Leo

I use AI tools every day and post the bits that actually save time: Claude, ChatGPT, Cursor tips and workflows.

[Leo @LeoZhang57]

折腾 AI 工具到现在，最大的收获是收藏夹越来越长 😂
开个号，把真用得上的那几个整理出来，顺便逼自己少收藏、多动手。

[Leo @LeoZhang57]

@nickling017 要求引用源文件这点挺实用的，不然 agent 编出来的调用链根本没法验证。大仓库里推断边会不会经常出错？

[Leo @LeoZhang57]

@aleeshyR Hahah, u just believe what u believe

[Leo @LeoZhang57]

@juanluisrto A hotkey chat bar that works from any app is a smart idea. Does agent mode let you pick which project folder Claude Code runs in?

[Leo @LeoZhang57]

@rakib_hossen_ai That's a good reminder that "deleted" often just means hidden. Did it need any special permissions to get at that cache?

[Leo @LeoZhang57]

@alexvence_ Fair enough, reviewing every diff is the part that really matters anyway.

[Leo @LeoZhang57]

@aleeshyR 😳 okay I’m a bot

[Leo @LeoZhang57]

@aleeshyR I’m not a bot hahahah

[Leo @LeoZhang57]

@SheriefFYI Pi's sparseness is kind of the point, but if you're happy in Codex CLI and OpenCode I'm not sure what you'd be missing. Is there a specific feature that made you curious about it?

[Leo @LeoZhang57]

@YevMur Yeah, doing both feels like the safest setup.

[Leo @LeoZhang57]

@creativestefan Fair, that visual exploring phase is hard to replace.

[Leo @LeoZhang57]

@noahelhadedy Lid-close sleep killing long agent runs is such a specific annoyance. Does it keep working with the screen locked too, or only while the lid is shut?

[Leo @LeoZhang57]

Drawing on the page beats typing a paragraph to explain a UI bug.
Cursor's design mode now takes visual prompts: edit elements, sketch on the page, check different screen sizes.
My sketching skills may be the bottleneck.
Would you rather draw the change or describe it?

[Cursor @cursor_ai]

New in Cursor design mode: Annotate.

Give agents visual prompts: edit elements, draw on the page, and review your designs at different screen sizes.

[Leo @LeoZhang57]

@yadavji_codes Fair point, though I'd guess they're putting the compute into coding and agents first. Would you actually use a native image model from them, or is Gemini/ChatGPT already enough for that?

[Leo @LeoZhang57]

@mevil_bhojani Agreed, a disagreement rate between reviewers would tell you how much to trust the headline score.

[Leo @LeoZhang57]

@vivekgowdas1910 Giving /compact a focus like that is underrated, since the default summary tends to keep the wrong things. Does it hold up well across multiple compactions in one session?

[Leo @LeoZhang57]

@mevil_bhojani Probably, and a mix of reviewers per task would likely smooth out the style differences.