正在加载视频...

视频加载失败

Finally rolling out the full Agent Session Explorer OSS next week 👀 Visualize every coding agent trace Identify root cause errors Feed it back to your coding agents to improve with every agent trace

35,326 次观看 • 17 天前 •via X (Twitter)

20 条评论

Eshwar Prasad Yaddanapudi 的头像
Eshwar Prasad Yaddanapudi16 天前

@VaraadDurgaay

maxi.me 的头像
maxi.me16 天前

I’d love to see this applied to browser and workflow agents too. Anthropic’s latest report shows why. When an agent submits a real form or bypasses an access restriction, understanding the exact sequence of actions becomes pretty important. Especially when the final output looks perfectly normal

我与梦想有个约会 的头像
我与梦想有个约会16 天前

cool

Aranđel 的头像
Aranđel16 天前

events box, fix height 🙏

Simon Loewen 的头像
Simon Loewen17 天前

the feed it back to your coding agents part is what I'm curious about. when an agent reads its own trace, does it go after the real root cause or just blame whatever step threw the error last? that's where mine usually get it wrong, I think. will look for the OSS drop next week.

Aiswarya Sankar 的头像
Aiswarya Sankar16 天前

We cross check against pr review errors

Chen 的头像
Chen16 天前

feeding agent traces back to fix errors is such a smart closed loop. excited to see the oss drop next week.

Vera Ai | Tools & Updates 的头像
Vera Ai | Tools & Updates16 天前

Open-source visibility into agent traces could make debugging so much easier

Kevin Rajan 的头像
Kevin Rajan16 天前

bruh...pls anchor the heights of the rows in the last column im about to have a seizure 😭😂 overall looks super cool tho!

Aiswarya Sankar 的头像
Aiswarya Sankar16 天前

😅 ok ok

CryptoFro 的头像
CryptoFro16 天前

This is the interface very frontier labs has to peek into everyone’s chat

Sujal 的头像
Sujal17 天前

feeding traces back to the agent closes the loop most teams leave open

Edgar Arakelyan 的头像
Edgar Arakelyan16 天前

Would be nice to have something like this for every agentenv agent It seems possible to support through our usage of a2a

BitAres 的头像
BitAres16 天前

Can I jump from a failed tool call straight to its inputs and result?

Pavel Kerbel 的头像
Pavel Kerbel16 天前

How will feedback move from a trace into lasting repo context? I'd keep the trace as evidence and propose a narrow lesson for review, ideally with a regression test, so one failed run doesn't become a rule every later agent inherits.

Aiswarya Sankar 的头像
Aiswarya Sankar16 天前

Sessions are mined in mass and in doing so several repeat patterns show up

Benedikt Sanftl 的头像
Benedikt Sanftl16 天前

we already closed the loop, hillclimbing to the max

D 的头像
D16 天前

Looks amazing, well done!

Caelan C 的头像
Caelan C16 天前

Crazy

Zan 的头像
Zan16 天前

These look cool but are useless for humans to understand like so many of these tools that pop up with millions of data points.

相关视频