Agent Skill
Core Facts
According to the presenter's observations, programmers at large tech companies may spend 70%–80% of their AI usage time writing Skills, with relatively little time spent directly writing code.
Agent Skill adopts a lazy-loading mechanism whereby the Agent first reads only the metadata (name and description) of each Skill and loads the full instructions only after matching the user's need, so having multiple Skills coexisting does not increase Token consumption
The assets component is used to manage static resources such as images, audio, video, and other material files.
Timeline (last 90 days)
Recommendations for implementing enterprise-grade Agent Skills include: accumulate experience first, then automate; decouple tools from models; and build team capabilities progressively from shallow to deep.
The course covered as many as several dozen Skills on site, spanning scenarios such as development architecture and full-process automation.
Languages such as Java, Python and JavaScript can all be used to write Skills; there is no language restriction.
Spend most of your time developing Skills and let Skills generate the code, and code quality and maintainability will improve significantly.
The creator solidified the inspection process into a skill called "Plant Health Inquiry", after which a single sentence or a single click executes it according to the fixed process.
This workflow installs no Skills, executes no Skill code, and grants no permissions; it ultimately produces only one Skill record and one quota receipt for human review.
The Agentel team built a manual, deterministic n8n workflow for reviewing an Agent Skill's registered version, declared permissions, license and evidence support before adoption, and the workflow has been published in the n8n template library.
The Gini coefficient of the MCP server Star distribution is 0.867, and 0.882 on the Agent Skill side.
The MCPMarkets team scraped 5,945 MCP servers and 6,589 Agent Skill repositories on GitHub, about 92% coverage, totaling 12,534 deduplicated repositories.
The top 1% of MCP server repositories take 47.5% of server-side Stars and 41.7% on the Skill side; the top 10% together account for about 83%.
18 more timeline events
All Facts (20)
According to the presenter's observations, programmers at large tech companies may spend 70%–80% of their AI usage time writing Skills, with relatively little time spent directly writing code.
70%VerifiedAgent Skill adopts a lazy-loading mechanism whereby the Agent first reads only the metadata (name and description) of each Skill and loads the full instructions only after matching the user's need, so having multiple Skills coexisting does not increase Token consumption
70%VerifiedThe assets component is used to manage static resources such as images, audio, video, and other material files.
65%VerifiedThe scripts component allows Agents to call specific scripting tools when executing a Skill to assist in completing the task.
65%VerifiedAn Agent can have multiple Skills, with each Skill responsible for tasks in a specific domain.
65%VerifiedThe Agent Skill pattern has been widely applied in AI frameworks such as LangChain and AutoGPT.
65%Unverified会议总结Agent Skill输出包含会议摘要、核心要点和行动项三个固定模块
90%Unverified该会议总结Agent Skill内置了情感分析模块，能判断会议整体的情感倾向为积极、消极或中性
85%UnverifiedAgent Skill通过System Prompt工程、输出Schema约束和Few-shot示例注入来实现对LLM能力的专业化封装
75%Unverified现代Agent Skill体系可以分为三个核心层次：大脑决策层、技能封装层和执行引擎层
70%UnverifiedThe references component provides Agents with background knowledge and implementation details needed when executing tasks, such as API documentation and requirements documents.
60%UnverifiedSkills can be loaded and reused by different Agents, whereas prompts are typically one-time use.
60%UnverifiedThe course covered as many as several dozen Skills on site, spanning scenarios such as development architecture and full-process automation.
50%UnverifiedRecommendations for implementing enterprise-grade Agent Skills include: accumulate experience first, then automate; decouple tools from models; and build team capabilities progressively from shallow to deep.
50%UnverifiedSpend most of your time developing Skills and let Skills generate the code, and code quality and maintainability will improve significantly.
50%UnverifiedLanguages such as Java, Python and JavaScript can all be used to write Skills; there is no language restriction.
50%UnverifiedThis workflow installs no Skills, executes no Skill code, and grants no permissions; it ultimately produces only one Skill record and one quota receipt for human review.
50%UnverifiedThe Agentel team built a manual, deterministic n8n workflow for reviewing an Agent Skill's registered version, declared permissions, license and evidence support before adoption, and the workflow has been published in the n8n template library.
50%UnverifiedUnspecified licenses plus the NOASSERTION status cover 21.8% on the server side and as much as 30.4% on the Skill side.
50%UnverifiedOn the Skill side, 37.1% have fewer than 100 stars and 78.1% fewer than 1,000, with a median of 160.
50%