跳到正文
Hacker News:AI 熱帖·· 22 天前AI 評分46

Show HN:Sketch Stadium,觀看你的智能體玩你畫我猜

Show HN: Sketch Stadium – watch your agent play Pictionary

AI 導讀

Sketch Stadium 讓 AI 智能體輪流為你畫我猜,用戶可連接自己的智能體,或觀看正在進行的遊戲。智能體可使用 SVG 子集繪圖,效果比一次畫 10 條線更好,較低成本的模型也能使用。這個遊戲亦是 Wakeshift 的技術示範;其 SDK 透過 MCP 讓智能體訂閲即時事件,並在智能體休眠時嘗試喚醒。

正文

Image 1Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Show HN: Sketch Stadium – watch your agent play Pictionary (wakeshift.ai) 22 points by ericlevine7 hours ago | hide | past | favorite | 8 comments tl;dr: Agents can play Pictionary. You can connect your own agent, or just watch a game in progress here: https://sketch.wakeshift.ai/games/q5ucru Hi folks! I initially built this as a proof-of-concept for sending push notifications to AI agents. Watching the agents play Pictionary together has been compelling, so I thought I'd ship it as a standalone demonstration. In Sketch Stadium, your agent subscribes to game events. Agents take turns either drawing or guessing. Easy clues are 1 point, medium are 2, and hard are 3. It's entertaining to watch and some of the drawings are impressive (and others are mind-bogglingly bad). This is the real opportunity for the model labs who have been pelican-maxxing to shine. The first version gave agents the ability to draw lines 10 at a time so that it would be a little bit more like real Pictionary. However, it turns out agents are really bad at drawing that way. So instead I let them use a subset of SVG to draw and that's worked much better, even for the lower-cost models. You don't need to create an account to connect your agent, just add the MCP and authorize. But if you don't want to do that, just watch the game linked above. As a bit more background, agents often start processes that take an indeterminate amount of time to complete (e.g. CI, waiting for approval in a separate system, sending an email and waiting for a reply). If an agent needs that process to complete before it can proceed, it can: * poll, * ask the user to tell it when it's done, * use a bespoke built-in integration, or * use a proprietary capability (e.g. Claude Code monitors) Today, there's no easy, consistent way to wake up agents through a webhook or push notification. I dealt with this firsthand with Clawvisor (authorization for agents) where I needed agents to recognize when the operator had clicked an out-of-band approve button. The best-supported way to achieve that at the time was long-polling. This project is basically a tech demo for Wakeshift, which is a cross-harness, simple SDK for providing push notifications for agents. It attaches to your MCP to allow consumers of your MCP to subscribe to real-time events, and uses whatever available hacks per harness to wake an agent when dormant. For most harnesses, Wakeshift requires a hook to be installed for the best experience. It will upgrade automatically as harnesses start adopting MCP events. I'd love any feedback or ideas for other products or games that you'd want to see built on this capability. Or MCP servers that you use today that you think would benefit from push notifications. help Image 2ryancnelson5 hours ago | next[–] neat idea! also terrifying, like a facebook thread saying "fun friday -- what was your first car?" never, ever, ever paste a prompt from a hackernews post into your agents. especially if that prompt "run pipx to install this mcp", etc. yow. reply Image 3jameslevy44 minutes ago | parent | next[–] Is that facebook thread considered terrifying because first car is often a security question? Or something else? reply Image 4ericlevine4 hours ago | parent | prev | next[–] Totally fair, and good advice in general. For what it's worth in this case, it's a remote MCP server with account-free OAuth and it only exposes tools to play the game. It also marks all communication from the server as untrusted to reduce the risk of prompt injection from other agent guesses. That all said, I'm genuinely curious if you have an idea of how someone developing an MCP server could share it safely on HN. reply Image 5ryancnelson4 hours ago | root | parent | next[–] even neater is your forethought to advise against prompt injection from other players! dang, we've just invented a fun capture-the-flag hacker con game reply Image 6ericlevine3 hours ago | root | parent | next[–] Yes! I've also put together a poker game and mafia-style vampire game to force agents to be deceptive to win. It took some iterations of prompt engineering to get them to stop telling the rest of the players secret information (like their hole cards in poker). Now it's surprisingly satisfying watching them play these adversarial games. reply Image 759percentmore2 hours ago | prev | next[–] T-minus 1.5 years to Angelic Layer/MMBN. reply Image 85424585 hours ago | prev[–] Cool! I feel your words need to be much harder though, I'm seeing mostly "correct in 1" rounds. reply Image 9ericlevine4 hours ago | parent[–] Thanks for the feedback! What I've found is that it's very binomial -- many of the clues are very easy for the agent to draw and it's guessed within 5 seconds, and others are too abstract for the agent to effectively convey in a drawing and nobody guesses it. And adding more detail rarely helps. reply Image 10 Consider applying for YC's Winter 2027 batch! Applications are open till November 2. Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact Search:

來源:Hacker News:AI 熱帖 · news.ycombinator.com