Run need_generate sampling via batch chance in auditor maintain path.

Auditor packets can request generation-rule pools without LLM randomness; results land in 运行.本轮.旁观.生成抽样 and clear each play turn.

Co-authored-by: Cursor <cursoragent@cursor.com>

tests/maintain-need-generate.test.ts

@@ -0,0 +1,43 @@
+import { describe, expect, it } from "vitest";
+import {
+  buildChanceBatchFromGenerationRule,
+  runMaintainNeedGenerateSampling,
+} from "../src/skills/maintain-need-generate.js";
+
+const SAMPLE_RULES = JSON.stringify({
+  schema: "context-fragment.v1",
+  正文: {
+    rules: [
+      {
+        rule_id: "npc-trait",
+        池: [
+          {
+            pool_id: "mood",
+            条目: [
+              { id: "calm", 权重: 2 },
+              { id: "angry", 权重: 1 },
+            ],
+          },
+        ],
+      },
+    ],
+  },
+});
+
+describe("maintain need_generate sampling", () => {
+  it("builds pick requests from generation rule pools", () => {
+    const batch = buildChanceBatchFromGenerationRule(SAMPLE_RULES, "npc-trait");
+    expect(batch).toHaveLength(1);
+    expect(batch![0]!.id).toBe("pool:mood");
+    expect(batch![0]!.op).toBe("pick");
+  });
+
+  it("runs batch sampling for need_generate", () => {
+    const result = runMaintainNeedGenerateSampling({
+      need: { rule_id: "npc-trait", reason: "补一条" },
+      generationRulesRaw: SAMPLE_RULES,
+    });
+    expect(result?.schema).toBe("chance.batch.v1");
+    expect(result?.results[0]?.id).toBe("pool:mood");
+  });
+});