02 / IT · Analysis
Postman drill examines where agent-handled tokens end up
A staged drill using a fake token shifts attention from whether an agent completes a replacement to where secrets remain.

Postman presented a staged token-replacement drill in which a coding agent never saw the new token, according to its DEV Community post. The exercise concerns developers who delegate work involving leaked credentials to coding agents.
The post frames the issue around where secrets travel, rather than whether the agent completes the task. Its use of a fake token makes the exercise a simulation of a leak, rather than evidence of handling an actual incident.
The post identifies the model's context, chat transcripts and shell command history as places where secrets may end up. These give developers a separate basis for assessing delegated work: what information enters the agent's input and what remains in records.
Where token values remain
Postman's title pairs replacement of a leaked token with keeping the new token out of the agent's view. That framing draws a distinction between assigning a task and revealing the secret value needed within the workflow.
The post extends its concern beyond the model's input to conversation records and shell history. On that reading, completion of the replacement and the contents of the resulting records are separate matters to assess.
A replacement drill with a fake token
According to the DEV Community post, the exercise used a fake token in a staged leaked-credential scenario. It presents a simulated workflow, rather than a record of a response to a real leak.
Postman connects the drill to the path secrets take through an agent's work. The resulting criterion for developers is broader than task completion: it also concerns how much secret information the agent needs to receive.