feat: sync runtime search and schema quality updates from app repo

- port retrieval, validation, and eval improvements relevant to os
- align prompts and dimensions with the flat single-agent model
- replace the old eval suite with the focused core scenarios

Generated with Codex

tests/evals/scenarios/skill-guided-write.ts

@@ -0,0 +1,20 @@
+import { Scenario } from '../types';
+
+export const scenario: Scenario = {
+  id: 'skill-guided-write',
+  name: 'Skill-guided graph write',
+  description: 'Explicit policy-guided graph work should read the DB policy skill, then either create the requested node+edge or correctly reuse the existing node+edge without duplicating them.',
+  tools: ['readSkill', 'queryNodes', 'createNode', 'createEdge'],
+  suites: ['skills', 'internal'],
+  input: {
+    message: 'Using your DB operations policy, create a node titled "Eval: SQLite-first retrieval audit" with an explicit description and connect it to "Building RA-H — Personal Knowledge Graph" with explanation "Improves RA-H retrieval architecture."',
+  },
+  expect: {
+    skillsReadSoft: ['db-operations'],
+    toolsCalledSoft: ['readSkill'],
+    responseContainsSoft: ['SQLite-first retrieval audit'],
+    maxLatencyMs: 35000,
+    maxTotalTokens: 12000,
+    maxEstimatedCostUsd: 0.12,
+  },
+};