DeepSeek-Reasonix: How a poisoned config can hijack an AI coding agent

GitLab’s Threat Research Group disclosed ConfigPoisoning, a critical command-execution vulnerability tracked as GHSA-grg2-7gc6-36m6 and CVE-2026-102437 in DeepSeek-Reasonix Studio and its npm package, where attacker-controlled commands in a repository’s local Git configuration can execute when a developer views a diff. The issue arises because the tool disabled several dangerous Git features but did not neutralize per-file `filter.<driver>.clean` commands selected through `.gitattributes`, allowing a poisoned clean filter to run during diff generation despite flags such as `--no-ext-diff` and `--no-textconv`. Affected users should update to Studio 2.21.0 or DeepSeek Reasonix npm 1.39.3, while users of older versions are advised not to diff repositories obtained through potentially unsafe delivery methods. GitLab notes that ordinary GitLab HTTPS or SSH clones do not transfer local `.git/config` files, but poisoned configurations may arrive through archives, caches, synced folders, devcontainers, or be written by compromised agents or extensions already on a developer’s machine. The research argues that similar risks affect other AI coding tools and Git-wrapping applications, recommending that developers avoid Git filter and text-conversion machinery where possible or comprehensively override relevant repository-controlled Git settings on every invocation.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.