Leit-motif/claude-vantage-point
我为 Claude Code 做了一个 HUD:你的智能体正在等待的问题会显示在提示框上方,点一下就能回答
一个 macOS 上的 Claude Code 桌面端模组,将待处理的智能体问题展示在提示框上方,让用户一键回答、将决策排队,并跟踪会话活动。
关于这个 mod
Vantage Point 是 macOS 上 Claude Code 桌面应用的 HUD 模组。它在提示框上方放置一条横条,显示智能体问题和可点击选项,支持决策队列,为复杂选择提供按选项分列的图示,包含用于回复的 RSVP 速读器,并显示未解决问题的前沿。它还会跟踪会话文件编辑,将行变更与决策记录关联,显示子智能体活动,并包含提示缓存倒计时。
安装
安装方法请查看原始来源。
原文 / README
Nobody is reading every reply their agents write anymore. I'm not. But the question the session is waiting on is always buried somewhere in the response.
So I made Vantage Point, a mod for the Claude Code desktop app. A strip sits above the prompt. When a reply ends by asking me to choose, the options show up there with the recommended one in green, and one click sends my answer. When I can't just let a yes/no rip, I click the question and get a diagram for each option: why, how it would work, what could go wrong.
And when I do need the whole reply, there's a speed reader built on RSVP (rapid serial visual presentation). It plays the reply one word at a time, the agent's short gist first, so I can chew through it instead of rawdogging the transcript.
As many of you may have experienced, I'm the bottleneck. Agents need rulings from me more often than I'm at the keyboard, so their asks land in a queue: I hit yes, no or comment, the agent gets the ruling back and keeps going.
A question I'm constantly asking myself is "what's the next step?" That's why I built the frontier. One press shows every open issue sorted into what's waiting on me, what's blocked and what the agents pick up next, for one repo or all of them.
The whole point is to move as fast as possible, but I still want to know the blast radius of my decisions in real time. So the session banner shows every file the session edits, with lines added and removed, each tied to the decision record behind it. If Claude drifts from what I decided, I see it while it's happening.
The rest of the strip is for glancing: what each subagent is doing, and a countdown on the session's prompt cache so I know when stepping away is going to cost me.
Claude Code built the whole thing, and I set it up to validate its own work as it went: screenshotting the app, reading its own screenshots, restarting the desktop app to check the mod loaded and drew right, then fixing what it saw.
Like every Claude Code mod, it's completely free. Claude Code desktop app on macOS only for now, early 0.x preview, and cloud sessions on claude.ai don't draw mods. The video is a real session.
https://github.com/Leit-motif/claude-vantage-point
What do you actually look at while your agents run?
获取徽章
这是你的项目吗?把这个徽章放进 README 或网站,链接回本页。


