Reviewable lineage for high-stakes ADK workflows #4908
Ratnaditya-J
started this conversation in
Show and tell
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Has anyone here used ADK in workflows that need external review or audit?
I've been testing that setup and ran into a gap between what
rewindgives you operationally and what reviewers usually ask for later.ADK already gives a strong runtime foundation with:
rewindresumeWhat I needed on top of that was something more concrete: a layer that diffs two runs, lets a human approve the change, and exports a signed evidence package.
I built a small layer for this, and the fit with ADK feels pretty natural.
ADK handles runtime continuity. The extra layer handles review, approval, and evidence.
flowchart LR U["Operator / Reviewer"] --> SF["Review / Evidence Layer"] SF --> ADK["Google ADK Runtime"] ADK --> M["Model"] ADK --> T["Tools"] ADK --> E["External Systems"] ADK --> H["Session History / Events / Rewind"] H --> SFThe reason this stood out to me is that ADK's docs are already pretty clear about the boundary:
That all makes sense for a runtime. It also creates a clean place for a reviewable lineage layer above the runtime.
The practical questions I kept running into were:
That feels adjacent to ADK rather than part of ADK itself.
So the framing that has been useful for me is:
I have a small integration example (SpineFrame) if that's useful to folks working on regulated or review-heavy workflows.
Reference Docs
Beta Was this translation helpful? Give feedback.
All reactions