Support assigned reviewers in LLM approval reviews - #72157
Conversation
b63e1b0 to
8e8b27c
Compare
8e8b27c to
3030164
Compare
3030164 to
db4da4a
Compare
db4da4a to
13467ee
Compare
13467ee to
9f16fdd
Compare
The review was always open to every user with the HITL permission, so an operator could not route an approval to the people who actually own the decision. The Task SDK already accepts assigned users for a review; the LLM operators simply never passed any.
… rules A malformed approval_assigned_users entry only failed inside defer_for_approval, after the model call was billed, and the version guard tested the raw argument while the mixin tested the normalized one. The old-core guard test never ran on any lane because its skipif contradicted the class-level one. The docs did not say that id is the auth manager user id, that 3.1.0 to 3.1.5 compare name as well, or that the list is fixed once the review row exists.
9f16fdd to
65c6efd
Compare
|
Round 1 is all addressed at One non-blocking doc thought, nothing gating:
|
|
Added in 6e25e2f: with the default |
Why
How
approval_assigned_usersto name the reviewers allowed to answer.HITLOperatoralready does.Was generative AI tooling used to co-author this PR?
Generated-by: Claude Code (Opus 5) following the guidelines