@deepseek-ai/dsh-llm-mock-server 0.1.6-alpha.2 → 0.1.7-alpha.1

package/README.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write packages/test-support/llm-mock-server/README.md
-README.md: be311d9c83bcd35ad5bd0c003f6e95861b6d6976
-README.zh.md: 7a022131d9ef09d9ef89c91cae418f093f77aca1
+README.md: 2f83cf66c02728820a4f8aa6f14b1a091d092626
+README.zh.md: ad85853fb4a20c3ccdb444508da8032c18d7538a

package/README.md

@@ -1,5 +1,5 @@
 ---
-description: "Scriptable OpenAI-compatible fault server for testing LLM adapters and recovery policy without a provider key, for test authors and demos."
+description: "Scriptable Messages-compatible fault server for testing LLM adapters and recovery policy without a provider key, for test authors and demos."
 kind: "package-library"
 ---
 
@@ -9,7 +9,7 @@ English | [中文](README.zh.md)
 
 ## Summary
 
-This package gives tests and demos a scriptable OpenAI-compatible HTTP/SSE endpoint, so they can exercise model-provider failures and successes without a provider key. Each accepted `/chat/completions` request consumes the next scripted behavior, including resets, stalls, malformed chunks, rate limits, server errors, completions, and tool calls. Test authors can run it with `pnpm run mock:llm` or call `startMockLlmServer`, which returns captured requests for assertions. Seeded `random` behavior supports reproducible mixed-failure stress runs.
+This package gives tests and demos a scriptable Messages-compatible HTTP/SSE endpoint, so they can exercise model-provider failures and successes without a provider key. Each accepted `/v1/messages` request consumes the next scripted behavior, including resets, stalls, malformed chunks, rate limits, server errors, completions, and tool calls. Test authors can run it with `pnpm run mock:llm` or call `startMockLlmServer`, which returns captured requests for assertions. Seeded `random` behavior supports reproducible mixed-failure stress runs.
 
 ## Table of Contents
 
@@ -39,7 +39,7 @@ pnpm run mock:llm \
   --partial-text "discard this half"
 ```
 
-Point the shipping DeepSeek adapter at the server; it appends `/chat/completions` to the configured base:
+Point the shipping DeepSeek adapter at the server; it sends requests to `/messages` beneath the server's `/v1` root:
 
 ```sh
 DEEPSEEK_BASE_URL=http://127.0.0.1:8000/v1 \
@@ -88,13 +88,13 @@ Omitting `--seed` generates one and prints it in the `ready` record. `--random-w
 
 ### Timing and content controls
 
-The CLI exposes `--success-text`, `--partial-text`, `--reasoning-text`, `--chunk-size`, `--chunk-delay-ms`, `--disconnect-delay-ms`, `--retry-after-ms`, `--request-id`, `--tool-name`, and `--tool-arguments`. Millisecond delays are bounded integers within Node's timer range; `retryAfterMs` must also be positive. The library accepts the same camel-case options. An optional exact `apiKey` validates `Authorization: Bearer <token>`; omission accepts any token.
+The CLI exposes `--success-text`, `--partial-text`, `--reasoning-text`, `--chunk-size`, `--chunk-delay-ms`, `--disconnect-delay-ms`, `--retry-after-ms`, `--request-id`, `--tool-name`, and `--tool-arguments`. Millisecond delays are bounded integers within Node's timer range; `retryAfterMs` must also be positive. The library accepts the same camel-case options. An optional exact `apiKey` validates `x-api-key: <token>`; omission accepts any token.
 
 ### What can go wrong
 
 - **The script runs out** — exhaustion returns a structured HTTP 500; set `--repeat-last` or lengthen the sequence when a run needs more requests.
 - **Random weights without a positive concrete behavior are rejected** — every entry must name an existing behavior and at least one must carry positive weight.
-- **Invalid requests do not consume the script** — wrong methods, paths, bearer tokens, and malformed JSON get ordinary 4xx responses, so a misconfigured client can burn retries without advancing the sequence.
+- **Invalid requests do not consume the script** — wrong methods, paths, API keys, and malformed JSON get ordinary 4xx responses, so a misconfigured client can burn retries without advancing the sequence.
 
 -----
 
@@ -108,7 +108,7 @@ This section explains the design of the server; the observable behavior is fully
 
 ### Design
 
-The server is built on one rule: each accepted chat-completions request consumes exactly one behavior from an arrival-ordered FIFO cursor, and the server never retries or interprets harness policy. Validation happens before the cursor advances — only a `POST` whose path ends in `/chat/completions`, with a valid bearer token when one is configured and a parseable JSON body, consumes the script; everything else receives an ordinary 4xx. `random` entries resolve at request time through a seeded PRNG over the configured weights, so a run is reproducible from its printed seed.
+The server is built on one rule: each accepted Messages request consumes exactly one behavior from an arrival-ordered FIFO cursor, and the server never retries or interprets harness policy. Validation happens before the cursor advances — only a `POST` whose path ends in `/v1/messages`, with a valid API key when one is configured and a parseable JSON body, consumes the script; everything else receives an ordinary 4xx. `random` entries resolve at request time through a seeded PRNG over the configured weights, so a run is reproducible from its printed seed.
 
 ### Source map
 

package/README.zh.md

@@ -1,5 +1,5 @@
 ---
-description: "用于在无提供方密钥的情况下测试 LLM（大语言模型）适配器与恢复策略的可通过脚本控制的 OpenAI 兼容故障服务器，面向测试作者与演示。"
+description: "用于在无提供方密钥的情况下测试 LLM（大语言模型）适配器与恢复策略的可通过脚本控制的 Messages 兼容故障服务器，面向测试作者与演示。"
 kind: "package-library"
 ---
 
@@ -9,7 +9,7 @@ kind: "package-library"
 
 ## 概述
 
-本包为测试与演示提供可编脚本的 OpenAI 兼容 HTTP／SSE（Server-Sent Events）端点，使其无需提供方密钥即可检验模型提供方的失败与成功。每个已接受的 `/chat/completions` 请求依次消费下一个脚本行为，包括重置、停滞、畸形分片、限流、服务器错误、补全与工具调用。测试作者可以通过 `pnpm run mock:llm` 运行服务器，也可以调用 `startMockLlmServer`，后者会返回捕获的请求供断言使用。带种子的 `random` 行为支持可复现的混合故障压力运行。
+本包为测试与演示提供可编脚本的 Messages 兼容 HTTP／SSE（Server-Sent Events）端点，使其无需提供方密钥即可检验模型提供方的失败与成功。每个已接受的 `/v1/messages` 请求依次消费下一个脚本行为，包括重置、停滞、畸形分片、限流、服务器错误、补全与工具调用。测试作者可以通过 `pnpm run mock:llm` 运行服务器，也可以调用 `startMockLlmServer`，后者会返回捕获的请求供断言使用。带种子的 `random` 行为支持可复现的混合故障压力运行。
 
 ## 目录
 
@@ -39,7 +39,7 @@ pnpm run mock:llm \
   --partial-text "discard this half"
 ```
 
-将发布的 DeepSeek 适配器指向服务器；它会将 `/chat/completions` 追加到已配置 base：
+将发布的 DeepSeek 适配器指向服务器；它会向服务器 `/v1` 根路径下的 `/messages` 发送请求：
 
 ```sh
 DEEPSEEK_BASE_URL=http://127.0.0.1:8000/v1 \
@@ -88,13 +88,13 @@ pnpm run mock:llm \
 
 ### 时序与内容控制
 
-CLI 公开 `--success-text`、`--partial-text`、`--reasoning-text`、`--chunk-size`、`--chunk-delay-ms`、`--disconnect-delay-ms`、`--retry-after-ms`、`--request-id`、`--tool-name` 与 `--tool-arguments`。毫秒延迟是 Node 定时器范围内的有界整数；`retryAfterMs` 还必须为正数。库接受相同的 camel-case 选项。可选的 `apiKey` 会精确验证 `Authorization: Bearer <token>`；省略时接受任何 token。
+CLI 公开 `--success-text`、`--partial-text`、`--reasoning-text`、`--chunk-size`、`--chunk-delay-ms`、`--disconnect-delay-ms`、`--retry-after-ms`、`--request-id`、`--tool-name` 与 `--tool-arguments`。毫秒延迟是 Node 定时器范围内的有界整数；`retryAfterMs` 还必须为正数。库接受相同的 camel-case 选项。可选的 `apiKey` 会精确验证 `x-api-key: <token>`；省略时接受任何 token。
 
 ### 可能出什么问题
 
 - **脚本耗尽**——耗尽时返回结构化 HTTP 500；当一次运行需要更多请求时设置 `--repeat-last` 或加长序列。
 - **没有正权重具体行为的随机权重会被拒绝**——每个条目都必须命名现有行为，且至少一个条目带正权重。
-- **无效请求不消费脚本**——错误方法、路径、Bearer token 与畸形 JSON 会收到普通 4xx 响应，因此配置错误的客户端可能耗尽重试却不推进序列。
+- **无效请求不消费脚本**——错误方法、路径、API key 与畸形 JSON 会收到普通 4xx 响应，因此配置错误的客户端可能耗尽重试却不推进序列。
 
 -----
 
@@ -108,7 +108,7 @@ CLI 公开 `--success-text`、`--partial-text`、`--reasoning-text`、`--chunk-s
 
 ### 设计
 
-服务器建立在一个规则之上：每个已接受的 chat-completions 请求从按到达顺序排列的 FIFO 游标消费恰好一个行为，服务器从不重试或解读 harness 策略。校验先于游标推进——只有 `POST` 且路径以 `/chat/completions` 结尾、配置密钥时携带有效 Bearer token、且 JSON 正文可解析的请求才消费脚本；其余请求都收到普通 4xx。`random` 条目在请求时通过带种子的 PRNG 按配置权重解析，因此一次运行可由其打印出的种子复现。
+服务器建立在一个规则之上：每个已接受的 Messages 请求从按到达顺序排列的 FIFO 游标消费恰好一个行为，服务器从不重试或解读 harness 策略。校验先于游标推进——只有 `POST` 且路径以 `/v1/messages` 结尾、配置密钥时携带有效 API key、且 JSON 正文可解析的请求才消费脚本；其余请求都收到普通 4xx。`random` 条目在请求时通过带种子的 PRNG 按配置权重解析，因此一次运行可由其打印出的种子复现。
 
 ### 源码地图
 

package/lib/index.js

@@ -4,8 +4,8 @@ import { isIP } from "node:net";
 import { setTimeout } from "node:timers/promises";
 //#region lib/types/index.js
 /**
-* Scriptable OpenAI-compatible HTTP/SSE server for transport, protocol, and
-* semantic-empty LLM recovery tests. Each accepted chat-completions request
+* Scriptable Messages HTTP/SSE server for transport, protocol, and
+* semantic-empty LLM recovery tests. Each accepted Messages request
 * consumes one behavior; the server never retries or interprets harness policy.
 *
 * @module @deepseek-ai/dsh-llm-mock-server
@@ -153,7 +153,7 @@ function writeSse(record, response, payload) {
 	record.chunksSent += 1;
 }
 function writeDone(record, response) {
-	writeSse(record, response, "[DONE]");
+	writeSse(record, response, { type: "message_stop" });
 }
 function finishRecord(options, record, outcome) {
 	if (record.outcome !== void 0) return;
@@ -179,17 +179,40 @@ function httpError(options, record, response, status, message, code, type = "moc
 	} }));
 	finishRecord(options, record, "completed");
 }
+function startMessage(record, response) {
+	writeSse(record, response, {
+		type: "message_start",
+		message: {
+			id: "mock-message",
+			type: "message",
+			role: "assistant",
+			model: "mock-model",
+			content: [],
+			usage: {
+				input_tokens: 3,
+				output_tokens: 0
+			}
+		}
+	});
+}
+function startText(record, response, index) {
+	writeSse(record, response, {
+		type: "content_block_start",
+		index,
+		content_block: {
+			type: "text",
+			text: ""
+		}
+	});
+}
 function terminalChunk(reason, outputTokens) {
 	return {
-		choices: [{
-			index: 0,
-			delta: { content: "" },
-			finish_reason: reason
-		}],
-		usage: {
-			prompt_tokens: 3,
-			completion_tokens: outputTokens
-		}
+		type: "message_delta",
+		delta: {
+			stop_reason: reason,
+			stop_sequence: null
+		},
+		usage: { output_tokens: outputTokens }
 	};
 }
 async function pause(milliseconds, response) {
@@ -208,22 +231,30 @@ async function pause(milliseconds, response) {
 		response.off("close", stop);
 	}
 }
-async function streamText(options, record, response, text, delayMs) {
+async function streamText(options, record, response, text, delayMs, index = 0) {
 	for (const chunk of splitText(text, options.chunkSize)) {
-		writeSse(record, response, { choices: [{
-			index: 0,
-			delta: { content: chunk },
-			finish_reason: null
-		}] });
+		writeSse(record, response, {
+			type: "content_block_delta",
+			index,
+			delta: {
+				type: "text_delta",
+				text: chunk
+			}
+		});
 		if (!await pause(delayMs, response)) return false;
 	}
 	return true;
 }
-async function completeText(options, record, response, reason, delayMs) {
-	if (!await streamText(options, record, response, options.successText, delayMs)) {
+async function completeText(options, record, response, reason, delayMs, index = 0) {
+	startText(record, response, index);
+	if (!await streamText(options, record, response, options.successText, delayMs, index)) {
 		finishRecord(options, record, "client_closed");
 		return;
 	}
+	writeSse(record, response, {
+		type: "content_block_stop",
+		index
+	});
 	writeSse(record, response, terminalChunk(reason, Array.from(options.successText).length));
 	writeDone(record, response);
 	response.end();
@@ -239,26 +270,30 @@ async function disconnect(options, record, response) {
 }
 function toolCallChunks(options) {
 	const midpoint = Math.max(1, Math.floor(options.toolArguments.length / 2));
-	return [{ choices: [{
-		index: 0,
-		delta: { tool_calls: [{
+	return [
+		{
+			type: "content_block_start",
 			index: 0,
-			id: "mock-call-1",
-			type: "function",
-			function: {
+			content_block: {
+				type: "tool_use",
+				id: "mock-call-1",
 				name: options.toolName,
-				arguments: options.toolArguments.slice(0, midpoint)
+				input: {}
 			}
-		}] },
-		finish_reason: null
-	}] }, { choices: [{
-		index: 0,
-		delta: { tool_calls: [{
+		},
+		...[options.toolArguments.slice(0, midpoint), options.toolArguments.slice(midpoint)].map((partial) => ({
+			type: "content_block_delta",
 			index: 0,
-			function: { arguments: options.toolArguments.slice(midpoint) }
-		}] },
-		finish_reason: null
-	}] }];
+			delta: {
+				type: "input_json_delta",
+				partial_json: partial
+			}
+		})),
+		{
+			type: "content_block_stop",
+			index: 0
+		}
+	];
 }
 async function runBehavior(options, record, request, response) {
 	switch (record.behavior) {
@@ -275,7 +310,8 @@ async function runBehavior(options, record, request, response) {
 			return;
 		case "empty":
 			openSse(response);
-			writeSse(record, response, terminalChunk("stop", 0));
+			startMessage(record, response);
+			writeSse(record, response, terminalChunk("end_turn", 0));
 			writeDone(record, response);
 			response.end();
 			finishRecord(options, record, "completed");
@@ -287,22 +323,22 @@ async function runBehavior(options, record, request, response) {
 			return;
 		case "stream_eof":
 			openSse(response);
-			writeSse(record, response, { choices: [{
-				index: 0,
-				delta: { role: "assistant" },
-				finish_reason: null
-			}] });
+			startMessage(record, response);
 			response.end();
 			finishRecord(options, record, "completed");
 			return;
 		case "partial_eof":
 			openSse(response);
+			startMessage(record, response);
+			startText(record, response, 0);
 			await streamText(options, record, response, options.partialText, 0);
 			response.end();
 			finishRecord(options, record, "completed");
 			return;
 		case "partial_disconnect":
 			openSse(response);
+			startMessage(record, response);
+			startText(record, response, 0);
 			if (!await streamText(options, record, response, options.partialText, options.chunkDelayMs)) return;
 			await disconnect(options, record, response);
 			return;
@@ -319,14 +355,19 @@ async function runBehavior(options, record, request, response) {
 			return;
 		case "malformed_event":
 			openSse(response);
-			writeSse(record, response, { choices: [null] });
+			writeSse(record, response, {
+				type: "content_block_start",
+				index: 0,
+				content_block: null
+			});
 			writeDone(record, response);
 			response.end();
 			finishRecord(options, record, "completed");
 			return;
 		case "wrong_content_type":
 			openSse(response, "application/json");
-			await completeText(options, record, response, "stop", 0);
+			startMessage(record, response);
+			await completeText(options, record, response, "end_turn", 0);
 			return;
 		case "rate_limit":
 			httpError(options, record, response, 429, "mock rate limit", "rate_limit");
@@ -351,32 +392,52 @@ async function runBehavior(options, record, request, response) {
 			return;
 		case "success":
 			openSse(response);
-			await completeText(options, record, response, "stop", 0);
+			startMessage(record, response);
+			await completeText(options, record, response, "end_turn", 0);
 			return;
 		case "reasoning_success":
 			openSse(response);
-			for (const chunk of splitText(options.reasoningText, options.chunkSize)) writeSse(record, response, { choices: [{
+			startMessage(record, response);
+			writeSse(record, response, {
+				type: "content_block_start",
+				index: 0,
+				content_block: {
+					type: "thinking",
+					thinking: ""
+				}
+			});
+			for (const chunk of splitText(options.reasoningText, options.chunkSize)) writeSse(record, response, {
+				type: "content_block_delta",
 				index: 0,
-				delta: { reasoning_content: chunk },
-				finish_reason: null
-			}] });
-			await completeText(options, record, response, "stop", 0);
+				delta: {
+					type: "thinking_delta",
+					thinking: chunk
+				}
+			});
+			writeSse(record, response, {
+				type: "content_block_stop",
+				index: 0
+			});
+			await completeText(options, record, response, "end_turn", 0, 1);
 			return;
 		case "tool_call_success":
 			openSse(response);
+			startMessage(record, response);
 			for (const chunk of toolCallChunks(options)) writeSse(record, response, chunk);
-			writeSse(record, response, terminalChunk("tool_calls", 2));
+			writeSse(record, response, terminalChunk("tool_use", 2));
 			writeDone(record, response);
 			response.end();
 			finishRecord(options, record, "completed");
 			return;
 		case "max_tokens":
 			openSse(response);
-			await completeText(options, record, response, "length", 0);
+			startMessage(record, response);
+			await completeText(options, record, response, "max_tokens", 0);
 			return;
 		case "slow_success":
 			openSse(response);
-			await completeText(options, record, response, "stop", options.chunkDelayMs);
+			startMessage(record, response);
+			await completeText(options, record, response, "end_turn", options.chunkDelayMs);
 			return;
 	}
 }
@@ -401,9 +462,9 @@ function chooseRandomBehavior(weights, random) {
 	return weights.at(-1)[0];
 }
 /**
-* Start a local chat-completions server that consumes one configured behavior
-* per accepted request. Only a `POST` path ending in `/chat/completions` consumes the script;
-* invalid routes, methods, authorization, and JSON receive ordinary 4xx
+* Start a local Messages server that consumes one configured behavior
+* per accepted request. Only a `POST` path ending in `/v1/messages` consumes the script;
+* invalid routes, methods, API keys, and JSON receive ordinary 4xx
 * responses. Closing the handle terminates stalled connections.
 *
 * @param options - listener, script, response content, timing, and telemetry options.
@@ -430,14 +491,14 @@ async function startMockLlmServer(options) {
 			response.writeHead(405, { allow: "POST" }).end();
 			return;
 		}
-		if (!path.endsWith("/chat/completions")) {
+		if (!path.endsWith("/v1/messages")) {
 			response.writeHead(404).end();
 			return;
 		}
-		if (resolved.apiKey !== void 0 && request.headers.authorization !== `Bearer ${resolved.apiKey}`) {
+		if (resolved.apiKey !== void 0 && request.headers["x-api-key"] !== resolved.apiKey) {
 			response.writeHead(401, { "content-type": "application/json" });
 			response.end(JSON.stringify({ error: {
-				message: "invalid mock bearer token",
+				message: "invalid mock API key",
 				code: "invalid_api_key"
 			} }));
 			return;

package/lib/types/index.d.ts

@@ -1,6 +1,6 @@
 /**
- * Scriptable OpenAI-compatible HTTP/SSE server for transport, protocol, and
- * semantic-empty LLM recovery tests. Each accepted chat-completions request
+ * Scriptable Messages HTTP/SSE server for transport, protocol, and
+ * semantic-empty LLM recovery tests. Each accepted Messages request
  * consumes one behavior; the server never retries or interprets harness policy.
  *
  * @module @deepseek-ai/dsh-llm-mock-server
@@ -40,7 +40,7 @@ export type MockLlmServerEvent = {
 };
 /** Captured wire request and its final server-side outcome. */
 export interface MockLlmRequestRecord {
-    /** One-based accepted chat-completions request number. */
+    /** One-based accepted Messages request number. */
     readonly attempt: number;
     /** Script entry consumed for this request before random resolution. */
     readonly scriptBehavior: MockLlmBehavior | 'script_exhausted';
@@ -63,7 +63,7 @@ export interface MockLlmServerOptions {
     readonly host?: string;
     /** TCP port; zero requests an OS-assigned port. */
     readonly port?: number;
-    /** Optional exact bearer token; omission accepts any authorization header. */
+    /** Optional exact API key; omission accepts any x-api-key header. */
     readonly apiKey?: string;
     /** Ordered request behaviors; exhaustion fails loud unless `repeatLast` is true. */
     readonly sequence: readonly MockLlmBehavior[];
@@ -98,7 +98,7 @@ export interface MockLlmServerOptions {
 }
 /** Running mock server and captured request state. */
 export interface MockLlmServer {
-    /** Base URL without `/v1`; both root and `/v1` chat-completions paths are accepted. */
+    /** Base URL without `/v1`; the endpoint is `/v1/messages`. */
     readonly baseURL: string;
     /** Actual bound port, including an OS-assigned value. */
     readonly port: number;
@@ -110,9 +110,9 @@ export interface MockLlmServer {
     close(): Promise<void>;
 }
 /**
- * Start a local chat-completions server that consumes one configured behavior
- * per accepted request. Only a `POST` path ending in `/chat/completions` consumes the script;
- * invalid routes, methods, authorization, and JSON receive ordinary 4xx
+ * Start a local Messages server that consumes one configured behavior
+ * per accepted request. Only a `POST` path ending in `/v1/messages` consumes the script;
+ * invalid routes, methods, API keys, and JSON receive ordinary 4xx
  * responses. Closing the handle terminates stalled connections.
  *
  * @param options - listener, script, response content, timing, and telemetry options.

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@deepseek-ai/dsh-llm-mock-server",
-  "description": "Scriptable OpenAI-compatible HTTP/SSE fault server for LLM recovery tests",
-  "version": "0.1.6-alpha.2",
+  "description": "Scriptable Messages HTTP/SSE fault server for LLM recovery tests",
+  "version": "0.1.7-alpha.1",
   "publishConfig": {
     "access": "public"
   },
@@ -27,9 +27,9 @@
   ],
   "license": "MIT",
   "peerDependencies": {
-    "@deepseek-ai/cordis": "^4.0.2"
+    "@deepseek-ai/cordis": "^4.0.3"
   },
   "devDependencies": {
-    "@deepseek-ai/cordis": "^4.0.2"
+    "@deepseek-ai/cordis": "^4.0.3"
   }
 }