“Claude Skills很棒，可能比 MCP 更重要”

Core Insights - Anthropic has launched Claude Skills, a new mode that allows its model to acquire new functionalities through the use of organized folders containing instructions, scripts, and resources [2][5][12] Summary by Sections Skills Overview - Skills are essentially Markdown files that instruct the model on how to perform specific tasks while allowing for additional documentation and pre-written scripts [4][5] - The new document generation feature of Claude is implemented through Skills, enabling the model to handle various file formats like .pdf, .docx, .xlsx, and .pptx [4][5] Functionality and Implementation - Claude can improve its task execution by loading relevant Skills only when necessary, which enhances efficiency [5][6] - At the start of a session, Claude scans all available Skill files and reads brief descriptions from the YAML front matter, minimizing token usage [6] Practical Application - An example of a Skill is the slack-gif-creator, which generates GIFs optimized for Slack, demonstrating the practical utility of Skills in real-world applications [7][10] - Skills are designed to be easily shared, with simpler Skills potentially implemented as single files and more complex ones as folders [21][24] Comparison with MCP - The Model Context Protocol (MCP) has shown limitations, particularly in token consumption, which can hinder the model's effectiveness [18][20] - Skills offer a more efficient alternative, allowing for task completion without the extensive token usage required by MCP [20][24] Future Potential - The potential for Skills is vast, with possibilities for creating a "data journalism agent" that can analyze and publish census data using just a folder of Markdown files and Python scripts [16][19] - Skills are expected to lead to a significant expansion in the ecosystem, surpassing the previous excitement surrounding MCP [24] Design Philosophy - The simplicity of Skills is a key advantage, allowing for straightforward implementation without the complexity of full protocols like MCP [25][27] - Skills focus on leveraging the model's capabilities to solve problems with minimal input, aligning with the essence of large models [27]