<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>RAG on Joe Bollen Security</title><link>https://joe-b-security.github.io/tags/rag/</link><description>Recent content in RAG on Joe Bollen Security</description><generator>Hugo -- gohugo.io</generator><language>en</language><copyright>Yes, it's a real three-body problem simulation</copyright><lastBuildDate>Fri, 10 Apr 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://joe-b-security.github.io/tags/rag/index.xml" rel="self" type="application/rss+xml"/><item><title>Improving a Coding Agent Harness: Part 2.5, Securely Writing Code</title><link>https://joe-b-security.github.io/posts/2026-04-10-improving-coding-agent-harness-part2-5/</link><pubDate>Fri, 10 Apr 2026 00:00:00 +0000</pubDate><guid>https://joe-b-security.github.io/posts/2026-04-10-improving-coding-agent-harness-part2-5/</guid><description>&lt;p&gt;In &lt;a
href="https://joe-b-security.github.io/posts/2026-04-09-improving-coding-agent-harness-part2/"&gt;Part 2&lt;/a&gt;, I added an OODA loop with a rule engine and a verify phase so the harness could check the model&amp;rsquo;s output against correctness gates before accepting it. Correctness is only one half of what &amp;ldquo;good code&amp;rdquo; means, and this part extends the same loop to cover the other half, security, which needs a different set of inputs from the harness to work. I&amp;rsquo;m a big believer in security being a measure of quality.&lt;/p&gt;</description></item></channel></rss>