devdoc83

Termaxa gates the shell commands AI coding agents run: allow/ask/deny, previews the blast radius, backs up before destructive ops.

The folder read-write config is the right primitive — that's what actually contains an agent when it goes off-script. And +1 on sandbox-exec being the daunting part; the macOS story is the hard bit.

Happy to answer questions about the Cloudflare Workers architecture or the Playwright integration.

This resonates. Built ZeroDrop in 2 days, spent 2 weeks on distribution and hit every wall — HN karma, Reddit ban, Stripe India restriction.

The building part was the easy part. The distribution not so much.

Happened to me too.. still new here, figuring out how HN works.. I though feel my effort was genuine and the product has real usage.. but such is life..

How do you handle the risk of stripping out the exact stack trace the agent needed? That seems like the hard tradeoff here.

Saw this play out firsthand this week. Launched a small developer tool and within 48 hours had traffic from 38 countries — Netherlands and Singapore near the top, which matches the bot-heavy regions in this data.

The SSL cert observation in another comment here is accurate too. The second a domain goes live it gets discovered.

Interesting idea — does this work in headless CI environments where there's no persistent browser storage?

This is exactly the problem I'm running into right now. Built something useful, getting organic npm installs, but turning that into actual distribution is a different skill entirely. Signed up.