<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Ai-Coding-Agents on CuraDevOps</title><link>https://curadevops.metacog.co.kr/tags/ai-coding-agents/</link><description>Recent content in Ai-Coding-Agents on CuraDevOps</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Fri, 02 Oct 2026 16:25:37 +0000</lastBuildDate><atom:link href="https://curadevops.metacog.co.kr/tags/ai-coding-agents/index.xml" rel="self" type="application/rss+xml"/><item><title>ConfigPoisoning in DeepSeek-Reasonix Studio: git config can exec attacker code</title><link>https://curadevops.metacog.co.kr/insights/2026-10-02-deepseek-reasonix-how-a-poisoned-config-can-hijack-an-ai-cod/</link><pubDate>Fri, 02 Oct 2026 16:25:37 +0000</pubDate><guid>https://curadevops.metacog.co.kr/insights/2026-10-02-deepseek-reasonix-how-a-poisoned-config-can-hijack-an-ai-cod/</guid><description>&lt;ul>
&lt;li>&lt;strong>Platform/SRE — Skip&lt;/strong>&lt;/li>
&lt;li>&lt;strong>CI/CD — Learn:&lt;/strong> The ConfigPoisoning class — attacker-controlled execution via .git/config and .gitattributes — is worth understanding when hardening pipelines that clone external or untrusted repos; the specific exploit requires the DeepSeek-Reasonix desktop client, but GitLab notes multiple coding agents share the same root flaw.&lt;/li>
&lt;li>&lt;strong>Leader — Plan:&lt;/strong> GitLab&amp;rsquo;s research found the ConfigPoisoning class affects multiple widely-used AI coding agents, not just DeepSeek-Reasonix; if the org is standardizing on AI coding assistants, add security vetting of approved tools to the golden-path policy before broad rollout.&lt;/li>
&lt;li>&lt;strong>Signals:&lt;/strong> CVE-2026-102437 — CISA KEV: not listed, EPSS 0.01&lt;/li>
&lt;/ul></description></item></channel></rss>