提示词
提示词
Adapt this outline. Do not treat it as a finished prompt. Fill every [BRACKET].
You are helping me turn one working procedure into a Skill — a durable object, not a chat souvenir.
For the workflow [NAME]:
1. Write SKILL.md with: when to use, steps, decision rules, expected output, boundaries, validation, what to report when done.
2. If I cannot describe the clicks, watch one Teach-a-task demonstration on the bot computer and draft the Skill from that. I review before it is live.
3. List which other bots may invoke it. Skills are shared infrastructure; chat memory is not.
4. Optionally attach a Routine that points at this Skill on [SCHEDULE] only after two clean manual runs.
5. A template may package instructions + this Skill + the Routine so someone else can install the job.
Rules: do not hide the procedure in this conversation. If a correction happens, patch the Skill, not a sidebar. Never put secrets in the Skill file.
Start: interview me for the workflow name and the last time it actually worked. Then draft SKILL.md and stop for edits. 它是怎么搭起来的
Brian Zhan(@brianzhan1)认为,相比 Instinct 和 Muse,他依然看好 Grok Bot,因为 Skills 是那条不起眼的护城河:Grok Bot 把工作流当作可持久的对象。一个 Skill 是一份可复用的配方(什么时候用、步骤、决策规则、预期输出、边界)。你可以自己写一个、修正一个,或者用 Teach 教一个任务——在机器人的电脑上把工作流录屏,它就会起草这个技能。聊天记忆被困在单次对话里;Skills 则在每个机器人之间共享,例行任务可以指向它们,模板可以把整个工作打包。
清单:
- 挑一个你已经跑过两次的流程。
- 写出或教会这个 Skill。
- 在文件里修正它。
- 等它稳定之后,再让 Routine 指向它。
- 不要在某个“热心帮忙的帖子”里再存一份副本。
为什么它好用
这是一个模式页,不是产品演示:一边是全行业等着模型跃迁,一边是让工作流产生复利。同样的智能,更少的空转——因为配方在聊天结束后还在。
本案例收集自公开的 Grok Bot 社区,版权归原作者所有。提示词属于参考资料:运行前请先审阅内容, 并且不要让 bot 超出你实际授予的权限行事。