The report
A deployment command usually needs permission to push a container image, and a temporary registry token is one way to provide it. In the affected LangGraph CLI versions, verbose mode logged subprocess standard input while the deployment routine was feeding a token to Docker login. Anyone allowed to read the resulting terminal capture or CI logs could potentially recover a credential that should have remained transient.
The GitHub advisory names langgraph-cli versions 0.4.15 through 0.4.31 and lists 0.4.32 as the security fix. It additionally notes that 0.4.37 stops verbose logging of input to other launched commands. The leaked token was limited to the deployment's image repository and lasted no more than twelve hours. That does not grant arbitrary control of the application's other secrets or permission to activate a new image in an existing digest-pinned deployment.
The dangerous transition was from stdin to logs
Verbose diagnostics are often treated as troubleshooting data, yet this case shows how the logging layer can become a second credential store. Rotating a token after its maximum lifetime cannot repair a previously replaced image, but a past log by itself also does not prove that any image was modified. The Case separates potential token use from observed log exposure.
Recheck deployment logs produced by affected releases, constrain access to those logs and update the CLI. For incident reconstruction, identify the image repository, token validity interval, actual pushes and subsequent deployments rather than reporting an unspecified compromise of the AI agent.