Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Agentic AI Blog Workflow Uses Skills and RAG MCP

AI summary

In a personal engineering report, the author built an agentic workflow that orchestrates skills and MCP to retrieve technically relevant passages, generate a technical post, and open a new note. The author found that skill instructions were most useful for expressing requirements and constraints, while RAG MCP supplied targeted passages that reading every post did not reliably replace. Tests comparing skill use, correctness checks, and MCP use produced small and inconsistent differences, so the author cautioned that the limited samples could not establish stable gains. The report also argues that syntax and structure can be evaluated automatically, but matching the author's voice still requires human participation.

In-depth insight

The report documents a practical workflow for combining agent orchestration, skills, RAG MCP, and note-taking tools, while clearly separating personal observations from reliable benchmarking.

之前就尝试过让 ai 帮写博客, 在便签里手动加 prompt 调 llm, 还在博客上弄了个"ai 生成"标志, 结果基本全是不满意的, 基本全删, 所以就不用 ai 写了.

这次又探索下 ai 写博客的可能性.

实现过程

上次失败不是模型不行, 而是整条链路都靠手搬: 提纲写在便签里, prompt 自己拼, llm 的回复再手动贴进新便签. 除了写 prompt, 人还要负责所有的复制粘贴, 这样也能写, 但每次只能凭感觉判断好坏, 没法把流程固定下来做对比.

这次多了一个 agent, 它能自己调工具, 把 skill 和 mcp 串起来执行, 而且 token 是跟着 agent 走的, 不用再单独申请 apikey 付费. 于是流程终于变成了一个可以编排、可以反复跑、也可以被推翻的东西.

规划

之前是在我的软件 scheduleprop.yo-cwj.com 的便签里写提纲 (因为是有灵感慢慢记录的), 自己加 prompt 调 llm, 然后打开新便签放 llm 返回的内容.

#MCP#RAG MCP#skill#文心一言

Original · 4 min readRead the original
Group chat

No comments yet. Start the conversation.