Technology
AI coding assistants execute background git commands in untrusted repositories
Developers face potential risks from background git processes reading local repository settings, according to Hacker News.
The short version
- Researchers found that multiple AI coding assistants execute background git commands to collect context prior to user prompts or authentication.[Hacker News]
- Configurations within local repositories, such as core.fsmonitor settings, can be read during automated background git operations.[Hacker News]
- Vendors and vulnerability authorities assigned CVE identifiers to specific affected tools, including Goose and Hermes Agent.[Hacker News]
Key facts
- Researchers reported that AI coding assistants execute background git operations to collect workspace context before authentication or trust confirmation prompts appear.[Hacker News]
- Git evaluates performance configuration options such as core.fsmonitor directly from a repository's local .git/config file.[Hacker News]
- Maintainers assigned CVE-2026-72718 with a 7.0 severity rating to Goose following the security notification.[Hacker News]
- Independent authority VulnCheck designated CVE-2026-71963 for an issue involving Hermes Agent.[Hacker News]
What remains uncertain
- The full scope of remediation across all affected agent platforms has not been universally detailed.[Hacker News]
Sources
Outlet counts describe coverage, not independent confirmation. Reports may share a wire service or original source.