Many users praised the founder of OneCLI for the insight of keeping credentials entirely out of the AI agent's context window to block prompt injection, while some dismissed the encryption and proxy approach as too weak for real security.
Based on 3 visible X reactions from 6 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@jonathanfishner the key insight is keeping the credential out of the context window entirely. if the agent never reads the key, prompt injection can't leak it. a credential store the agent talks to by id avoids that
@jonathanfishner AES-256-GCM is military grade encryption. setting a proxy is putting a padlock on a screen door. convenience
Many users praised the founder of OneCLI for the insight of keeping credentials entirely out of the AI agent's context window to block prompt injection, while some dismissed the encryption and proxy approach as too weak for real security.
Based on 3 visible X reactions from 6 accounts; directional sample.
Ask a question below.
Published answers will appear here.