正在加载视频...

视频加载失败

Vlog 🎬 I tried vibe coding but failed... Claude Code kept doing wrong things in this session. Maybe I should improve my prompting skill 😂 Please share your advice or thoughts in the comments.

19,947 次观看 • 1 年前 •via X (Twitter)

11 条评论

Takuya 🐾 devaslife 的头像
Takuya 🐾 devaslife1 年前

Watch on YouTube:

Konrad Zapałowicz 的头像
Konrad Zapałowicz1 年前

I've found this article to be very useful: Also, this oauth library by cloudflare has been entirely written by ai As far as I remember the commit messages include prompts so this is something you might want to read through too

Takuya 🐾 devaslife 的头像
Takuya 🐾 devaslife1 年前

Interesting!! Thanks for sharing

Takuya 🐾 devaslife 的头像
Takuya 🐾 devaslife1 年前

this article looks helpful

Matthew Barlow 的头像
Matthew Barlow1 年前

I feel your pain, especially with the agent not using built-in bun features. Try providing the agent with Bun's llms.txt file As for the "pwd" "ls" etc. errors I don't know why this happens but it usually recovers. I have the agent create a plan, work on just the first step while I do something else, ring a "terminal bell" when it is finished (notification settings in Claude Code), then I review its work, iterate until I like it, git commit, "/clear" the context, and kick off the next task.

Takuya 🐾 devaslife 的头像
Takuya 🐾 devaslife1 年前

thanks! i forgot that Bun has llms.txt😂👍 I'll try it. the notification settings look nice.

Iggy 的头像
Iggy1 年前

Here are some articles from devs who are jumping into using it, including a pdf from @AnthropicAI and some of their videos: 1. 2. 3. 4. 5.

boisselle 的头像
boisselle1 年前

haha. the LLMs love adding things, so you have to specifically tell them to ONLY do what you're asking it to do 🤣

Bryan Randal Arendt 的头像
Bryan Randal Arendt1 年前

1. read everything you can that anthropic has published in their docs: 2: start small. do a small green field project. use a planning prompt. do not turn on auto accept until you have held its hand for a while. say no often and correct what it's doing. once it's on the right track then let it go an auto accept or even --dangerously-skip-permissions 3: connect it to gh cli tool and have it branch, commit and push changes 4: playwright MCP is the only MCP you need. have it tdd or at least verify/validate it self via the cli 5: read again 6: understand claude code's philosophy. raw modal access 7: find and watch all the @bcherny interviews you can

Jake Manger 的头像
Jake Manger1 年前

Try Cursor! There's a reason everyone's using it. Claude 4 on there is amazing

Takuya 🐾 devaslife 的头像
Takuya 🐾 devaslife1 年前

yep but I don't want to leave nvim😂😂

相关视频