<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Ratchet Code on Architect PARK JUN WOO</title><link>https://www.parkjunwoo.com/tags/ratchet-code/</link><description>Recent content in Ratchet Code on Architect PARK JUN WOO</description><generator>Hugo</generator><language>en</language><lastBuildDate>Sat, 16 May 2026 14:00:00 +0900</lastBuildDate><atom:link href="https://www.parkjunwoo.com/tags/ratchet-code/index.xml" rel="self" type="application/rss+xml"/><item><title>Why Coding Agents Work and Why They Break</title><link>https://www.parkjunwoo.com/opinion/why-agents-work-and-break/</link><pubDate>Sat, 16 May 2026 14:00:00 +0900</pubDate><guid>https://www.parkjunwoo.com/opinion/why-agents-work-and-break/</guid><description>The same model hallucinates in web chat yet ships 200-line features in a coding agent. It is not the model that changed &amp;ndash; it is the topology. Generation can be probabilistic. Verification must be deterministic.</description></item><item><title>Ratchet Pattern — How to Make an Agent Finish the Job</title><link>https://www.parkjunwoo.com/tech/ratchet-pattern/</link><pubDate>Fri, 15 May 2026 14:00:00 +0900</pubDate><guid>https://www.parkjunwoo.com/tech/ratchet-pattern/</guid><description>I asked an AI agent to write tests for 527 functions. It stopped at 40 and declared &amp;lsquo;done.&amp;rsquo; The Ratchet Pattern forces completion by delegating the done/not-done decision to a mechanical verifier — so the agent keeps going until the machine says stop.</description></item><item><title>Feedback Topology Over Model IQ</title><link>https://www.parkjunwoo.com/opinion/feedback-topology/</link><pubDate>Thu, 14 May 2026 18:00:00 +0900</pubDate><guid>https://www.parkjunwoo.com/opinion/feedback-topology/</guid><description>The same model stalls at 40 or completes all 527. The difference is not the model — it is the feedback structure. LLM performance depends far more on how fast and deterministic the feedback loop is than on the model itself.</description></item><item><title>tsma -- Regression Defense Line for Legacy Code</title><link>https://www.parkjunwoo.com/tech/tsma/</link><pubDate>Thu, 14 May 2026 09:00:00 +0900</pubDate><guid>https://www.parkjunwoo.com/tech/tsma/</guid><description>A CLI tool that indexes every function, detects test presence, measures coverage, and gives precise feedback to LLM agents. One command builds a regression defense line around legacy code.</description></item><item><title>filefunc — One File, One Concept</title><link>https://www.parkjunwoo.com/tech/filefunc/</link><pubDate>Mon, 16 Mar 2026 13:50:00 +0900</pubDate><guid>https://www.parkjunwoo.com/tech/filefunc/</guid><description>The navigation unit for an AI code agent is the file. filefunc is a Go code structure convention and CLI tool that enforces one concept per file.</description></item></channel></rss>