Executive brief
CodexBar, a tool used for managing AI provider integrations and costs, was found to have a security flaw in how it handles network requests. An attacker could potentially trick the application into sending sensitive information, such as login cookies or API keys, to a server they control. This could lead to the theft of user credentials or unauthorized access to AI services like OpenAI.
Technical details
A credential forwarding vulnerability exists in CodexBar's ProviderHTTPClient transport before version 0.33.0. The shared production client used default URLSession redirect handling, which failed to strip sensitive headers (cookies, bearer tokens, API keys) when following redirects to different origins or downgraded HTTP destinations. An attacker capable of influencing the destination of a provider request (e.g., via a man-in-the-middle or a compromised provider endpoint) could redirect the authenticated request to an attacker-controlled host to capture credentials. The fix, implemented in version 0.33.0, introduces a redirect guard that blocks redirects leaving the original HTTPS origin while preserving same-origin redirects.
Affected products
- steipete CodexBar before 0.33.0
Timeline
- 2026-05-31: disclosed: Initial pull request submitted by researcher
- 2026-06-10: patched: Fix merged into main branch
- 2026-06-11: advisory: CVE published and version 0.33.0 released