Skip to content

armgit / cindy

Public Imported from GitHub
0

Consider it done. The open-source AI agent that works out of the box · 想到,就能做到。开源、开箱即用的 AI Agent。

test(pi): 守护快照外模型热切后的档位显式下发

zyjisdog59a8f62bc135
diff --git a/packages/maker-core/src/agents/pi/__tests__/pi-provider-routing.test.ts b/packages/maker-core/src/agents/pi/__tests__/pi-provider-routing.test.ts
index 997b58155a..4550381175 100644
--- a/packages/maker-core/src/agents/pi/__tests__/pi-provider-routing.test.ts
+++ b/packages/maker-core/src/agents/pi/__tests__/pi-provider-routing.test.ts
@@ -4037,6 +4037,89 @@ describe("Pi provider-aware model routing", () => {
await handle.close();
});
+ it("re-declares the session effort when the switch target only exists after session start", async () => {
+ // 2026-10-09 实测:用户在会话中途新建 provider/模型后直接热切。该路由不在
+ // 启动快照里——preview 必须走 live-refresh 重建能力表,收敛块随后按会话档位
+ // 显式下发 set_thinking_level。这条链一旦断裂(快照外就沉默、或误判目标不
+ // 支持思考而发 off),切模后的首轮就没有思考声明,模型会把推理写进正文。
+ let includeLate = false;
+ const lateProvider = {
+ id: "late-prov",
+ sourceProviderId: "late-prov",
+ name: "Late Provider",
+ baseUrl: "http://late.test",
+ api: "openai-completions" as const,
+ models: [
+ {
+ id: "late-model",
+ wireId: "late-model",
+ reasoning: true,
+ contextWindow: 200_000,
+ thinkingLevelMap: {
+ minimal: null,
+ low: "low",
+ medium: null,
+ high: "high",
+ xhigh: null,
+ max: "max",
+ },
+ },
+ ],
+ };
+ const startupProviders = {
+ providers: [
+ {
+ id: "native-a",
+ sourceProviderId: "native-a",
+ name: "Native A",
+ baseUrl: "http://a.test",
+ api: "openai-completions" as const,
+ models: [{ id: "local-model", contextWindow: 200_000 }],
+ },
+ ],
+ env: {},
+ };
+ const agent = new PiAgent(
+ byomDeps(
+ async () => (includeLate ? { providers: [lateProvider], env: {} } : startupProviders),
+ [
+ { id: "local-model", displayName: "Local", contextWindow: 200_000, efforts: [], defaultEffort: null },
+ {
+ id: "late-model",
+ displayName: "Late",
+ contextWindow: 200_000,
+ efforts: ["low", "high", "max"],
+ defaultEffort: null,
+ },
+ ],
+ ),
+ );
+ const handle = await agent.startSession({
+ sessionId: "late-provider-thinking",
+ workingDir: cwd,
+ model: "local-model",
+ providerId: "native-a",
+ effort: "max",
+ });
+ includeLate = true;
+ const beforeSwitch = captured.requests.length;
+
+ // 无 intent 载体(旁路入口 / UI 只选模型的形状):档位仍必须按会话档位显式
+ // 重发。2026-10-06 lead 案例就是切换没带载体 → 上游 off 原样继承 →
+ // usage.reasoning 归零、推理整段写进正文。
+ await handle.setModel!("late-model", { providerId: "late-prov" });
+
+ const switched = captured.requests.slice(beforeSwitch);
+ // 目标不在启动快照 → 必须发生一次 native catalog 重建(get_available_models
+ // 或 models.json 重写),否则后面的档位收敛无从谈起。
+ expect(switched.some((request) => request.type === "set_model")).toBe(true);
+ // 档位必须显式下发为会话档位:既不能沉默(一条都没有),更不能误判成
+ // 「不支持思考」而主动发 off。
+ expect(switched).toContainEqual({ type: "set_thinking_level", level: "max" });
+ expect(switched).not.toContainEqual({ type: "set_thinking_level", level: "off" });
+ await handle.close();
+ });
+
it("reopens thinking at the session effort instead of a hardcoded xhigh", async () => {
const denseMap = {
minimal: "minimal",