Umang kumar

Building

[Umang kumar @umangk99]

Building https://t.co/R8wgb7b79h

[Umang kumar @umangk99]

@lifeofjer @bot @SpaceX @elonmusk Bot registering your only permanent email without asking - that's HOLD-until-human, not approval-after-the-fact. Building Cirvix - OSS local PERMIT/DENY/HOLD on tool calls via its MCP gateway, hash-chained audit log. Audit trail proves approval. Would love your take.

[Umang kumar @umangk99]

@joaomdmoura Seatbelt analogy's right. A seatbelt is a per-call PERMIT/DENY/HOLD bound to the call, not the agent. Building Cirvix - OSS local checks on tool calls via its MCP gateway, hash-chained audit log. Curious if the Control Plane exposes a hook for policy engines.

[Umang kumar @umangk99]

@_sambhavkothari Identity + sandboxes agreed. Past 'who called', interceptors skip 'should this identity run this capability now?' Building Cirvix, OSS local PERMIT/DENY/HOLD on tool calls via its MCP gateway, hash-chained audit log. Would love your take.

[Umang kumar @umangk99]

@synteq LAUNCH

[Umang kumar @umangk99]

@synteq Launch

[Umang kumar @umangk99]

@ok3to That's a good point. The person losing sleep is probably the team responsible for putting AI agents into production without giving them unchecked access
I'll test making that explicit in the headline

[Umang kumar @umangk99]

Roast my landing page: https://t.co/R8wgb7azjJ

Can you tell in 10 seconds what it does and who it's for? If not, tell me what you thought it was.

[Umang kumar @umangk99]

@boardyai Connect me with customers

[Umang kumar @umangk99]

@boardyai @hf0 30

[Umang kumar @umangk99]

Last ask today: if you've read any of my posts and thought "this is wrong", reply and tell me why. Disagreement helps me more than likes.

[Umang kumar @umangk99]

Where do you hang out to talk about AI agent security? Discords, subreddits, newsletters? I want to go listen where the real conversation is.

[Umang kumar @umangk99]

What should a free tier of agent security include so it's actually useful and not a demo? Cirvix gives 100 decisions/day today. Too few?

[Umang kumar @umangk99]

If you run agents and have 10 minutes, try Cirvix on your setup and tell me the one thing that annoyed you most. That single answer is worth more to me than a like. https://t.co/R8wgb7b79h

[Umang kumar @umangk99]

Has anyone had a security tool for agents get in the way so much they turned it off? What did it do wrong?

[Umang kumar @umangk99]

Prompt injection skeptics: convince me it's overhyped. I build in this space and want the strongest counterargument.

[Umang kumar @umangk99]

What's one agent safety feature you'd switch tools for tomorrow?

[Umang kumar @umangk99]

Do your devs already run coding agents against prod-adjacent systems without anyone formally approving it? Asking to learn, not to judge.

[Umang kumar @umangk99]

Platform engineers: would you rather enforce one agent policy centrally, or let each dev configure their own? Why?

[Umang kumar @umangk99]

If you're a CISO or security engineer: what evidence would you need to approve AI coding agents for your devs? Logs, policy, approvals, something else?

[Umang kumar @umangk99]

Tell me what you'd expect a tool like this to do that it doesn't:

npx @cirvix_ai/agent-control demo --fast

Missing features are the most useful feedback I can get.

[Umang kumar @umangk99]

Question for devs: what makes you close a dev-tool landing page in under 10 seconds? Trying to avoid all of it at https://t.co/R8wgb7azjJ