niceeval 0.6.1 → 0.6.2

package/src/agents/coding-cli-versions.test.ts

@@ -0,0 +1,15 @@
+// cases: docs/engineering/unit-tests/adapters/cases.md
+import { readFile } from "node:fs/promises";
+import { describe, expect, it } from "vitest";
+import {
+  DEFAULT_CLAUDE_CODE_CLI_VERSION,
+  DEFAULT_CODEX_CLI_VERSION,
+} from "./coding-cli-versions.ts";
+
+describe("official coding-agent versions", () => {
+  it("keeps the non-TypeScript Docker recipe in sync", async () => {
+    const dockerfile = await readFile(new URL("../../sandbox/docker/Dockerfile", import.meta.url), "utf8");
+    expect(dockerfile).toContain(`ARG CODEX_VERSION=${DEFAULT_CODEX_CLI_VERSION}`);
+    expect(dockerfile).toContain(`ARG CLAUDE_CODE_VERSION=${DEFAULT_CLAUDE_CODE_CLI_VERSION}`);
+  });
+});

package/src/agents/coding-cli-versions.ts

@@ -0,0 +1,3 @@
+/** Coding-agent CLI versions used by runtime fallback installs and official sandbox recipes. */
+export const DEFAULT_CODEX_CLI_VERSION = "0.144.1";
+export const DEFAULT_CLAUDE_CODE_CLI_VERSION = "2.1.207";

package/src/agents/index.ts

@@ -4,6 +4,11 @@ export { defineAgent, defineSandboxAgent } from "../define.ts";
 export { shared } from "./shared.ts";
 export type { Shared } from "./shared.ts";
 
+// 证据覆盖声明:官方 SDK 适配器声明全通道 complete 用 completeCoverage;
+// 手写映射按实际情况声明(见 docs/feature/adapters/architecture/evidence.md)。
+export { completeCoverage } from "../scoring/coverage.ts";
+export type { CoverageStatus, CoverageDeclaration, EvidenceCoverage } from "../types.ts";
+
 // span → canonical GenAI 归一(只服务瀑布图,不喂断言)。私有埋点写自己的 spanMapper 时用:
 // tagSpan 把判定写回 span(原属性只增不改),heuristicTag 是通用兜底判定;mapCodexSpans 是
 // 现成的参考实现(无侵入接 codex 后端时直接声明 `spanMapper: mapCodexSpans`)。
@@ -43,6 +48,10 @@ export type {
   CodexThreadStream,
 } from "./sdk-streams.ts";
 
+// LangGraph 官方事件流转换器(不绑定 transport,不提供 langGraphAgent 工厂)。
+export { fromLangGraphEvents } from "./langgraph.ts";
+export type { LangGraphEventLike, LangGraphContentBlockLike, LangGraphStream } from "./langgraph.ts";
+
 // 通用「拼装方式」件:逐帧驱动循环、逐 token/参数增量累加器。见 docs-site/zh/guides/write-send.mdx——
 // 这些和任何具体协议无关,自己写 adapter 时优先拿这些拼,只有 transport(怎么发)与
 // 「帧类型 → 操作」这张映射表才是真正要手写的。会话续接与 HITL 停轮现场不再是可选件,
@@ -71,9 +80,11 @@ export { BUILTIN_AGENTS } from "./builtin.ts";
 export { claudeCodeAgent } from "./claude-code.ts";
 export { codexAgent } from "./codex.ts";
 export { bubAgent } from "./bub.ts";
+export { openClawAgent } from "./openclaw.ts";
 export type { ClaudeCodeConfig, ClaudeCodePluginSpec } from "./claude-code.ts";
 export type { CodexConfig, CodexPluginSpec } from "./codex.ts";
 export type { BubConfig, PythonPluginSpec } from "./bub.ts";
+export type { OpenClawConfig } from "./openclaw.ts";
 
 // 安装 manifest 的落点:adapter 写(shared.writeAgentSetup),运行器读并抬成 attempt artifact。
 export { AGENT_SETUP_MANIFEST_PATH } from "./manifest.ts";

package/src/agents/langgraph.test.ts

@@ -0,0 +1,204 @@
+// cases: docs/engineering/unit-tests/adapters/cases.md
+import { describe, expect, it } from "vitest";
+
+import { fromLangGraphEvents } from "./langgraph.ts";
+
+describe("fromLangGraphEvents", () => {
+  it("messages+tools 全链:content blocks → 事件,tools started/finished 按 call id 配对,usage 在 finish 上", () => {
+    const s = fromLangGraphEvents();
+
+    // messages finish:reasoning / text / tool_call blocks 按原始顺序落事件
+    expect(
+      s.add({
+        channel: "messages",
+        event: "finish",
+        data: {
+          message: {
+            role: "assistant",
+            content: [
+              { type: "reasoning", reasoning: "先查天气" },
+              { type: "text", text: "我来查一下" },
+              { type: "tool_call", id: "call_1", name: "get_weather", args: { city: "北京" } },
+            ],
+            usage_metadata: { input_tokens: 100, output_tokens: 20 },
+          },
+        },
+      }),
+    ).toEqual([
+      { type: "thinking", text: "先查天气" },
+      { type: "message", role: "assistant", text: "我来查一下" },
+      { type: "action.called", callId: "call_1", name: "get_weather", input: { city: "北京" }, tool: "unknown" },
+    ]);
+    expect(s.usage).toEqual({ inputTokens: 100, outputTokens: 20, requests: 1 });
+
+    // tools/started 描述同一次调用:按 call ID 去重,不重发 action.called
+    expect(
+      s.add({ channel: "tools", event: "started", data: { id: "call_1", name: "get_weather", input: { city: "北京" } } }),
+    ).toEqual([]);
+
+    expect(s.add({ channel: "tools", event: "finished", data: { id: "call_1", output: { temp: 21 } } })).toEqual([
+      { type: "action.result", callId: "call_1", output: { temp: 21 }, status: "completed" },
+    ]);
+
+    expect(s.add({ channel: "lifecycle", event: "completed" })).toEqual([]);
+    expect(s.status).toBe("completed");
+  });
+
+  it("messages partial(逐 token 增量)整个忽略;user 消息也落 message 事件", () => {
+    const s = fromLangGraphEvents();
+    expect(s.add({ channel: "messages", event: "partial", data: { message: { role: "assistant", content: "我" } } })).toEqual([]);
+    expect(s.add({ channel: "messages", event: "finish", data: { message: { role: "user", content: "东京天气" } } })).toEqual([
+      { type: "message", role: "user", text: "东京天气" },
+    ]);
+    expect(s.usage).toBeUndefined(); // 协议没报 usage 就不编造
+  });
+
+  it("tools/started 单独出现也成对;error → failed,markRejected 后 → rejected", () => {
+    const s = fromLangGraphEvents();
+    expect(s.add({ channel: "tools", event: "started", data: { id: "c1", name: "search_docs", input: { q: "x" } } })).toEqual([
+      { type: "action.called", callId: "c1", name: "search_docs", input: { q: "x" }, tool: "unknown" },
+    ]);
+    expect(s.add({ channel: "tools", event: "error", data: { id: "c1", error: "boom" } })).toEqual([
+      { type: "action.result", callId: "c1", output: "boom", status: "failed" },
+    ]);
+
+    s.add({ channel: "tools", event: "started", data: { id: "c2", name: "send_email", input: {} } });
+    s.markRejected("c2");
+    expect(s.add({ channel: "tools", event: "error", data: { id: "c2", error: "denied" } })).toEqual([
+      { type: "action.result", callId: "c2", output: "denied", status: "rejected" },
+    ]);
+  });
+
+  it("interrupt → input.requested + waiting;HITL 请求形状映射 action/input/display/options", () => {
+    const s = fromLangGraphEvents();
+    const events = s.add({
+      channel: "lifecycle",
+      event: "interrupted",
+      data: {
+        interrupts: [
+          {
+            id: "int_1",
+            value: {
+              action_request: { action: "send_email", args: { to: "a@b.c" } },
+              description: "需要人工批准发送邮件",
+              config: { allow_accept: true, allow_ignore: true },
+            },
+          },
+        ],
+      },
+    });
+    expect(events).toEqual([
+      {
+        type: "input.requested",
+        request: {
+          id: "int_1",
+          action: "send_email",
+          input: { to: "a@b.c" },
+          display: "需要人工批准发送邮件",
+          options: [{ id: "accept" }, { id: "ignore" }],
+        },
+      },
+    ]);
+    expect(s.status).toBe("waiting");
+
+    // 同一 interrupt 也从 input 通道到达:按 id 去重,只产一条请求事件
+    expect(s.add({ channel: "input", data: { id: "int_1", value: "approve?" } })).toEqual([]);
+  });
+
+  it("input 通道的字符串 interrupt value → prompt", () => {
+    const s = fromLangGraphEvents();
+    expect(s.add({ channel: "input", data: { id: "q1", value: "要继续吗?" } })).toEqual([
+      { type: "input.requested", request: { id: "q1", prompt: "要继续吗?" } },
+    ]);
+  });
+
+  it("lifecycle failed → status failed + error 事件", () => {
+    const s = fromLangGraphEvents();
+    expect(s.add({ channel: "lifecycle", event: "failed", data: { error: { message: "graph exploded" } } })).toEqual([
+      { type: "error", message: "graph exploded" },
+    ]);
+    expect(s.status).toBe("failed");
+  });
+
+  it("subgraph namespace → subagent 层级:首见补 called,该层 lifecycle 闭合成 completed", () => {
+    const s = fromLangGraphEvents();
+
+    // 嵌套 namespace 的首个事件:自外向内逐级补 subagent.called(段内 : 后缀不进展示名)
+    const events = s.add({
+      channel: "tools",
+      event: "started",
+      namespace: ["research:ckpt-1", "web:ckpt-2"],
+      data: { id: "t1", name: "fetch_page", input: { url: "https://x" } },
+    });
+    expect(events).toEqual([
+      { type: "subagent.called", callId: "research:ckpt-1", name: "research" },
+      { type: "subagent.called", callId: "research:ckpt-1/web:ckpt-2", name: "web" },
+      { type: "action.called", callId: "t1", name: "fetch_page", input: { url: "https://x" }, tool: "unknown" },
+    ]);
+
+    // 内层 lifecycle completed 只闭合内层
+    expect(s.add({ channel: "lifecycle", event: "completed", namespace: ["research:ckpt-1", "web:ckpt-2"] })).toEqual([
+      { type: "subagent.completed", callId: "research:ckpt-1/web:ckpt-2", status: "completed" },
+    ]);
+
+    // 根图 completed:仍未闭合的层级一起闭合
+    expect(s.add({ channel: "lifecycle", event: "completed" })).toEqual([
+      { type: "subagent.completed", callId: "research:ckpt-1", status: "completed" },
+    ]);
+    expect(s.status).toBe("completed");
+  });
+
+  it("subgraph lifecycle failed 闭合成 failed,并先闭合更深的打开层级", () => {
+    const s = fromLangGraphEvents();
+    s.add({ channel: "messages", event: "finish", namespace: ["a", "b"], data: { message: { role: "assistant", content: "hi" } } });
+    expect(s.add({ channel: "lifecycle", event: "failed", namespace: ["a"], data: { error: "sub failed" } })).toEqual([
+      { type: "subagent.completed", callId: "a/b", status: "failed" },
+      { type: "subagent.completed", callId: "a", status: "failed" },
+    ]);
+  });
+
+  it("seq 乱序帧:超前暂存,缺口补齐按 seq 顺序放出;落后的重复帧丢弃", () => {
+    const s = fromLangGraphEvents();
+    const msg = (seq: number, text: string) => ({
+      seq,
+      channel: "messages" as const,
+      event: "finish",
+      data: { message: { role: "assistant", content: text } },
+    });
+
+    expect(s.add(msg(1, "一"))).toEqual([{ type: "message", role: "assistant", text: "一" }]);
+    // seq 3 超前:暂存
+    expect(s.add(msg(3, "三"))).toEqual([]);
+    // seq 2 补上缺口:2、3 按协议顺序一起放出
+    expect(s.add(msg(2, "二"))).toEqual([
+      { type: "message", role: "assistant", text: "二" },
+      { type: "message", role: "assistant", text: "三" },
+    ]);
+    // 重连补发的旧帧(seq 已消费):丢弃
+    expect(s.add(msg(2, "二"))).toEqual([]);
+  });
+
+  it("end() 放出 seq 缺口后仍压着的乱序帧", () => {
+    const s = fromLangGraphEvents();
+    const msg = (seq: number, text: string) => ({
+      seq,
+      channel: "messages" as const,
+      event: "finish",
+      data: { message: { role: "assistant", content: text } },
+    });
+    s.add(msg(1, "一"));
+    expect(s.add(msg(4, "四"))).toEqual([]); // seq 2、3 永远没来
+    expect(s.add(msg(3, "三"))).toEqual([]);
+    expect(s.end()).toEqual([
+      { type: "message", role: "assistant", text: "三" },
+      { type: "message", role: "assistant", text: "四" },
+    ]);
+  });
+
+  it("values / updates 等其它通道与无 channel 帧返回 []", () => {
+    const s = fromLangGraphEvents();
+    expect(s.add({ channel: "values", data: { messages: [] } })).toEqual([]);
+    expect(s.add({ channel: "updates", data: {} })).toEqual([]);
+    expect(s.add({ data: {} })).toEqual([]);
+  });
+});