📚 Stacked PRs

• fix(BA-6887): reference the real from_status/to_status columns in kernel history conditions #12857 — BA-6887 fix: kernel history conditions reference the real status columns ✅
• feat(BA-6889): add kernel scheduling-history query conditions and orders #12867 — BA-6889 models: kernel query conditions and orders ✅
• feat(BA-6890): add kernel scheduling-history repository search scope #12868 — BA-6890 repository: kernel search scope and scoped search ✅
• feat(BA-6891): add kernel scheduling-history v2 DTOs and service actions #12869 — BA-6891 DTO + service: kernel v2 DTOs and search actions ← you are here
• feat(BA-6892): add kernel scheduling-history adapter and REST v2 endpoints #12870 — BA-6892 adapter + REST v2: kernel endpoints
• feat(BA-6893): add kernel scheduling-history SDK v2 client and CLI commands #12866 — BA-6893 SDK + CLI: kernel commands

Merge in order, bottom-up. Each PR is based on the one above it, so its diff shows only its own layer.

What

The v2 DTOs and the service layer for kernel scheduling history.

Review notes

The admin search is a global action. It is system-wide and unscoped, so its authorization is a SUPERADMIN role gate rather than RBAC scope resolution — which is what BaseGlobalAction / GlobalActionProcessor model. The processor applies the gate on the action path, behind the route's superadmin_required middleware. The scoped counterpart resolves against an RBAC scope and stays on the plain action path.

This makes search_kernel_history the only one of the four admin history searches on the global path; session/deployment/route predate GlobalActionProcessor and are untouched here. Converting them is a separate change — 44 REST v2 domains use superadmin_required and only retention_policy has adopted the new processor so far.

The DTOs are shared with GraphQL. Per common/dto/manager/v2/AGENTS.md these are the single source of truth for REST v2 and the Strawberry schema, so BA-6883 builds its GraphQL types on these rather than defining its own. That is why the scope's non-emptiness rule lives in KernelHistoryScopeDTO.model_validator — one rule, both surfaces, rather than restated per surface.

KernelHistoryNode has no sub_steps, deliberately. kernel_scheduling_history defines no such column — the 84b901f69d16 migration added sub_steps to the deployment and route tables only. Worth flagging: BEP-1061 (§2.3) and BA-6852 both describe kernel history as carrying sub_steps, which does not match the schema. That discrepancy belongs to the writer side (BA-6852) and is left alone here.

Only the two inputs that get wired are added. There is no offset-only SearchKernelHistoryInput mirroring SearchSessionHistoryInput — that one has no callers, and mirroring it would add dead code.

The service methods are thin passthroughs, consistent with the existing ones. SearchKernelScopedHistoryAction is a plain @dataclass, not frozen=True: the shared SchedulingHistoryAction base is non-frozen, so a frozen subclass is a TypeError. (services/AGENTS.md asks for frozen; the base and all six existing siblings already diverge. Out of scope here.)

The adapter that converts these DTOs and drives the processors is #12870.

🤖 Generated with Claude Code

📚 Documentation preview 📚: https://sorna--12869.org.readthedocs.build/en/12869/

📚 Documentation preview 📚: https://sorna-ko--12869.org.readthedocs.build/ko/12869/