,000–260,000 stars. To determine whether these figures refer to GitHub stars or installation totals from a particular registry, a GitHub API response or archived page from the same point in time would be required. Therefore, this article does not republish those numbers as verified current values.\n\n### Information Needed for a Reproducible Popularity Ranking\n\nTo create a reliable ranking, at least the following information should be disclosed together:\n\n1. Measurement time and time zone\n2. The repository’s exact owner, name, and commit SHA\n3. Original responses for metrics such as GitHub stars, forks, and contributor counts\n4. If installation counts are used, how duplicate installations and automated traffic are handled\n5. The aggregation unit distinguishing entire repositories from individual skills\n6. Rules for handling deleted or renamed repositories\n\nIn particular, if a single repository contains dozens of skills, repository stars alone cannot reveal which skills are popular. “Popular repositories” and “the most widely used individual skills” are different questions.\n\n## Design Philosophies of the Five Popular Candidates\n\n### 1. obra/superpowers: Enforcing Process Before Implementation\n\n`superpowers` has a strong workflow orientation designed to discourage agents from writing code immediately after receiving a request and instead require them to define the problem and develop a plan first. It centers on stages such as brainstorming, planning, test-driven development, debugging, and validation.\n\nThis approach is advantageous for tasks with unclear requirements or a high risk of change. By contrast, strictly applying the same process to clearly scoped tasks such as typo corrections may result in more questions and documentation than actual implementation.\n\n### 2. Everything Claude Code: Providing a Bundled Development Environment\n\nEverything Claude Code is a project that gathers various Claude Code configuration assets—including commands, agents, skills, hooks, and rules—in one place. It is useful for finding a broad range of examples spanning planning, implementation, review, testing, and record-keeping.\n\nHowever, “install everything” is not necessarily the best approach. Similar rules may be duplicated, or different workflows may conflict. In actual adoption, it is safer to select only the components the current team needs and verify when each item is activated.\n\n### 3. mattpocock/skills: Improving the Quality of Questions Rather Than Answers\n\nA distinctive feature of this collection is that it provides not only output templates but also questioning procedures that help users clarify their thinking. It encourages users to surface goals, assumptions, edge cases, and success criteria before implementation.\n\nThe fact that source code is publicly available does not mean every file can be used freely for commercial purposes. Repository-level licenses may differ from separate conditions specified for particular directories or files, so the current license must be checked before actual copying, modification, or distribution.\n\n### 4. andrej-karpathy-skills: Turning Concise Development Principles into Behavioral Rules\n\nAccording to the provided materials, this project was created by a third party that organized development principles publicly discussed by Andrej Karpathy into a skill format. Its central ideas include thinking before coding, choosing simple solutions, modifying only the requested scope with precision, and setting verifiable goals.\n\nThe important point is to distinguish the source from endorsement. Using a famous person’s name or reformulating their public statements does not mean that person directly created the repository or guarantees its results. The original statements and the reformulator’s interpretation must be evaluated separately.\n\n### 5. anthropics/skills: Examining Official Structures and Output Examples\n\nAnthropic’s official repository is a starting point for examining skill directory structures and examples of work involving outputs such as documents, spreadsheets, presentations, and PDFs. It allows comparisons of how materials and scripts needed for actual file generation tasks are organized, beyond simple behavioral rules.\n\nThe number of `SKILL.md` files stated in the provided materials may vary depending on whether translations, duplicates, examples, and branches are included. To compare scale, files must be counted at a specific commit, and the criteria for identifying canonical versions must also be disclosed.\n\n## Three Principles Repeated Across Leading Skills\n\n### They Constrain Behavior Before Expanding It\n\nGood skills do not indiscriminately expand what an agent can do. Instead, they control actions with a high probability of failure. Representative examples include not implementing before reaching agreement, not claiming completion without testing, and not modifying unrelated surrounding code that was not requested.\n\nThese constraints are not intended to eliminate agent autonomy. They are designed to reduce the cost of errors by placing checkpoints before actions that are difficult to reverse.\n\n### They Turn Expert Judgment Criteria into Procedures\n\nThe value of a skill lies less in sentence format than in decision-making criteria. By specifying the sequence in which experts confirm requirements, the way they handle uncertainty, and the standards they use to validate results, an agent can repeat similar reasoning procedures.\n\nHowever, imitating a particular person’s writing style is different from reproducing a proven working method. Evaluation cases and failure conditions should be examined rather than names or authority.\n\n### They Confirm “Why” Before “How”\n\nCode generation itself is becoming increasingly easy, but what should be built and what state should count as success are not determined automatically. This is also why the leading candidates focus on questions, planning, scope control, and validation.\n\nAdding sophisticated implementation procedures without clear goals can result in solving a poorly defined problem more quickly.\n\n## Focus on Context Structure Rather Than Token Counts\n\nThe provided materials include examples in which some bundles use approximately 17,000–22,000 tokens per session. However, token counts vary depending on the following conditions:\n\n- The model and tokenizer used\n- The scope of activated skills, rules, and hooks\n- Whether the client reads only metadata or injects the entire body\n- The length of the conversation history and project instructions\n- Whether caching and context compression are applied\n\nTherefore, a specific figure cannot be generalized as a fixed cost across all installation environments. When measuring, use an empty session as the baseline and add skills one at a time while comparing input tokens, response latency, and task success rates together.\n\nPractical ways to reduce context include the following:\n\n- Separate organizational policies that are always required from task-specific skills.\n- Include only the essential information needed for routing in skill descriptions.\n- Move lengthy specifications and examples into separate files so they are read only when necessary.\n- Consolidate overlapping rules into a single shared skill.\n- Disable skills that are rarely used or have not demonstrated performance improvements.\n\n## Security and Supply Chain Risks Overlooked by Popularity Rankings\n\nSkills may look like ordinary Markdown documents, but they can include shell commands, Python code, package installations, and external network requests. If an agent executes them, they create supply chain risks similar to those of ordinary software dependencies.\n\nBefore adopting an external skill, check the following:\n\n1. Verify the repository owner and whether it is an official project.\n2. Review installation scripts and executable files directly.\n3. Check the permissions required for environment variables, authentication tokens, the home directory, and network access.\n4. Pin the reviewed commit rather than using the latest state of a branch.\n5. If possible, run it first in a container or restricted test environment.\n6. Review the changes and any new permissions again before automatic updates.\n7. Confirm that the license permits copying, modification, internal use, and commercial distribution.\n\nEven a large number of GitHub stars does not eliminate the possibility of malicious changes, account compromise, or discontinued maintenance. Popularity is not a substitute for a security audit.\n\n## Criteria for Choosing Skills That Fit Your Environment\n\n| Evaluation item | Question to ask | Positive sign |\n|---|---|---|\n| Problem fit | Does it solve a real task that repeatedly fails? | Tasks where it applies and does not apply are clearly defined |\n| Routing | When should the skill be loaded? | Descriptions and triggers are specific |\n| Verifiability | Can quality before and after use be compared? | Test cases and success criteria are provided |\n| Context efficiency | Is there a large amount of content that must always be read? | Lengthy materials are loaded only when needed |\n| Safety | Does it require code execution or external access? | Uses minimal permissions and a clearly defined execution scope |\n| Maintenance | Can recent changes and issue handling be verified? | Change history and contribution procedures are public |\n| License | Is it compatible with the organization’s intended use? | File-specific conditions are clear |\n\nPopular repositories are best used as references for learning how to write skills, while production skills should be kept small and tailored to the organization’s codebase, review procedures, and tool permissions. Starting with a single recurring task and retaining only rules that pass evaluation is also advantageous for maintenance.\n\n## Conclusion\n\nAlthough the five provided Top 5 candidates take different forms, they all encourage agents to confirm requirements, plans, scope, and validation before rushing into implementation. The key to their popularity lies less in adding numerous commands than in turning failure-prone decision points into procedures.\n\nHowever, repository stars or registry installation counts alone cannot definitively identify the world’s most widely used individual skills. When citing rankings, the measurement time, original metrics, and aggregation units must be disclosed, while adoption decisions should be based on quality evaluation, context cost, licensing, and security review.","content_html":"\u003cp\u003eAgent skills are reusable work packages that enable AI agents to perform specific tasks consistently. They are broader in scope than simple prompts and can include not only instructions but also scripts, templates, examples, and reference materials.\u003c/p\u003e\n\u003cp\u003eThis article analyzes five popular candidates presented as of August 10, 2026. However, because there is no raw data or point-in-time archive that would allow the exact figures in the provided materials to be independently reproduced, they should not be interpreted as a definitive “global usage ranking.”\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#what-are-agent-skills\" class=\"anchor\" id=\"what-are-agent-skills\"\u003e\u003c/a\u003eWhat Are Agent Skills?\u003c/h2\u003e\n\u003cp\u003eThe core concept of Agent Skills, as described by Anthropic, is \u003cstrong\u003eprogressive disclosure\u003c/strong\u003e. Instead of reading every work instruction from the outset, an agent first reviews the names and descriptions of skills, then identifies only those relevant to the current task and loads their main content and supporting materials.\u003c/p\u003e\n\u003cp\u003eA typical skill folder may contain the following elements:\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e\n\u003ccode\u003eSKILL.md\u003c/code\u003e: The skill’s purpose, conditions for use, and work procedures\u003c/li\u003e\n\u003cli\u003eScripts: Tasks to be executed deterministically, such as validation, conversion, and generation\u003c/li\u003e\n\u003cli\u003eReference materials: API specifications, organizational policies, data structures, and domain knowledge\u003c/li\u003e\n\u003cli\u003eTemplates and examples: The desired output format and quality standards\u003c/li\u003e\n\u003cli\u003eEvaluation materials: Test cases for comparing results before and after using the skill\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eSkills are not model training that teaches an agent new foundational capabilities. They are closer to context assets that provide task knowledge and procedures at execution time.\u003c/p\u003e\n\u003ch3\u003e\n\u003ca href=\"#differences-from-prompts-rules-and-mcp\" class=\"anchor\" id=\"differences-from-prompts-rules-and-mcp\"\u003e\u003c/a\u003eDifferences from Prompts, Rules, and MCP\u003c/h3\u003e\n\u003cdiv class=\"overflow-x-auto\"\u003e\u003ctable\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth\u003eComponent\u003c/th\u003e\n\u003cth\u003ePrimary role\u003c/th\u003e\n\u003cth\u003eWhen it is usually loaded\u003c/th\u003e\n\u003cth\u003ePoints to note\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Component\"\u003eGeneral prompt\u003c/td\u003e\n\u003ctd data-label=\"Primary role\"\u003eSpecifies a single request and its desired result\u003c/td\u003e\n\u003ctd data-label=\"When it is usually loaded\"\u003eWhen the user makes a request\u003c/td\u003e\n\u003ctd data-label=\"Points to note\"\u003eMay offer low reusability and consistency\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Component\"\u003ePersistent rules\u003c/td\u003e\n\u003ctd data-label=\"Primary role\"\u003ePolicies and behavioral constraints applied to every session\u003c/td\u003e\n\u003ctd data-label=\"When it is usually loaded\"\u003eAt session start or at all times\u003c/td\u003e\n\u003ctd data-label=\"Points to note\"\u003eContinuously consume context and may conflict with one another\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Component\"\u003eAgent Skills\u003c/td\u003e\n\u003ctd data-label=\"Primary role\"\u003eProvide procedures, materials, and scripts for specific tasks\u003c/td\u003e\n\u003ctd data-label=\"When it is usually loaded\"\u003eWhen a relevant task is identified\u003c/td\u003e\n\u003ctd data-label=\"Points to note\"\u003eDepend on routing accuracy and skill quality\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Component\"\u003eMCP\u003c/td\u003e\n\u003ctd data-label=\"Primary role\"\u003eA standard interface for connecting to external tools and data\u003c/td\u003e\n\u003ctd data-label=\"When it is usually loaded\"\u003eWhen a tool call is needed\u003c/td\u003e\n\u003ctd data-label=\"Points to note\"\u003eAuthentication, permissions, and external system security are important\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\u003c/div\u003e\n\u003cp\u003eWhile MCP primarily addresses “what can be connected to,” skills explain “which procedures and standards should be used” with connected tools and materials. Rather than competing, the two technologies are complementary and can be used together.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#how-should-the-top-5-ranking-be-interpreted\" class=\"anchor\" id=\"how-should-the-top-5-ranking-be-interpreted\"\u003e\u003c/a\u003eHow Should the Top 5 Ranking Be Interpreted?\u003c/h2\u003e\n\u003cp\u003eThe provided materials ranked the following five candidates by the number of stars on their GitHub repositories. However, GitHub stars are closer to indicators of interest and bookmarking; they do not directly measure downloads, active users, or task success rates.\u003c/p\u003e\n\u003cdiv class=\"overflow-x-auto\"\u003e\u003ctable\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth\u003eProvided rank\u003c/th\u003e\n\u003cth\u003eRepository\u003c/th\u003e\n\u003cth\u003eType\u003c/th\u003e\n\u003cth\u003eRepresentative strength\u003c/th\u003e\n\u003cth\u003eRisk to verify\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Provided rank\"\u003e1\u003c/td\u003e\n\u003ctd data-label=\"Repository\"\u003e\u003ccode\u003eobra/superpowers\u003c/code\u003e\u003c/td\u003e\n\u003ctd data-label=\"Type\"\u003eA workflow collection that governs development procedures\u003c/td\u003e\n\u003ctd data-label=\"Representative strength\"\u003ePrioritizes requirement confirmation, planning, testing, and validation\u003c/td\u003e\n\u003ctd data-label=\"Risk to verify\"\u003eIts procedures may be excessive for simple tasks\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Provided rank\"\u003e2\u003c/td\u003e\n\u003ctd data-label=\"Repository\"\u003e\u003ccode\u003eaffaan-m/everything-claude-code\u003c/code\u003e\u003c/td\u003e\n\u003ctd data-label=\"Type\"\u003eA collection of configurations, commands, and agents for Claude Code\u003c/td\u003e\n\u003ctd data-label=\"Representative strength\"\u003eBroadly covers multiple stages of the development lifecycle\u003c/td\u003e\n\u003ctd data-label=\"Risk to verify\"\u003eInstalling everything may cause instruction conflicts and increase context usage\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Provided rank\"\u003e3\u003c/td\u003e\n\u003ctd data-label=\"Repository\"\u003e\u003ccode\u003emattpocock/skills\u003c/code\u003e\u003c/td\u003e\n\u003ctd data-label=\"Type\"\u003eA skill collection focused on questions and design reviews\u003c/td\u003e\n\u003ctd data-label=\"Representative strength\"\u003eProvides an entry point for clarifying ideas before implementation\u003c/td\u003e\n\u003ctd data-label=\"Risk to verify\"\u003eFile-specific licenses and commercial-use conditions must be checked\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Provided rank\"\u003e4\u003c/td\u003e\n\u003ctd data-label=\"Repository\"\u003e\u003ccode\u003emultica-ai/andrej-karpathy-skills\u003c/code\u003e\u003c/td\u003e\n\u003ctd data-label=\"Type\"\u003eA third-party project that reformulates publicly shared development principles as skills\u003c/td\u003e\n\u003ctd data-label=\"Representative strength\"\u003eEmphasizes simplicity, minimal changes, and verifiable goals\u003c/td\u003e\n\u003ctd data-label=\"Risk to verify\"\u003eMust distinguish whether the person named in the project officially created or endorsed it\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Provided rank\"\u003e5\u003c/td\u003e\n\u003ctd data-label=\"Repository\"\u003e\u003ccode\u003eanthropics/skills\u003c/code\u003e\u003c/td\u003e\n\u003ctd data-label=\"Type\"\u003eAnthropic’s official examples and document-production skills\u003c/td\u003e\n\u003ctd data-label=\"Representative strength\"\u003eUseful for examining skill structures and examples of output generation\u003c/td\u003e\n\u003ctd data-label=\"Risk to verify\"\u003eLicenses for individual directories must be checked separately, even within the official repository\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\u003c/div\u003e\n\u003cp\u003eThe provided materials state that each repository has approximately 160,000–260,000 stars. To determine whether these figures refer to GitHub stars or installation totals from a particular registry, a GitHub API response or archived page from the same point in time would be required. Therefore, this article does not republish those numbers as verified current values.\u003c/p\u003e\n\u003ch3\u003e\n\u003ca href=\"#information-needed-for-a-reproducible-popularity-ranking\" class=\"anchor\" id=\"information-needed-for-a-reproducible-popularity-ranking\"\u003e\u003c/a\u003eInformation Needed for a Reproducible Popularity Ranking\u003c/h3\u003e\n\u003cp\u003eTo create a reliable ranking, at least the following information should be disclosed together:\u003c/p\u003e\n\u003col\u003e\n\u003cli\u003eMeasurement time and time zone\u003c/li\u003e\n\u003cli\u003eThe repository’s exact owner, name, and commit SHA\u003c/li\u003e\n\u003cli\u003eOriginal responses for metrics such as GitHub stars, forks, and contributor counts\u003c/li\u003e\n\u003cli\u003eIf installation counts are used, how duplicate installations and automated traffic are handled\u003c/li\u003e\n\u003cli\u003eThe aggregation unit distinguishing entire repositories from individual skills\u003c/li\u003e\n\u003cli\u003eRules for handling deleted or renamed repositories\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eIn particular, if a single repository contains dozens of skills, repository stars alone cannot reveal which skills are popular. “Popular repositories” and “the most widely used individual skills” are different questions.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#design-philosophies-of-the-five-popular-candidates\" class=\"anchor\" id=\"design-philosophies-of-the-five-popular-candidates\"\u003e\u003c/a\u003eDesign Philosophies of the Five Popular Candidates\u003c/h2\u003e\n\u003ch3\u003e\n\u003ca href=\"#1-obrasuperpowers-enforcing-process-before-implementation\" class=\"anchor\" id=\"1-obrasuperpowers-enforcing-process-before-implementation\"\u003e\u003c/a\u003e1. obra/superpowers: Enforcing Process Before Implementation\u003c/h3\u003e\n\u003cp\u003e\u003ccode\u003esuperpowers\u003c/code\u003e has a strong workflow orientation designed to discourage agents from writing code immediately after receiving a request and instead require them to define the problem and develop a plan first. It centers on stages such as brainstorming, planning, test-driven development, debugging, and validation.\u003c/p\u003e\n\u003cp\u003eThis approach is advantageous for tasks with unclear requirements or a high risk of change. By contrast, strictly applying the same process to clearly scoped tasks such as typo corrections may result in more questions and documentation than actual implementation.\u003c/p\u003e\n\u003ch3\u003e\n\u003ca href=\"#2-everything-claude-code-providing-a-bundled-development-environment\" class=\"anchor\" id=\"2-everything-claude-code-providing-a-bundled-development-environment\"\u003e\u003c/a\u003e2. Everything Claude Code: Providing a Bundled Development Environment\u003c/h3\u003e\n\u003cp\u003eEverything Claude Code is a project that gathers various Claude Code configuration assets—including commands, agents, skills, hooks, and rules—in one place. It is useful for finding a broad range of examples spanning planning, implementation, review, testing, and record-keeping.\u003c/p\u003e\n\u003cp\u003eHowever, “install everything” is not necessarily the best approach. Similar rules may be duplicated, or different workflows may conflict. In actual adoption, it is safer to select only the components the current team needs and verify when each item is activated.\u003c/p\u003e\n\u003ch3\u003e\n\u003ca href=\"#3-mattpocockskills-improving-the-quality-of-questions-rather-than-answers\" class=\"anchor\" id=\"3-mattpocockskills-improving-the-quality-of-questions-rather-than-answers\"\u003e\u003c/a\u003e3. mattpocock/skills: Improving the Quality of Questions Rather Than Answers\u003c/h3\u003e\n\u003cp\u003eA distinctive feature of this collection is that it provides not only output templates but also questioning procedures that help users clarify their thinking. It encourages users to surface goals, assumptions, edge cases, and success criteria before implementation.\u003c/p\u003e\n\u003cp\u003eThe fact that source code is publicly available does not mean every file can be used freely for commercial purposes. Repository-level licenses may differ from separate conditions specified for particular directories or files, so the current license must be checked before actual copying, modification, or distribution.\u003c/p\u003e\n\u003ch3\u003e\n\u003ca href=\"#4-andrej-karpathy-skills-turning-concise-development-principles-into-behavioral-rules\" class=\"anchor\" id=\"4-andrej-karpathy-skills-turning-concise-development-principles-into-behavioral-rules\"\u003e\u003c/a\u003e4. andrej-karpathy-skills: Turning Concise Development Principles into Behavioral Rules\u003c/h3\u003e\n\u003cp\u003eAccording to the provided materials, this project was created by a third party that organized development principles publicly discussed by Andrej Karpathy into a skill format. Its central ideas include thinking before coding, choosing simple solutions, modifying only the requested scope with precision, and setting verifiable goals.\u003c/p\u003e\n\u003cp\u003eThe important point is to distinguish the source from endorsement. Using a famous person’s name or reformulating their public statements does not mean that person directly created the repository or guarantees its results. The original statements and the reformulator’s interpretation must be evaluated separately.\u003c/p\u003e\n\u003ch3\u003e\n\u003ca href=\"#5-anthropicsskills-examining-official-structures-and-output-examples\" class=\"anchor\" id=\"5-anthropicsskills-examining-official-structures-and-output-examples\"\u003e\u003c/a\u003e5. anthropics/skills: Examining Official Structures and Output Examples\u003c/h3\u003e\n\u003cp\u003eAnthropic’s official repository is a starting point for examining skill directory structures and examples of work involving outputs such as documents, spreadsheets, presentations, and PDFs. It allows comparisons of how materials and scripts needed for actual file generation tasks are organized, beyond simple behavioral rules.\u003c/p\u003e\n\u003cp\u003eThe number of \u003ccode\u003eSKILL.md\u003c/code\u003e files stated in the provided materials may vary depending on whether translations, duplicates, examples, and branches are included. To compare scale, files must be counted at a specific commit, and the criteria for identifying canonical versions must also be disclosed.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#three-principles-repeated-across-leading-skills\" class=\"anchor\" id=\"three-principles-repeated-across-leading-skills\"\u003e\u003c/a\u003eThree Principles Repeated Across Leading Skills\u003c/h2\u003e\n\u003ch3\u003e\n\u003ca href=\"#they-constrain-behavior-before-expanding-it\" class=\"anchor\" id=\"they-constrain-behavior-before-expanding-it\"\u003e\u003c/a\u003eThey Constrain Behavior Before Expanding It\u003c/h3\u003e\n\u003cp\u003eGood skills do not indiscriminately expand what an agent can do. Instead, they control actions with a high probability of failure. Representative examples include not implementing before reaching agreement, not claiming completion without testing, and not modifying unrelated surrounding code that was not requested.\u003c/p\u003e\n\u003cp\u003eThese constraints are not intended to eliminate agent autonomy. They are designed to reduce the cost of errors by placing checkpoints before actions that are difficult to reverse.\u003c/p\u003e\n\u003ch3\u003e\n\u003ca href=\"#they-turn-expert-judgment-criteria-into-procedures\" class=\"anchor\" id=\"they-turn-expert-judgment-criteria-into-procedures\"\u003e\u003c/a\u003eThey Turn Expert Judgment Criteria into Procedures\u003c/h3\u003e\n\u003cp\u003eThe value of a skill lies less in sentence format than in decision-making criteria. By specifying the sequence in which experts confirm requirements, the way they handle uncertainty, and the standards they use to validate results, an agent can repeat similar reasoning procedures.\u003c/p\u003e\n\u003cp\u003eHowever, imitating a particular person’s writing style is different from reproducing a proven working method. Evaluation cases and failure conditions should be examined rather than names or authority.\u003c/p\u003e\n\u003ch3\u003e\n\u003ca href=\"#they-confirm-why-before-how\" class=\"anchor\" id=\"they-confirm-why-before-how\"\u003e\u003c/a\u003eThey Confirm “Why” Before “How”\u003c/h3\u003e\n\u003cp\u003eCode generation itself is becoming increasingly easy, but what should be built and what state should count as success are not determined automatically. This is also why the leading candidates focus on questions, planning, scope control, and validation.\u003c/p\u003e\n\u003cp\u003eAdding sophisticated implementation procedures without clear goals can result in solving a poorly defined problem more quickly.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#focus-on-context-structure-rather-than-token-counts\" class=\"anchor\" id=\"focus-on-context-structure-rather-than-token-counts\"\u003e\u003c/a\u003eFocus on Context Structure Rather Than Token Counts\u003c/h2\u003e\n\u003cp\u003eThe provided materials include examples in which some bundles use approximately 17,000–22,000 tokens per session. However, token counts vary depending on the following conditions:\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eThe model and tokenizer used\u003c/li\u003e\n\u003cli\u003eThe scope of activated skills, rules, and hooks\u003c/li\u003e\n\u003cli\u003eWhether the client reads only metadata or injects the entire body\u003c/li\u003e\n\u003cli\u003eThe length of the conversation history and project instructions\u003c/li\u003e\n\u003cli\u003eWhether caching and context compression are applied\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eTherefore, a specific figure cannot be generalized as a fixed cost across all installation environments. When measuring, use an empty session as the baseline and add skills one at a time while comparing input tokens, response latency, and task success rates together.\u003c/p\u003e\n\u003cp\u003ePractical ways to reduce context include the following:\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eSeparate organizational policies that are always required from task-specific skills.\u003c/li\u003e\n\u003cli\u003eInclude only the essential information needed for routing in skill descriptions.\u003c/li\u003e\n\u003cli\u003eMove lengthy specifications and examples into separate files so they are read only when necessary.\u003c/li\u003e\n\u003cli\u003eConsolidate overlapping rules into a single shared skill.\u003c/li\u003e\n\u003cli\u003eDisable skills that are rarely used or have not demonstrated performance improvements.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch2\u003e\n\u003ca href=\"#security-and-supply-chain-risks-overlooked-by-popularity-rankings\" class=\"anchor\" id=\"security-and-supply-chain-risks-overlooked-by-popularity-rankings\"\u003e\u003c/a\u003eSecurity and Supply Chain Risks Overlooked by Popularity Rankings\u003c/h2\u003e\n\u003cp\u003eSkills may look like ordinary Markdown documents, but they can include shell commands, Python code, package installations, and external network requests. If an agent executes them, they create supply chain risks similar to those of ordinary software dependencies.\u003c/p\u003e\n\u003cp\u003eBefore adopting an external skill, check the following:\u003c/p\u003e\n\u003col\u003e\n\u003cli\u003eVerify the repository owner and whether it is an official project.\u003c/li\u003e\n\u003cli\u003eReview installation scripts and executable files directly.\u003c/li\u003e\n\u003cli\u003eCheck the permissions required for environment variables, authentication tokens, the home directory, and network access.\u003c/li\u003e\n\u003cli\u003ePin the reviewed commit rather than using the latest state of a branch.\u003c/li\u003e\n\u003cli\u003eIf possible, run it first in a container or restricted test environment.\u003c/li\u003e\n\u003cli\u003eReview the changes and any new permissions again before automatic updates.\u003c/li\u003e\n\u003cli\u003eConfirm that the license permits copying, modification, internal use, and commercial distribution.\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eEven a large number of GitHub stars does not eliminate the possibility of malicious changes, account compromise, or discontinued maintenance. Popularity is not a substitute for a security audit.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#criteria-for-choosing-skills-that-fit-your-environment\" class=\"anchor\" id=\"criteria-for-choosing-skills-that-fit-your-environment\"\u003e\u003c/a\u003eCriteria for Choosing Skills That Fit Your Environment\u003c/h2\u003e\n\u003cdiv class=\"overflow-x-auto\"\u003e\u003ctable\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth\u003eEvaluation item\u003c/th\u003e\n\u003cth\u003eQuestion to ask\u003c/th\u003e\n\u003cth\u003ePositive sign\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Evaluation item\"\u003eProblem fit\u003c/td\u003e\n\u003ctd data-label=\"Question to ask\"\u003eDoes it solve a real task that repeatedly fails?\u003c/td\u003e\n\u003ctd data-label=\"Positive sign\"\u003eTasks where it applies and does not apply are clearly defined\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Evaluation item\"\u003eRouting\u003c/td\u003e\n\u003ctd data-label=\"Question to ask\"\u003eWhen should the skill be loaded?\u003c/td\u003e\n\u003ctd data-label=\"Positive sign\"\u003eDescriptions and triggers are specific\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Evaluation item\"\u003eVerifiability\u003c/td\u003e\n\u003ctd data-label=\"Question to ask\"\u003eCan quality before and after use be compared?\u003c/td\u003e\n\u003ctd data-label=\"Positive sign\"\u003eTest cases and success criteria are provided\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Evaluation item\"\u003eContext efficiency\u003c/td\u003e\n\u003ctd data-label=\"Question to ask\"\u003eIs there a large amount of content that must always be read?\u003c/td\u003e\n\u003ctd data-label=\"Positive sign\"\u003eLengthy materials are loaded only when needed\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Evaluation item\"\u003eSafety\u003c/td\u003e\n\u003ctd data-label=\"Question to ask\"\u003eDoes it require code execution or external access?\u003c/td\u003e\n\u003ctd data-label=\"Positive sign\"\u003eUses minimal permissions and a clearly defined execution scope\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Evaluation item\"\u003eMaintenance\u003c/td\u003e\n\u003ctd data-label=\"Question to ask\"\u003eCan recent changes and issue handling be verified?\u003c/td\u003e\n\u003ctd data-label=\"Positive sign\"\u003eChange history and contribution procedures are public\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Evaluation item\"\u003eLicense\u003c/td\u003e\n\u003ctd data-label=\"Question to ask\"\u003eIs it compatible with the organization’s intended use?\u003c/td\u003e\n\u003ctd data-label=\"Positive sign\"\u003eFile-specific conditions are clear\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\u003c/div\u003e\n\u003cp\u003ePopular repositories are best used as references for learning how to write skills, while production skills should be kept small and tailored to the organization’s codebase, review procedures, and tool permissions. Starting with a single recurring task and retaining only rules that pass evaluation is also advantageous for maintenance.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#conclusion\" class=\"anchor\" id=\"conclusion\"\u003e\u003c/a\u003eConclusion\u003c/h2\u003e\n\u003cp\u003eAlthough the five provided Top 5 candidates take different forms, they all encourage agents to confirm requirements, plans, scope, and validation before rushing into implementation. The key to their popularity lies less in adding numerous commands than in turning failure-prone decision points into procedures.\u003c/p\u003e\n\u003cp\u003eHowever, repository stars or registry installation counts alone cannot definitively identify the world’s most widely used individual skills. When citing rankings, the measurement time, original metrics, and aggregation units must be disclosed, while adoption decisions should be based on quality evaluation, context cost, licensing, and security review.\u003c/p\u003e\n","tags":["AI Agents","Claude Code","Anthropic","Agent Skills","GitHub","Coding Agent"],"faqs":[{"question":"How are agent skills different from simple prompts?","answer":"A prompt usually expresses a single request, whereas an agent skill bundles the instructions, scripts, reference materials, examples, and evaluation criteria needed for recurring tasks into a folder. It can also be configured to load only the necessary content when a relevant task arises."},{"question":"Can the repository with the most GitHub Stars be considered the best skill?","answer":"No. GitHub Stars indicate interest and awareness, but they do not measure actual installations, active use, usage of individual skills, or task success rates. To assess quality, evaluation results, maintenance status, security, and licensing must also be considered."},{"question":"Is this Top 5 an official global ranking?","answer":"No. It is a summary of popular candidates at the repository level as of the specified date, not an officially recognized global usage ranking. Reproducing an accurate ranking would require the measurement time, original API responses, aggregation unit, and a snapshot of each repository."},{"question":"Can the popularity of an individual skill be determined from a repository's total Stars?","answer":"No. If a single repository contains multiple skills and configurations, there is no way to determine which item prompted a user to Star it. Statistics on installations or invocations of individual skills would need to be disclosed separately to compare popularity at the skill level."},{"question":"Are Agent Skills and MCP the same technology?","answer":"No. MCP provides an interface for agents to connect to external data and tools, while Agent Skills provide procedures and decision criteria for performing specific tasks. They can be used together, such as by including instructions in a skill on how to use MCP tools correctly."},{"question":"Does installing more skills make an agent more capable?","answer":"Not always. Activating even unused instructions increases context costs and can cause different rules to conflict. It is better to select only the necessary skills and then compare input tokens, latency, and task success rates."},{"question":"Are external skills safe because they are Markdown files?","answer":"They cannot be assumed to be safe. Skills may include shell commands, executable code, package installation, and instructions for external communication. Executable files and permissions should be reviewed, a verified commit should be pinned, and the skill should first be tested in a restricted environment."},{"question":"Can publicly available skills be freely used commercially?","answer":"Being publicly available is not the same as being open source. You must check not only the license for the entire repository but also any separate terms that apply to individual skills or directories. If the terms are unclear, it is safer to refrain from copying, modifying, redistributing, or using them in paid services."},{"question":"What is the most important requirement for a good agent skill?","answer":"It must clearly define the tasks to which it applies and does not apply, and its success criteria must be evaluable. It should also use minimal context and permissions and require the agent to confirm the requirements, scope, and validation method before acting hastily."}],"sources":[{"url":"https://github.com/obra/superpowers","title":"obra/superpowers GitHub repository","type":"source"},{"url":"https://github.com/affaan-m/everything-claude-code","title":"affaan-m/everything-claude-code GitHub repository","type":"source"},{"url":"https://github.com/mattpocock/skills","title":"mattpocock/skills GitHub repository","type":"source"},{"url":"https://github.com/anthropics/skills","title":"anthropics/skills GitHub repository","type":"source"},{"url":"https://skills.sh/","title":"skills.sh Agent Skills directory","type":"data_point"},{"url":"https://docs.github.com/en/repositories/viewing-activity-and-data-for-your-repository/about-stars","title":"GitHub Docs: About stars","type":"source"}],"images":[{"id":665,"url":"https://injoys.com/rails/active_storage/blobs/proxy/eyJfcmFpbHMiOnsiZGF0YSI6ODI3OSwicHVyIjoiYmxvYl9pZCJ9fQ==--b50bbcc847bc2d7b61ded00be971783aa85a29a4/ai-e0d0c4d7.webp","is_representative":true,"generation_method":"ai_image","license":"ai_generated","mime_type":"image/webp","translations":{"ko":{"alt":"별이 달린 5개 에이전트 스킬 카드와 돋보기, 평가 기준 아이콘이 놓인 저울","caption":"인기 순위와 세부 평가 기준을 함께 살펴 에이전트 스킬을 비교하는 모습을 나타낸다.","description":null},"en":{"alt":"Five starred agent skill cards on podiums, with a magnifier and a scale of evaluation icons","caption":"The illustration compares five agent skills by looking beyond popularity to detailed selection criteria.","description":null},"ja":{"alt":"星付きの5つのエージェントスキルカードと虫眼鏡、評価基準アイコンを載せた天秤","caption":"人気順位だけでなく、詳細な評価基準から5つのエージェントスキルを比較している。","description":null},"es":{"alt":"Cinco tarjetas de habilidades de agente con estrellas, una lupa y una balanza con criterios","caption":"La ilustración compara cinco habilidades de agente más allá de su popularidad mediante varios criterios.","description":null},"id":{"alt":"Lima kartu skill agen berbintang di podium, dengan kaca pembesar dan timbangan ikon penilaian","caption":"Ilustrasi ini membandingkan lima skill agen berdasarkan popularitas dan berbagai kriteria pemilihan.","description":null},"pt":{"alt":"Cinco cartões de habilidades de agentes com estrelas, lupa e balança com critérios de avaliação","caption":"A ilustração compara cinco habilidades de agentes além da popularidade, usando diferentes critérios.","description":null},"zh-hant":{"alt":"五張帶星號的代理技能卡、放大鏡，以及放有評估指標圖示的天平","caption":"插圖呈現不只看熱門排名，也依多項選擇標準比較五種代理技能。","description":null},"de":{"alt":"Fünf Agenten-Skill-Karten mit Sternen, Lupe und Waage mit Bewertungskriterien","caption":"Die Illustration vergleicht fünf Agenten-Skills anhand von Beliebtheit und weiteren Auswahlkriterien.","description":null}}},{"id":666,"url":"https://injoys.com/rails/active_storage/blobs/proxy/eyJfcmFpbHMiOnsiZGF0YSI6ODI4NSwicHVyIjoiYmxvYl9pZCJ9fQ==--baba657ae44c9c7abd0b42f0a8a282655bf5665f/ai-58954fc2.webp","is_representative":false,"generation_method":"ai_image","license":"ai_generated","mime_type":"image/webp","translations":{"ko":{"alt":"다섯 개의 에이전트 스킬 카드가 필터와 검증 단계를 거쳐 선택되는 흐름도","caption":"여러 에이전트 스킬을 기능, 보안, 실행 과정에 따라 비교하고 선별하는 구조를 보여준다.","description":null},"en":{"alt":"Flowchart of five agent skill cards passing through filtering and validation steps","caption":"The diagram shows agent skills being compared and selected by function, security, and workflow.","description":null},"ja":{"alt":"5つのエージェントスキルカードを絞り込み、検証して選ぶフロー図","caption":"複数のエージェントスキルを機能、セキュリティ、実行工程で比較・選別する流れを示している。","description":null},"es":{"alt":"Diagrama de cinco tarjetas de habilidades de agentes filtradas y validadas","caption":"El diagrama muestra la comparación y selección de habilidades según su función, seguridad y flujo de trabajo.","description":null},"id":{"alt":"Diagram lima kartu keterampilan agen yang melewati tahap penyaringan dan validasi","caption":"Diagram ini menunjukkan keterampilan agen dibandingkan dan dipilih berdasarkan fungsi, keamanan, dan alur kerja.","description":null},"pt":{"alt":"Fluxograma de cinco cartões de habilidades de agentes passando por filtros e validação","caption":"O diagrama mostra habilidades de agentes sendo comparadas e selecionadas por função, segurança e fluxo de trabalho.","description":null},"zh-hant":{"alt":"五張代理技能卡片經過篩選與驗證步驟的流程圖","caption":"圖中呈現依功能、安全性與執行流程比較並選擇代理技能的方式。","description":null},"de":{"alt":"Flussdiagramm mit fünf Agenten-Skill-Karten, die gefiltert und geprüft werden","caption":"Das Diagramm zeigt den Vergleich und die Auswahl von Agenten-Skills nach Funktion, Sicherheit und Ablauf.","description":null}}}],"published_at":"2026-08-16T04:15:57+09:00","updated_at":"2026-08-16T04:15:57+09:00","license":"cc_by","translation_status":"reviewed","available_locales":["ko","en","ja","es"],"data_locales":["ko","en","ja","es","id","pt","zh-hant","de"],"url":"https://injoys.com/en/articles/top-5-ai-agent-skills-comparison"}