The highlighted Expo occurrence is a documentation example sending an OpenAI key to OpenAI, not a shared helper sending that key to another host. The runtime TypeSafe judge uses a key configured for TypeSafe and sends it to the TypeSafe endpoint for its prompt-classification feature; no malicious credential flow was found.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
The CLI runs when invoked; the optional judge runs from the installed prompt hook on non-status turns when a TypeSafe key is configured.
Impact
The inspected flows do not show credentials being sent to an unrelated or publisher-controlled destination outside the relevant service.
Mechanism
The judge redacts credential-shaped prompt values and sends the resulting prompt plus stack context to the fixed TypeSafe API using the configured TypeSafe key. The Expo skill contains an example that sends an OpenAI key to OpenAI.
Evidence
stacks/_shared/hooks/jev-judge.tsstacks/_shared/hooks/user-prompt-submit.tsstacks/react-native/skills/expo-official/upstream/eas-hosting/SKILL.mddist/installer.js