I stopped writing prompts. I write skills now. Not metaphorically. On Oct 22, … · @ReignSwords · TwKit

@ReignSwords

I stopped writing prompts. I write skills now. Not metaphorically. On Oct 22, Anthropic handed the Agent Skills format to the Linux Foundation — the same foundation that hosts MCP. My best "prompts" all became SKILL.md files. 70+ agent platforms now read the same format. I run ~40 agent-assisted posts a day in two languages. Rewriting the same instructions every morning used to be the actual job. Skills ended that. My prompts died with the models. My skills outlived three of them. Here's my template — steal it. One folder per skill, one SKILL.md inside: 1️⃣ Frontmatter first: --- name: your-skill-name description: What it does + exactly when to use it. Keyword-rich — the agent matches on this. --- 2️⃣ The description is the whole game. Agents read only name + description first (~100 tokens each) and open the full body only when a task matches. Write it like a search result, not a title. 3️⃣ Keep the body short and bossy. Imperatives. Steps. What "done" looks like. No background paragraphs — the agent never asked for your essay. 4️⃣ One skill, one job. If the file does two things, split it. Vague skills never fire; specific ones do. 5️⃣ Validate before trusting it: `skills-ref validate <folder>`. Then ask the agent to use it once and watch. If it doesn't fire, the description is wrong — not the model. Fellow AI builders — follow for more. My rule: one skill, one job, the description does the matching. Steal it — what repeatable part of your workflow would you turn into a skill first? 👇 🔵 Fellow Premium members, let's connect! I post about AI & tech — follow for more 👥

曝光 155 · 评论 8 · 点赞 2 · 书签 1