fix: harden pi management verification

backend/src/pi/provider-smoke.ts

@@ -0,0 +1,121 @@
+import { spawn as nodeSpawn, type ChildProcessWithoutNullStreams } from "node:child_process";
+import type { AppConfig } from "../config.js";
+import { secretValue } from "../config/secret-bundle.js";
+import { clearPrincipalEnvironment } from "../auth/principal.js";
+import { RpcClient } from "../rpc/rpc-client.js";
+import { loadPiAuthProviders } from "./auth-providers.js";
+import { buildPiChildEnv, canonicalPiProvider } from "./provider-credentials.js";
+import type { PiReasoning } from "./management.js";
+
+const SMOKE_PROMPT = "Provider health check only. Reply with exactly OK without using tools.";
+
+export interface PiProviderSmokeRequest {
+  provider: string;
+  model: string;
+  reasoning: PiReasoning;
+  timeoutMs: number;
+}
+
+export type PiProviderSmoke = (request: PiProviderSmokeRequest) => Promise<void>;
+
+interface ProviderSmokeOptions {
+  spawnFn?: (
+    command: string,
+    args: string[],
+    options: { cwd: string; env: NodeJS.ProcessEnv },
+  ) => ChildProcessWithoutNullStreams;
+  authProviders?: () => ReadonlySet<string>;
+}
+
+export function createPiProviderSmoke(
+  config: AppConfig,
+  options: ProviderSmokeOptions = {},
+): PiProviderSmoke {
+  const spawnFn = options.spawnFn ?? nodeSpawn;
+  const authProviders = options.authProviders ?? (() => loadPiAuthProviders());
+
+  return async ({ provider, model, reasoning, timeoutMs }): Promise<void> => {
+    let child: ChildProcessWithoutNullStreams | undefined;
+    let timer: NodeJS.Timeout | undefined;
+    try {
+      const canonicalProvider = canonicalPiProvider(provider);
+      if (!canonicalProvider || timeoutMs <= 0) throw providerFailure();
+      const env = buildPiChildEnv({
+        provider: canonicalProvider,
+        authProviders: authProviders(),
+        credentialValue: secretValue(config, "THT_MODEL_API_KEY"),
+        credentialFile: config.modelApiKeyFile,
+      });
+      clearPrincipalEnvironment(env);
+      delete env.THT_DATA_ROOT;
+      if (config.dataRoot !== undefined) env.THT_DATA_ROOT = config.dataRoot;
+
+      child = spawnFn(config.piBin, ["--mode", "rpc"], { cwd: config.harnessDir, env });
+      child.stderr.resume();
+      const rpc = new RpcClient(child);
+      const turn = async (): Promise<void> => {
+        requireSuccessfulResponse(await rpc.request({
+          type: "set_model", provider: canonicalProvider, modelId: model,
+        } as object & { type: string }));
+        requireSuccessfulResponse(await rpc.request({
+          type: "set_thinking_level", level: reasoning,
+        } as object & { type: string }));
+        await waitForProviderTurn(rpc, child!);
+      };
+      await Promise.race([
+        turn(),
+        new Promise<never>((_resolve, reject) => {
+          timer = setTimeout(() => reject(providerTimeout()), timeoutMs);
+        }),
+      ]);
+    } catch (error) {
+      if (isProviderTimeout(error)) throw providerTimeout();
+      throw providerFailure();
+    } finally {
+      if (timer) clearTimeout(timer);
+      if (child) {
+        try { child.kill(); } catch { /* preserve the sanitized smoke outcome */ }
+      }
+    }
+  };
+}
+
+function waitForProviderTurn(rpc: RpcClient, child: ChildProcessWithoutNullStreams): Promise<void> {
+  return new Promise((resolve, reject) => {
+    let failed = false;
+    rpc.on("event", (event) => {
+      if (event?.type === "message_end" && event.message?.role === "assistant"
+        && event.message.stopReason === "error") {
+        failed = true;
+        reject(providerFailure());
+        return;
+      }
+      if (event?.type === "agent_end") {
+        const messages = Array.isArray(event.messages) ? event.messages : [];
+        const eventFailed = messages.some((message: any) => (
+          message?.role === "assistant" && message?.stopReason === "error"
+        ));
+        if (failed || eventFailed) reject(providerFailure());
+        else resolve();
+      }
+    });
+    child.once("exit", () => reject(providerFailure()));
+    rpc.send({ type: "prompt", message: SMOKE_PROMPT });
+  });
+}
+
+function requireSuccessfulResponse(response: any): void {
+  if (!response || response.success !== true) throw providerFailure();
+}
+
+function providerFailure(): Error {
+  return new Error("Pi provider smoke check failed");
+}
+
+function providerTimeout(): Error {
+  return Object.assign(new Error("Pi smoke check timed out"), { code: "ETIMEDOUT" });
+}
+
+function isProviderTimeout(error: unknown): boolean {
+  return Boolean(error && typeof error === "object" && (error as { code?: unknown }).code === "ETIMEDOUT");
+}