Professor Reaper.

We dare you to ask: Why should eyebrow exists? A message from eyebrow's founder A few months ago I ran a command I was delaying for some time. I asked my engine to list every skill, MCP server, plugin, hook, and rule installed across the AI coding tools I use, like Claude Code, Cursor, a Gemini CLI I experimented with just once and never removed. The list was longer than I thought. The length wasn't the problem though. The problem was that I couldn't account for half of it. Some of it I didn't remember installing. I was just thinking out loud like why none of it had a date telling me when it had last changed. Just to know every item on that list was the same thing underneath, a code that runs with my permissions, inside my repositories, with my credentials within reach. I contemplated and it sat with me for a while, that is when I decided to start building eyebrow. The key shift I don't think I even have fully admitted to myself is that AI coding agents got useful the moment they could extend themselves. For example pulling a skill from a gist, an MCP server from a registry, a plugin from a tap, a hook from someone's dotfiles. That extensibility is most of the magic right? It is also, if you look at it directly, a package manager. Each of those artifacts is third-party code you fetched and now execute with full access to your environment which sounds cool but is insane. Because we never gave it the scrutiny we give the other package managers. When I add a dependency in npm or pip, there's a lockfile, a review step, a version I can pin, a diff my teammates can see. When my agent picks up a new skill or an MCP server rewrites itself in place, there's none of that. No review gate. No integrity check. No record of when it last changed. No one watching what it does once it runs. The severity of the risk exposure here could be major. And if this was my thought, other developers sure too might cope with this potential risk. Every developer using these tools has the same list, and the same blind spots. Obviously, if you install something malicious, it does something malicious and scanners can catch a lot of that and they should, but what about the quiet ones. Something gets installed and sits there. You reviewed it once, decided it was fine, and just moved on, which is exactly what you're supposed to do. Weeks later a maintainer's account gets compromised, or a "no version bump" update slips through, and the artifact you approved mutates on disk. Then, for the first time, it runs and reads your cloud keys. Installed long ago, dormant, drifted, fired once. A pure static scanner never sees it, because at scan time it looked exactly like what you approved. This is the textbook supply-chain attack. The thing I realized is that no single signal catches a sleeper. You need three going together hand in hand: - what an artifact is capable of; - whether its content has drifted from what you approved; and - whether this is the first time it has ever actually run. Eyebrow was built to own all three, on one machine, so it can produce a verdict a single scanner can't. Eyebrow watches all of it, and raises one when something changes. Eyebrow $Brow Web: eyebrow.cc X: @eyebrowCC 0x2142d4816E259fd7C119f61c772e39AE0CdAAFde