MSRC's October release fixes CVE-2026-102232, a Visual Studio Code security feature bypass (Important, CVSS 7.0, CWE-863). Per Microsoft's FAQ, crafted content processed by the AI coding agent could modify a configured lifecycle-hook file without the expected confirmation, leading to commands running as the user. The chain requires prompt injection, Copilot agent interaction and then a build trigger (AC:H, UI:R). It is fixed in VS Code 1.141.1 (released Oct 13); versions 1.0.0 through 1.141.0 are affected. Microsoft marks it not publicly disclosed and not exploited. This is the second hook-file bypass this year: GHSA-w79w-rj9h-vg4f / CVE-2026-70335 (fixed in 1.132.1, August) added confirmation before agents write custom-agent hook configs. Treat agent-writable config (hooks, MCP and agent definitions in .github/ and ~/.copilot/) as code. Review diffs to those paths, and don't run agent mode on untrusted repos with pre-1.141.1 builds.