@downcity/agent 1.1.232 → 1.1.236

package/scripts/core-engine-context-compaction.test.mjs

@@ -0,0 +1,411 @@
+/**
+ * @file 验证 CoreEngine 基于真实 usage 的触发水位与 part/tool transaction 深度折叠。
+ */
+
+import assert from "node:assert/strict";
+import test from "node:test";
+import { MockLanguageModelV3 } from "ai/test";
+
+import {
+  deep_compact_model_messages,
+  resolve_model_usage_ratio,
+  resolve_model_usage_tokens,
+  should_compact_after_usage,
+} from "../bin/executor/core-engine/CoreEngineContextCompaction.js";
+import { CoreEngineRunner } from "../bin/executor/core-engine/CoreEngineRunner.js";
+import { LocalSessionContextComposer } from "../bin/executor/composer/context/LocalSessionContextComposer.js";
+
+function create_stream_text_result(text, input_tokens, output_tokens) {
+  return {
+    stream: new ReadableStream({
+      start(controller) {
+        controller.enqueue({ type: "stream-start", warnings: [] });
+        controller.enqueue({ type: "text-start", id: "text-1" });
+        controller.enqueue({ type: "text-delta", id: "text-1", delta: text });
+        controller.enqueue({ type: "text-end", id: "text-1" });
+        controller.enqueue({
+          type: "finish",
+          finishReason: { unified: "stop", raw: "stop" },
+          usage: {
+            inputTokens: {
+              total: input_tokens,
+              noCache: input_tokens,
+              cacheRead: 0,
+              cacheWrite: 0,
+            },
+            outputTokens: {
+              total: output_tokens,
+              text: output_tokens,
+              reasoning: 0,
+            },
+          },
+        });
+        controller.close();
+      },
+    }),
+  };
+}
+
+function create_runner() {
+  const context_composer = new LocalSessionContextComposer({
+    sessionId: "compact-runner-session",
+    getTools: () => ({}),
+  });
+  return new CoreEngineRunner({
+    history_store: { sessionId: "compact-runner-session" },
+    context_composer,
+    logger: { log: async () => {} },
+    should_compact_on_error: () => false,
+  });
+}
+
+function create_context_error_runner() {
+  const context_composer = new LocalSessionContextComposer({
+    sessionId: "compact-runner-session",
+    getTools: () => ({}),
+  });
+  return new CoreEngineRunner({
+    history_store: { sessionId: "compact-runner-session" },
+    context_composer,
+    logger: { log: async () => {} },
+    should_compact_on_error: (error) =>
+      String(error || "").includes("context length"),
+  });
+}
+
+function create_run_input(model, messages, context_window = 100) {
+  return {
+    execute_input: {
+      query: "latest request",
+      system: [],
+      messages,
+      tools: {},
+    },
+    model,
+    run_context: {
+      sessionId: "compact-runner-session",
+      injectedUserMessages: [],
+      deferredPersistedUserMessages: [],
+      pendingAssistantFileParts: [],
+    },
+    resolve_step_inputs: async () => ({
+      model,
+      system: [],
+      tools: {},
+      context_window,
+    }),
+  };
+}
+
+test("usage 优先读取 totalTokens，并在缺失时回退 input + output", () => {
+  assert.equal(
+    resolve_model_usage_tokens({
+      totalTokens: 95,
+      inputTokens: 80,
+      outputTokens: 10,
+    }),
+    95,
+  );
+  assert.equal(
+    resolve_model_usage_tokens({ inputTokens: 80, outputTokens: 15 }),
+    95,
+  );
+  assert.equal(resolve_model_usage_tokens({}), null);
+});
+
+test("普通调用使用 95% 触发，compact 验收使用 50% 目标", () => {
+  assert.equal(resolve_model_usage_ratio({ totalTokens: 94 }, 100), 0.94);
+  assert.equal(should_compact_after_usage(0.94, false), false);
+  assert.equal(should_compact_after_usage(0.95, false), true);
+  assert.equal(should_compact_after_usage(0.5, true), false);
+  assert.equal(should_compact_after_usage(0.5001, true), true);
+});
+
+test("最终 step 达到 95% 时通过 run result 请求 writer 收口后持久化 compact", async () => {
+  const model = new MockLanguageModelV3({
+    modelId: "usage-trigger-model",
+    doStream: async () => create_stream_text_result("done", 90, 5),
+  });
+  const result = await create_runner().run(create_run_input(model, [{
+    id: "user-1",
+    role: "user",
+    metadata: {
+      v: 1,
+      ts: 1,
+      sessionId: "compact-runner-session",
+      source: "ingress",
+      kind: "normal",
+    },
+    parts: [{ type: "text", text: "latest request" }],
+  }]));
+  assert.equal(result.success, true);
+  assert.equal(result.compact_required, true);
+});
+
+test("新的持久化 Summary 只按 50% 水位验收一次", async () => {
+  const model = new MockLanguageModelV3({
+    modelId: "usage-validation-model",
+    doStream: async () => create_stream_text_result("done", 55, 5),
+  });
+  const runner = create_runner();
+  const messages = [{
+    id: "summary-1",
+    role: "assistant",
+    metadata: {
+      v: 1,
+      ts: 1,
+      sessionId: "compact-runner-session",
+      source: "compact",
+      kind: "summary",
+    },
+    parts: [{ type: "text", text: "previous checkpoint" }],
+  }, {
+    id: "user-1",
+    role: "user",
+    metadata: {
+      v: 1,
+      ts: 2,
+      sessionId: "compact-runner-session",
+      source: "ingress",
+      kind: "normal",
+    },
+    parts: [{ type: "text", text: "latest request" }],
+  }];
+  const first = await runner.run(create_run_input(model, messages));
+  const second = await runner.run(create_run_input(model, messages));
+  assert.equal(first.compact_required, true);
+  assert.equal(second.compact_required, undefined);
+});
+
+test("显式 compact 后在下一次 provider 调用前重载 canonical history", async () => {
+  const provider_prompts = [];
+  const compacted_records = [{
+    id: "summary-reloaded",
+    role: "assistant",
+    metadata: {
+      v: 1,
+      ts: 2,
+      sessionId: "compact-runner-session",
+      source: "compact",
+      kind: "summary",
+    },
+    parts: [{ type: "text", text: "compacted checkpoint" }],
+  }];
+  const context_composer = new LocalSessionContextComposer({
+    sessionId: "compact-runner-session",
+    getTools: () => ({}),
+  });
+  const runner = new CoreEngineRunner({
+    history_store: {
+      sessionId: "compact-runner-session",
+      list_records: async () => compacted_records,
+    },
+    context_composer,
+    logger: { log: async () => {} },
+    should_compact_on_error: () => false,
+  });
+  let reload_requested = true;
+  const model = new MockLanguageModelV3({
+    modelId: "history-reload-model",
+    doStream: async (options) => {
+      provider_prompts.push(JSON.stringify(options.prompt));
+      return create_stream_text_result("done", 20, 5);
+    },
+  });
+  const input = create_run_input(model, [{
+    id: "old-user",
+    role: "user",
+    metadata: {
+      v: 1,
+      ts: 1,
+      sessionId: "compact-runner-session",
+      source: "ingress",
+      kind: "normal",
+    },
+    parts: [{ type: "text", text: "history before compact" }],
+  }]);
+  input.run_context.consume_history_reload = () => {
+    const requested = reload_requested;
+    reload_requested = false;
+    return requested;
+  };
+
+  const result = await runner.run(input);
+
+  assert.equal(result.success, true);
+  assert.equal(provider_prompts.length, 1);
+  assert.match(provider_prompts[0], /compacted checkpoint/);
+  assert.doesNotMatch(provider_prompts[0], /history before compact/);
+});
+
+test("Provider context-length error 在当前 tool-loop 内 deep compact 后重试", async () => {
+  let provider_calls = 0;
+  const model = new MockLanguageModelV3({
+    modelId: "context-error-model",
+    doStream: async () => {
+      provider_calls += 1;
+      if (provider_calls === 1) throw new Error("context length exceeded");
+      return create_stream_text_result("recovered", 40, 5);
+    },
+  });
+  const result = await create_context_error_runner().run(
+    create_run_input(model, [{
+      id: "user-1",
+      role: "user",
+      metadata: {
+        v: 1,
+        ts: 1,
+        sessionId: "compact-runner-session",
+        source: "ingress",
+        kind: "normal",
+      },
+      parts: [{ type: "text", text: "latest request" }],
+    }]),
+  );
+  assert.equal(result.success, true);
+  assert.equal(result.compact_required, true);
+  assert.equal(provider_calls, 2);
+});
+
+test("deep compact 删除 reasoning，并完整保留最新并行 tool transaction", () => {
+  const large_output = "tool-output-".repeat(8_000);
+  const messages = [
+    { role: "user", content: "older request" },
+    {
+      role: "assistant",
+      content: [
+        { type: "reasoning", text: "old reasoning".repeat(2_000) },
+        {
+          type: "tool-call",
+          toolCallId: "old-call",
+          toolName: "old_tool",
+          input: { query: "old" },
+        },
+      ],
+    },
+    {
+      role: "tool",
+      content: [
+        {
+          type: "tool-result",
+          toolCallId: "old-call",
+          toolName: "old_tool",
+          output: { type: "text", value: large_output },
+        },
+      ],
+    },
+    { role: "user", content: "latest request" },
+    {
+      role: "assistant",
+      content: [
+        { type: "reasoning", text: "latest reasoning".repeat(2_000) },
+        { type: "text", text: "running tools" },
+        {
+          type: "tool-call",
+          toolCallId: "call-a",
+          toolName: "tool_a",
+          input: { value: "a" },
+        },
+        {
+          type: "tool-call",
+          toolCallId: "call-b",
+          toolName: "tool_b",
+          input: { value: "b" },
+        },
+      ],
+    },
+    {
+      role: "tool",
+      content: [
+        {
+          type: "tool-result",
+          toolCallId: "call-a",
+          toolName: "tool_a",
+          output: { type: "text", value: large_output },
+        },
+        {
+          type: "tool-result",
+          toolCallId: "call-b",
+          toolName: "tool_b",
+          output: { type: "json", value: { large_output } },
+        },
+      ],
+    },
+  ];
+
+  const compacted = deep_compact_model_messages(messages, 0);
+  const serialized = JSON.stringify(compacted);
+  assert.equal(serialized.includes("old reasoning"), false);
+  assert.equal(serialized.includes("latest reasoning"), false);
+  assert.equal(serialized.includes("latest request"), true);
+  assert.equal(serialized.includes("call-a"), true);
+  assert.equal(serialized.includes("call-b"), true);
+  assert.ok(serialized.length < JSON.stringify(messages).length / 2);
+
+  const tool_call_ids = new Set();
+  const tool_result_ids = new Set();
+  const active_tool_text = [];
+  for (const message of compacted) {
+    if (!Array.isArray(message.content)) continue;
+    for (const part of message.content) {
+      if (part.type === "tool-call") tool_call_ids.add(part.toolCallId);
+      if (part.type === "tool-result") tool_result_ids.add(part.toolCallId);
+      if (part.type === "tool-call" || part.type === "tool-result") {
+        active_tool_text.push(JSON.stringify(part));
+      }
+    }
+  }
+  assert.equal(active_tool_text.join("\n").includes("old-call"), false);
+  assert.deepEqual([...tool_call_ids].sort(), ["call-a", "call-b"]);
+  assert.deepEqual([...tool_result_ids].sort(), ["call-a", "call-b"]);
+});
+
+test("单条 assistant 含大量 parts 时也会在消息内部折叠", () => {
+  const messages = [{
+    role: "assistant",
+    content: Array.from({ length: 40 }, (_, index) => ({
+      type: "text",
+      text: `part-${String(index)}:${"x".repeat(4_000)}`,
+    })),
+  }];
+  const compacted = deep_compact_model_messages(messages, 0);
+  assert.equal(compacted.length, 1);
+  assert.ok(JSON.stringify(compacted).length < JSON.stringify(messages).length / 4);
+});
+
+test("tool approval request/response 与对应 call 一起保留且不产生孤儿", () => {
+  const compacted = deep_compact_model_messages([
+    { role: "user", content: "approve the operation" },
+    {
+      role: "assistant",
+      content: [
+        {
+          type: "tool-call",
+          toolCallId: "approval-call",
+          toolName: "dangerous_tool",
+          input: { path: "/tmp/example" },
+        },
+        {
+          type: "tool-approval-request",
+          toolCallId: "approval-call",
+          approvalId: "approval-1",
+        },
+      ],
+    },
+    {
+      role: "tool",
+      content: [
+        {
+          type: "tool-approval-response",
+          approvalId: "approval-1",
+          approved: true,
+          providerExecuted: true,
+        },
+      ],
+    },
+  ]);
+  const serialized = JSON.stringify(compacted);
+  assert.equal(serialized.includes('"toolCallId":"approval-call"'), true);
+  assert.equal(serialized.includes('"approvalId":"approval-1"'), true);
+  assert.equal(serialized.includes('"approved":true'), true);
+});

package/scripts/logger-isolation.test.mjs

@@ -0,0 +1,55 @@
+/**
+ * @file 验证 logger 实例不会跨 workspace 共享落盘目录。
+ *
+ * 关键点（中文）
+ * - 每次 getLogger 调用都应返回独立实例。
+ * - 两个实例并发写日志时，只能写入各自项目的 `.downcity/logs`。
+ */
+
+import test from "node:test";
+import assert from "node:assert/strict";
+import os from "node:os";
+import path from "node:path";
+import fs from "node:fs/promises";
+
+import { getLogger } from "../bin/index.js";
+
+/**
+ * 读取项目日志目录内的全部 JSONL 文本。
+ */
+async function read_project_logs(project_root) {
+  const logs_dir = path.join(project_root, ".downcity", "logs");
+  const files = await fs.readdir(logs_dir);
+  const jsonl_files = files.filter((file) => file.endsWith(".jsonl"));
+  const chunks = await Promise.all(
+    jsonl_files.map((file) => fs.readFile(path.join(logs_dir, file), "utf8")),
+  );
+  return chunks.join("\n");
+}
+
+test("getLogger keeps concurrent workspace logs isolated", async () => {
+  const root_a = await fs.mkdtemp(
+    path.join(os.tmpdir(), "downcity-logger-isolation-a-"),
+  );
+  const root_b = await fs.mkdtemp(
+    path.join(os.tmpdir(), "downcity-logger-isolation-b-"),
+  );
+  const logger_a = getLogger(root_a);
+  const logger_b = getLogger(root_b);
+
+  assert.notEqual(logger_a, logger_b);
+  await Promise.all([
+    logger_a.log("info", "workspace_a_only"),
+    logger_b.log("info", "workspace_b_only"),
+  ]);
+  await Promise.all([logger_a.saveAllLogs(), logger_b.saveAllLogs()]);
+
+  const [logs_a, logs_b] = await Promise.all([
+    read_project_logs(root_a),
+    read_project_logs(root_b),
+  ]);
+  assert.match(logs_a, /workspace_a_only/);
+  assert.doesNotMatch(logs_a, /workspace_b_only/);
+  assert.match(logs_b, /workspace_b_only/);
+  assert.doesNotMatch(logs_b, /workspace_a_only/);
+});