Skip to main content
Un-blackbox a Flow: what definition actually ran? Handy for support (“what prompt did this agent actually see?”), for pre-run assertions (“is the model I expect wired up?”), and for embedding the “why” panel next to answers in your admin UI. Authenticated by PAT read scope.

resolve(slug)

Returns the resolved definition for every agent in the Flow — not just the entry point. Reflects the currently-published version by default; there’s no draft-vs-published toggle here (use the CLI’s nora flows versions for that).

Common patterns

Pre-run assertion

Fail loudly if the model you expect isn’t wired up, before you spend on a run:

The “why did this answer look like this?” support panel

Next to a trace in your admin UI, pull the current definition so support engineers see the prompt the agent actually ran with:

Audit for drift

Snapshot the resolved definition to your source of truth and diff on schedule — anything unexpected is a signal that someone edited the Flow outside change control.

Scope

resolve requires PAT read. Read tokens can surface prompts and tool config, which may contain end-user PII or credentials — treat any read token as admin-only and don’t hand it to browser code (SDK refuses to init in the browser regardless; see Overview).