Add Beryl Agent and harness source

2026-09-08 20:10:03 +08:00
parent 06d72d001c
commit 47cfbcd66a
70 changed files with 12558 additions and 0 deletions
@@ -0,0 +1,175 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+import { mkdtemp, mkdir, readFile, symlink } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import path from "node:path";
+import {
+  ApprovalBroker,
+  MemoryEventStore,
+  RunIdleWatchdog,
+  SafeWorkspace,
+  adsMetrics,
+  boundSwadsArguments,
+  compactSwadsText,
+  createTools,
+  validateBaseUrl,
+} from "../src/index.js";
+let roots: string[] = [];
+afterEach(async () => {
+  for (const root of roots)
+    await import("node:fs/promises").then((f) =>
+      f.rm(root, { recursive: true, force: true }),
+    );
+  roots = [];
+});
+async function ws() {
+  const root = await mkdtemp(path.join(tmpdir(), "asm-"));
+  roots.push(root);
+  const sessions = path.join(root, "sessions");
+  await mkdir(sessions);
+  return SafeWorkspace.create(sessions, "s1");
+}
+describe("harness core", () => {
+  it("assigns monotonic event sequence", () => {
+    const s = new MemoryEventStore();
+    expect(s.append("x", { type: "run.started", payload: {} }).sequence).toBe(
+      1,
+    );
+    expect(s.append("x", { type: "run.completed", payload: {} }).sequence).toBe(
+      2,
+    );
+  });
+  it("approves and is idempotent", async () => {
+    const b = new ApprovalBroker(1000);
+    const p = b.request(
+      {
+        sessionId: "s",
+        runId: "r",
+        toolCallId: "t",
+        toolName: "save_report_draft",
+        arguments: {},
+        riskLevel: "medium",
+      },
+      new AbortController().signal,
+    );
+    const id = b.get(
+      [...(b as unknown as { all: Map<string, unknown> }).all.keys()][0]!,
+    )!.approvalId;
+    b.resolve(id, "approved");
+    expect((await p).decision).toBe("approved");
+    expect(b.resolve(id, "approved").status).toBe("approved");
+  });
+  it("times out approval", async () => {
+    vi.useFakeTimers();
+    const b = new ApprovalBroker(10);
+    const p = b.request(
+      {
+        sessionId: "s",
+        runId: "r",
+        toolCallId: "t",
+        toolName: "save_report_draft",
+        arguments: {},
+        riskLevel: "medium",
+      },
+      new AbortController().signal,
+    );
+    await vi.advanceTimersByTimeAsync(11);
+    expect((await p).decision).toBe("expired");
+    vi.useRealTimers();
+  });
+  it("uses an activity-based watchdog and pauses while waiting", async () => {
+    vi.useFakeTimers();
+    const onTimeout = vi.fn();
+    const watchdog = new RunIdleWatchdog(100, onTimeout);
+    watchdog.touch();
+    await vi.advanceTimersByTimeAsync(75);
+    watchdog.touch();
+    await vi.advanceTimersByTimeAsync(75);
+    expect(onTimeout).not.toHaveBeenCalled();
+    watchdog.pause();
+    await vi.advanceTimersByTimeAsync(500);
+    expect(onTimeout).not.toHaveBeenCalled();
+    watchdog.touch();
+    await vi.advanceTimersByTimeAsync(101);
+    expect(onTimeout).toHaveBeenCalledTimes(1);
+    watchdog.stop();
+    vi.useRealTimers();
+  });
+  it("writes only report.md", async () => {
+    const w = await ws();
+    await w.writeReport("hello", "Demo");
+    expect(await readFile(path.join(w.root, "report.md"), "utf8")).toContain(
+      "hello",
+    );
+    expect(() => w.resolve("../escape")).toThrow("PATH_ESCAPE");
+  });
+  it("renders daily and weekly reports as PNG with a three-stage funnel", async () => {
+    const w = await ws();
+    const fixture = path.resolve("../../fixtures/ads-account.json");
+    const imageTool = createTools(fixture, w).find(
+      (tool) => tool.name === "save_report_image",
+    )!;
+    const result = await imageTool.execute(
+      "tool-1",
+      {
+        reportType: "weekly",
+        accountName: "Lemonz Home",
+        period: "2026-08-31 – 2026-09-06",
+        timezone: "Asia/Kuala_Lumpur",
+        currency: "MYR",
+        freshness: "2026-09-07 10:39",
+        outcome: "成交增长，素材效率出现分化。",
+        kpis: [
+          { label: "花费", value: "RM 100" },
+          { label: "成交", value: "RM 700" },
+          { label: "订单", value: "40" },
+          { label: "ROI", value: "7.00" },
+        ],
+        funnel: [
+          { label: "商品曝光", value: "10,000" },
+          { label: "商品点击", value: "500", rate: "曝光→点击 5.00%" },
+          { label: "订单", value: "40", rate: "点击→购买 8.00%" },
+        ],
+        funnelCaption: "同一商品指标族与归因窗口。",
+        highlights: ["高效商品贡献主要成交。"],
+        recommendations: ["复核低效素材。", "验证高效方向。", "继续观察低量样本。"],
+        limitations: ["平台归因不是因果增量。"],
+      },
+      new AbortController().signal,
+      undefined,
+      {} as never,
+    );
+    const png = await readFile(path.join(w.root, "swads-weekly-report.png"));
+    expect(png.subarray(1, 4).toString()).toBe("PNG");
+    expect((result.details as { funnelIncluded: boolean }).funnelIncluded).toBe(true);
+  });
+  it("rejects symlink report", async () => {
+    const w = await ws();
+    await symlink("/tmp/outside", path.join(w.root, "report.md"));
+    await expect(w.writeReport("x")).rejects.toThrow("SYMLINK_DENIED");
+  });
+  it("computes safe metrics", async () => {
+    const fixture = path.resolve("../../fixtures/ads-account.json");
+    const x = await adsMetrics(fixture, "demo-account", 7);
+    expect(x.roas).toBe(3);
+    expect(x.days).toBe(7);
+  });
+  it("rejects private model URLs", async () =>
+    await expect(validateBaseUrl("https://127.0.0.1/v1")).rejects.toThrow(
+      "BASE_URL_PRIVATE_DENIED",
+    ));
+  it("bounds SW Ads semantic queries", () => {
+    const result = boundSwadsArguments("metrics_semantic_query", {
+      query: { limit: 1000, metrics: ["spend"] },
+    });
+    expect((result.query as { limit: number }).limit).toBe(50);
+  });
+  it("compacts large SW Ads row results", () => {
+    const result = JSON.parse(
+      compactSwadsText(
+        JSON.stringify({ rows: Array.from({ length: 80 }, (_, i) => [i]) }),
+      ),
+    ) as { rows: unknown[]; response_compaction: { original_rows: number } };
+    expect(result.rows).toHaveLength(50);
+    expect(result.response_compaction.original_rows).toBe(80);
+  });
+});