Module agent 

Scoped, sequential Agent reviews with budgets and one tools-disabled verdict repair, and follow-ups that continue a saved review’s conversation.

Modules

error

Why an Agent review failed, as a request errorCode.

session

Saved Agent conversations: one append-only JSON Lines file per review session, which session show reads and session send continues.

verdict

Structs

Continuation

What a follow-up knows besides the person’s message.

Review

Enums

FollowUp

Preparation failure leaves the saved conversation unchanged; a recorded send has an answer (including failure/cancellation) and attempts. This internal result is not serialized.

Constants

FOLLOW_UP

What every follow-up’s question starts with, before the person’s message: the review’s system prompt still asks for a JSON verdict, and this turn sets that aside.

Functions

execute

state holds Codex credentials; the API-key backends read only the environment. recorder holds the review’s session_id and saves its conversation.

follow_up

Continue a saved review’s conversation with a person’s message, as recorder appends it: the backend, model, reasoning and session id the review used, the eval’s Agent tools resolved against the current workspace, and the review’s budgets. The answer is free text, and nothing about the recorded review changes.

session_id

A new review’s session id, a random UUID: every request of the review, its turns, retries and repair turn, carries it as the provider’s prompt-cache identity.