Release v0.6.0 · lusblead/dsh-Kingdom

v0.6.0 — M1-C Execution Truth

ExecutionProfile on Worker Binding (kingdom_set_execution_profile, OWNER-only under session-bound):
{ provider?, model? } decides who actually executes — the Worker Binding now truly
determines the runtime, not just the record.
ExecutorFactory is the single execution-resolution entry: assigned_binding_id → Binding → ExecutionProfile → executor; Core still depends only on WorkerExecutor.
Execution evidence columns on executions (SQL-queryable): executor_kind,
provider, provider_source (binding/global-fallback), requested_model,
resolved_model, model_source (binding/parent-inherited/unknown), plus an
immutable resolution snapshot JSON (requested/resolved/source).

Schema v2 (first versioned, transactional, idempotent migration): v1 DBs converge
on open; SCHEMA_VERSION=2 is now a truthful field.
binding.model_name is seat metadata only — ExecutorFactory MUST NOT read it
(hard invariant, test-locked).
resolved_model is what DSH runtime actually resolved (in-process observed);
null = seam has no evidence, never "no model". Future transport telemetry adds
observed_model.
No profile → compatible fallback to global workerProvider with explicit
global-fallback + parent-inherited evidence.

dsh-kingdom-0.6.0.tgz (npm dsh-kingdom@0.6.0, latest)
dsh-kingdom-gui-0.6.0.zip (execution-profile display; seat-model vs exec-model vs resolved-model kept distinct)

Plugin tests 27/27 (incl. Schema v2 migration, admin-authz for profiles, E1/E2/E3
resolution, 4+1 identity matrix, execution evidence chain)
GUI tests 42/42
M1-D Release Gate PASSED: live 3-run attribution experiment (E1 binding-specified,
E2 provider-only, E3 no-profile) — executions evidence columns match each case exactly,
resolved_model observed from the real subagent runtime
Design: docs/M1-C-EXECUTION-TRUTH-DESIGN.md (frozen v2); audit: docs/M1-TRUE-GOVERNANCE-AUDIT.md