Expand description
The human review surface over HTTP (lex-hub#92, first slice).
In an agent-native VCS humans are not the gate — agents produce and verify
continuously. The human’s job is to adjudicate exceptions and arbitrate
intent: not to read every diff, but to look at the changes a machine gate
couldn’t close, see why each was made (its recorded Intent), and record a
verdict. Those verdicts are Review attestations — a human decision lives in
the same typed attestation graph as lex-hub-ci’s TypeCheck or a Replay, and
the same gates consume it (a standing Reject blocks promotion).
Two endpoints, tenant-scoped (auth + store selection live in lex-hub):
GET /v1/review/inbox[?branch=<b>]— the head’s stages, each with its intent and current review state;needs_reviewflags the exceptions.POST /v1/review/verdict— record an Approve / Reject / RequestChanges verdict as aReviewattestation.
Functions§
- review_
inbox_ handler GET /v1/review/inbox[?branch=<name>]— the review inbox for a branch head: each head stage with its declaration name, the intent it was made under (the why), its latest review verdict, and whether it still needs a human look.- review_
verdict_ handler POST /v1/review/verdict— record a human review verdict as aReviewattestation on a stage. The verdict enters the same attestation graph the gates consume (a Reject blocks promotion vialatest_review_verdict).