53 points by omrimaya 3 hours ago | 37 comments
- For those unfamiliar, “CVE” stands for “CV Enrichment”, common slang in Posture Engineering
- "'They' just don't want you to know how incredibly secure this vibeslopped clawshit is."
- Pretty impressive to introduce 1400 CVEs in a project that's only ~7 months old.
- These are CVEs in the base image and in standard lib dependencies. For example, just scanned an unhardened image I built today:
Unhardened: docker.io/nanoco/nanoclaw:agent-alpha 71 packages, 344 unique CVEs, linux/arm64 PACKAGE VERSION TYP C H M L N TOT ----------------------------------------------------------- expat 2.5.0 deb 0 4 18 1 2 25 curl 7.88.1 deb 4 4 6 0 7 21 hono 4.12.14 npm 0 1 18 2 0 21 libtiff 4.5.0 deb 0 2 1 1 15 20 perl 5.36.0 deb 5 6 3 0 3 17 pnpm 10.33.0 npm 0 8 7 0 0 15 glibc 2.36 deb 1 2 2 1 7 13 openjpeg 2.5.0 deb 0 0 3 1 9 13 cups 2.4.2 deb 0 2 8 0 1 11 glib2 2.74.6 deb 1 7 1 0 1 10 tar 1.34(+2) deb 1 1 7 0 1 10 llvm 15.0.6 deb 0 0 0 1 9 10 sqlite3 3.40.1 deb 1 2 3 0 3 9 nss 3.87.1 deb 1 0 3 0 4 8 avahi 0.8 deb 0 0 8 0 0 8 util-linux 2.38.1 deb 0 0 3 0 2 7 elf 0.188 deb 0 0 0 0 7 7 libssh2 1.10.0 deb 1 4 1 0 0 6 openldap 2.5.13 deb 0 1 0 0 5 6 chromium 151.0.7922.108 deb 0 5 0 0 0 5 ----------------------------------------------------------- UNIQUE CVEs 16 68 121 17 119 344 (+51 more packages, 102 findings) C/H/M/L/N = critical/high/medium/low/negligible. Counts are unique CVEs: binaries from one source package are grouped (libcurl4 + libcurl3-gnutls + curl = curl), so a CVE hitting three of them counts once, not three times.- Are these real findings, or a situation in which fixes have been backported? At one place I worked, the corpsec guys were wildly incompetent and would try to bury me in "CVEs" in my systems that were nothing but "vulnerable" software versions with all of the "identified" vulnerabilities fixed by Debian backported patches.
- Most modern scanners know the backported versions for the major vendors (ubuntu/debian/rhel/alpine), but it was definitely an issue a decade ago.
- If the thing measuring whether there are CVEs is also the thing creating said CVEs, are we sure they are even CVEs? Deduped? Etc.
- Hold my beer
- It’s like it’s made of CVEs. First 50-100 should be a good sign if it’s cleaner to start over.
- If you're not a security person, the unspoken subtext here: the overwhelming majority of these "CVEs" do not matter to the project, and a very large number of them don't matter at all. They're pro-forma findings, like ReDOS in code paths that are rarely used, or, even more commonly, "prototype pollution" issues.
- I'm convinced you can tackle 5-10 "CVEs" a day, make a little dashboard, put some pretty graphs on it, and send it to your exec team and probably get accolades. Nevermind that the CVEs had nothing to do with your product.
- My favorite urgent must fix CVE from compliance was a bug in the Linux PCMCIA driver on some EC2 VMs.
- I don’t say it’s the case there (it probably isn’t, you probably need hardware access or root), but sometimes those sorts of things do actually matter because there’s a way of causing them to be run anyway.
This is why it’s good to exclude things you don’t need. The less there is, the fewer places there are for problems to lurk.
- Why is that even part of the image?
- This is how Vanta et al. make millions.
- I don't think Vanta itself even scans for CVEs? They scan for compliance which includes tooling that may look for CVEs, like Dependabot.
Scanning companies should be sophisticated enough to distinguish invalid CVEs and back ported fixes.
- Why is the Node ecosystem like this? Why do people continue to choose it for popular projects vs. anything else?
- Because it's written in Javascript and there are classes of highly-situational and not-especially-meaningful vulnerabilities that broadly impact the entire ecosystem ("prototype pollution" is the canonical example) that get counted individually in every occurrence in every NPM package.
It's not a real number, and, worse, it obscures the real figure of merit (/demerit). The Node ecosystem does have a real problem with its culture of sprawling dependencies. But it's not the first-order issue with this "1400" number.
- It's the opposite of NIH syndrome. Need to left pad a string? Just import a library from some rando on the internet!
- When node was new there were many browsers (different rendering and js engines, not just chrome reskins) and standards and "proto standards" were moving very fast. Different browsers and versions would have very different support for CSS directives, tag behavior, etc. This was a real pain in the ass to make a site consistent across different browsers - every other line of code would require a full switch statement based on browser and version it seemed, and all of these things would need updating every time some browser had an update.
The answer to this was something called polyfills, a library that did something as simple as element.center() with just 500 lines of code to make it consistent on all browsers, and all the places you want to center that element are updated by the polyfill authors and your code doesn't need to be touched.
All the sites that weren't using good polyfills broke (or at least looked terrible) for days every time there was a new $browser update.
Since thats the javascript environment node was born into, the style was carried over by inertia and habit, for better and worse.
- Primarily because of the original sin of JS being an awfully designed language. Partially because it's the most popular.
- Because, like it or not, it does Write Once, Use Anywhere better than Java ever did.
It is pretty much the lowest common denominator for code.
- Because it's written by a type of dev that thinks everything should use JS because that's what they used for 6 months while working at a FAANG company before getting laid off and since they worked at a FAANG company clearly they know better than you so if you attack them for using an extremely poor language you just aren't a seriously person so please shut up while I go fishing for some VC funds.
It's like asking why Claude Code is written with react, because the devs getting paid >$500k a year clearly know better d'uh.
- What is NanoClaw? Glad you asked:
> NanoClaw is a secure, lightweight alternative to OpenClaw.
- I'm pretty skeptical you can call any claw-like thing secure unless you solve prompt injection.
- If the lightweight alternative has 1400 CVEs, how many does OpenClaw have?
- I don't understand the 'custom patch' strategy over 'fix the app with a major version change' strategy.
- I wonder how many new CVEs were introduced while patching these
- That's what happens when you vibe code.
- Why hasn't looking at EPSS (Exploit Prediction Scoring System) become a more standard approach than just raw CVEs?
- so s/bookworm/trixie/g didn't work then?
Yes, this is mostly a joke, I am able to understand the difference between base distros.
- a spot on "joke". deb12 is gone for a month now.
- let me guess. they wrote a promt that told claude do undo all bugs?