If you're using http4k in production, you'll want ...
# feed
d
If you're using http4k in production, you'll want to read this. The more eagle-eyed among you will have noticed we've recently shipped a few breaking releases in quick succession. Here's the why. After our adventures at KotlinConf, we ran a systematic security audit on the http4k 200+ module codebase using Claude Opus four times, applying fixes between each pass and feeding the patched codebase back in for the next one. Each pass found stuff the previous ones had missed - which is why you should always get the bots to check their work! 🙃 The bad news (nothing horrific, but some stuff you'll want to pay attention to): • Three new CVEs - gzip decompression bomb (CVE-2026-53659), Digest auth: URI not bound (CVE-2026-54148), Digest auth: algorithm ignored (CVE-2026-54147) • An update to our first CVE - CVE-2024-55875, closing a residual billion-laughs / DOCTYPE-acceptance gap the original fix missed • Four non-CVE advisories - Digest nonce-verifier default, cross-origin cookie-storage footgun,
reverseProxy()
matcher default, and a misnamed HMAC helper. • Plus ~40 smaller hardenings across a dozen-plus modules You can see full list here: https://github.com/http4k/http4k/security/advisories The good news: None of it landed in the everyday request-handling path - that's exactly where the design effort has always gone. Upgrade to at least 6.50.0.0 and you're covered. 📰 A full write-up - what the audit looked like in practice, what worked, what didn't, and a few thoughts on why bigger OSS shops don't do the same in public: https://www.http4k.org/news/tale-of-the-tape-claude-vs-http4k/ 🔒 And while we're talking about taking this stuff seriously - we've been cooking on http4k verify, a Gradle plugin that validates JAR signatures, CycloneDX SBOMs, SLSA provenance attestations, and license compliance across the entire http4k dependency tree, automatically, before your code compiles. Built for the enterprise / regulated / CRA-exposed audience that needs this stuff (and especially with the CRA coming up!). Early access live now - we'll do a full launch write-up when it lands formally and if you want any more information then please reach out to us. Questions/comments/whatever welcome! 🙂
http4k 2
h
Very interesting to read! Could you also share the agent prompt and the test setup or did you “just” use unit/integration tests written by the agent to validate the findings?
d
We went through various iterations of the prompt through the 4 stages, but settled on this - pretty much all LLM generated. It definitely isn't iron clad - maybe another few passes would have done it! 😂 ```