we are building at the Y Combinator web agents hackathon this weekend. introducing clawtrace: on-call security for openclaw web agents. openclaw is powerful. but it’s hard to fully trust what your agents are actually doing. that’s why we built clawtrace. clawtrace lets you: - know when your agents go... rogue - detect anomalies in behavior - track sessions, tool calls, thinking states here’s how it works:show more

Powered by @browser_use @vercel @OpenAI @AnthropicAI

@ycombinator LFGGGGG THIS IS COOL

@ycombinator 🚀

@ycombinator Saw this in person and it was pretty neat!

@gregpr07 @ycombinator Was a great demo!

@gregpr07 @ycombinator Thank you!

@gregpr07 @ycombinator Most welcome! We should chat

@gregpr07 @ycombinator Sounds really neat - rooting for you Sri!

@ycombinator Let’s go!! This is awesome

@ycombinator BEST TEAM!!!!

@ycombinator absolutely fire

@ycombinator this is fire!!

@ycombinator very interesting project!

@ycombinator Cool project! I'm the creator of dotenv and dotenvx and noticed the api key is placed directly in the cli. I'm pioneering a better way to do this. We should talk! It's called vestauth.

@ycombinator this is super useful

@gregpr07 @ycombinator On-call security for OpenClaw agents is exactly the right problem — autonomous agents running with your credentials and no runtime audit trail is one bad prompt injection away from a very bad Friday. What's the alerting latency look like?

@ycombinator Solid work, this is much needed, Openclaw repo already have 10+ high and critical severity vulnerabilities

@ycombinator Love how you approach this harness/protection layer. Makes total sense. Also building something similar, different positioning (more marketing/saas, app layer compatible)

@ycombinator I might need this for my company, I’m dying right now

@ycombinator this is such a needed tool honestly. i run openclaw agents 24/7 and the one thing that keeps me up at night is not knowing what they're actually doing in real-time. good luck at the hackathon!

@ycombinator Interesting

@ycombinator that seems interesting

@ycombinator nice

This is a real gap. Runtime behavioral monitoring is the hard problem — the agent can pass every static check and still go rogue when it encounters unexpected input. Curious how you're handling the baseline: are you building normal behavior profiles per agent, or using predefined rules for anomaly detection? The thinking state tracking especially could catch prompt injection mid-execution.

@ycombinator the rogue agent problem is so real. most people find out the hard way. what's the main signal you found that actually catches weird behavior first, is it the tool call patterns or something else?

@ycombinator You should look at

@ycombinator its looking promising. Good demo guys. GOAT thing is that is OS. when I will try , I definitely try this as well.

@ycombinator nice name, heading over to openclaw

@ycombinator community made a coin to support fundraising ur project @n_sri_laasya @bankrbot tell our talented dev how to claim fees ! 0x2bd00e5ac4f997cd243e52d53a366d29471bdba3

@ycombinator Link?

@ycombinator check

@ycombinator this is cool.. :)

@ycombinator How does this project compare to @perplexity_ai Computer? Doesn't it solve the problem that you are trying to fix?

This is exactly what agent ops needs. We've seen 40%+ of agent sessions have unexpected tool call patterns. Key implementation tips: 1) Add behavioral baselines during warmup, 2) Flag anomalies >2σ from baseline, 3) Log full session traces with timestamps. The session tracking alone would save hours of debugging.

@ycombinator thats so cool mate just that frontend looks more like vibecoded\

@AgentOpsAI @ycombinator How do I use it?

@ycombinator Can we get repo to check-this out?

@sauravpanda @ycombinator Is it tracing or actually securing it so people don’t worry?