@qwen-code/arena-agent 0.0.10-dev.10 → 0.0.10-dev.11

package/dist/types.d.ts

@@ -11,6 +11,7 @@ export interface ToolCall {
     id: string;
     type: "function";
     function: FunctionCall;
+    extra_content?: Record<string, unknown>;
 }
 export interface SystemMessage {
     role: "system";
@@ -246,6 +247,7 @@ export interface ModelDeltaToolCall {
     index?: number;
     id?: string;
     type?: string;
+    extra_content?: ToolCall["extra_content"];
     function?: {
         name?: string;
         arguments?: string;

package/docs/MODES_AND_RELEASES.md

@@ -22,7 +22,10 @@ src/
   mode.ts                  # 四模式名称、别名与默认 step limit
   workspace.ts             # React/Full Stack scaffold 初始化
 
-spec/                      # Arena 生产 fixtures（保持原路径）
+spec/                      # Agent Arena 当前提示词与工具契约源文件
+  message.json             # 普通模式系统提示词在 messages[0].content
+  tools.json               # 工具契约与捕获的生成默认值
+  session-memory/          # 当前 rolling 记忆提示词源文件
 modes/code-arena/
   model.json               # 三种 Code Arena 共用模型参数
   react/
@@ -42,6 +45,34 @@ modes/code-arena/
 contract 一起编译到 `src/canonical-data.ts`。正常 npm 运行不读取 prompt/schema
 源文件；scaffold 仍作为包资源保留，SEA 构建时则作为 assets 嵌入二进制。
 
+## 源码入口与历史资料
+
+普通 Agent Arena 的提示词正文只在 `spec/message.json` 维护。构建将它嵌入
+`src/canonical-data.ts`，运行时由 `src/spec.ts` 构造请求。`src/prompts.ts` 的
+普通模式审计模板也从这份生成数据取得，不再手工保存另一份相同正文。
+该文件中直接写出的 Repo 提示词用于 Git 仓库会话，是另一种场景的提示词。
+
+| 要查看的内容 | 当前入口 | 用途与维护方式 |
+| --- | --- | --- |
+| Agent Arena 普通模式提示词 | `spec/message.json` → `src/spec.ts` 的 `systemMessage()` | `messages[0].content` 是正式源文本，构造请求时填入会话信息 |
+| Agent Arena Repo 模式提示词 | `src/prompts.ts` 的 `repoAgentSystemPromptTemplate`、`repoClosedSessionNotice` | 当前运行时直接使用的模板 |
+| Agent Arena 普通模式审计模板 | `src/prompts.ts` 的 `agentSystemPromptTemplate` | 自动取自正式源的生成数据，只把日期、时区、位置换成占位符；不另存正文 |
+| Agent Arena 工具契约 | `spec/tools.json` | 描述、schema、顺序与捕获的生成默认值；实现位于 `src/tools.ts` |
+| Code Arena 提示词与工具契约 | `modes/code-arena/<mode>/system-prompt.txt`、`tools.json` 及共用 `model.json` | 三个模式分别维护；实现位于 `src/code-tools.ts` |
+| Agent Arena rolling 记忆提示词 | `spec/session-memory/*.txt` | 当前构建源文件；由 `generate-session-memory-prompts.mjs` 生成运行时数据 |
+| 摘要模式实现 | `src/summary-compaction.ts` | Agent Arena 的 `summary` 模式和 Code Arena 仍在使用 |
+| 生成文件 | `src/canonical-data.ts`、`src/session-memory-prompts.ts` | 由 `npm run build` 前置步骤生成，不手动改正文 |
+| Repo 提示词测试样本 | `test/fixtures/agent-arena/message-repo*.json` | 验证模板能逐字还原捕获；不作为运行时配置 |
+| 历史输入与报告 | `tasks/`、`docs/history/`、`docs/evidence/` | 按记录的时间、批次和代码身份阅读，不当成当前配置或最新结论 |
+
+`spec/session-memory/` 的捕获来源记录保留在
+`tasks/v2/contracts/compaction/capture.json`，原始样本在同目录的 `dump/`。
+来源证据与当前构建源文件分开保存，迁移目录不改变提示词字节。
+
+工具脚本按用途从 [scripts/README.md](../scripts/README.md) 进入。旧脚本仍可能被
+回归测试或历史批次使用；没有静态引用不等于可以删除。`code-harness`、旧环境变量
+别名和生成文件也属于兼容或打包路径，不作为重复实现清理。
+
 ## 模式选择
 
 统一 CLI 使用显式参数：

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@qwen-code/arena-agent",
-  "version": "0.0.10-dev.10",
+  "version": "0.0.10-dev.11",
   "description": "A four-mode TypeScript agent harness for Arena and Code Arena traces.",
   "type": "module",
   "types": "./dist/sdk.d.ts",
@@ -73,6 +73,7 @@
     "typecheck": "tsc -p tsconfig.json --noEmit && tsc -p tsconfig.web.json --noEmit && npm run typecheck:code-arena-replay"
   },
   "dependencies": {
+    "@modelcontextprotocol/sdk": "1.32.1",
     "@stacksjs/ts-webp": "^0.1.2",
     "ajv": "^8.20.0",
     "ajv-formats": "^3.0.1",

package/scripts/README.md

@@ -1,9 +1,27 @@
-# 轨迹复刻脚本
+# 脚本入口
 
-本目录保存 DataWorks SQL Arena 轨迹数据的复刻与环境准备脚本。数据集位于
-`tasks/`，运行产物位于 `tasks/reproductions/`。
+本目录包含构建、发布、对齐审计和历史实验脚本。先按用途选择入口，不能将
+旧实验脚本的默认样本或配置当作当前运行设置。
+
+| 用途 | 入口 | 说明 |
+| --- | --- | --- |
+| 构建 CLI、Web 与 SDK | `npm run build` | 先生成当前契约和记忆提示词数据，再编译 TypeScript |
+| 校验与发布二进制 | [发版流程](../docs/MODES_AND_RELEASES.md#发布节奏) | 包含版本、changelog 和平台包检查 |
+| 新批次 Agent Arena / Code Arena 对齐 | [arena-align](../skills/arena-align/SKILL.md) | 按用户指定输入组织四个阶段；不让模型重新决策 |
+| Agent Arena 环境与契约审计 | `audit-arena-first-two.ts` | 当前阶段入口，保留不匹配样本；运行方式见对应 Skill |
+| 旧 Agent Arena 审计 | `audit-arena-capture.ts` | 依赖旧 v2 工具基准，候选带精确契约过滤；不用于新批次完整审计 |
+| 从旧 v2 数据重建契约 | `refresh-arena-v2-contract.mjs` | 显式覆盖当前 fixtures 的历史维护脚本；不是构建步骤，也不获取最新官方配置 |
+| 旧任务的模型重跑实验 | `reproduce_case.mjs`、`reproduce_all.mjs` | 会调用 agent 模型重新执行任务，见下文；与历史工具回放的验证范围不同 |
+
+`replay-*`、`audit-*`、`probe-*` 等工具中还有各批次的专项诊断入口。按关联的
+Skill、测试和报告核对输入格式、执行身份及调用范围，不以文件是否出现在 npm
+命令列表里判断是否仍有用途。当前源码和生成文件的位置见
+[源码入口与历史资料](../docs/MODES_AND_RELEASES.md#源码入口与历史资料)。
 
-## 通用脚本
+## 旧任务的模型重跑实验
+
+以下说明仅适用于 DataWorks SQL Arena 数据的 `reproduce_*` 脚本。数据集位于
+`tasks/`，运行产物位于 `tasks/reproductions/`。
 
 `reproduce_case.mjs` 运行单个 case，`reproduce_all.mjs`
 批量运行数据集中标记为可复刻的 case。

package/dist/legacy-compaction.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"legacy-compaction.js","sourceRoot":"","sources":["../src/legacy-compaction.ts"],"names":[],"mappings":"AAUA,MAAM,oBAAoB,GACxB,8EAA8E;IAC9E,+EAA+E;IAC/E,2EAA2E;IAC3E,6EAA6E;IAC7E,6EAA6E;IAC7E,yBAAyB,CAAC;AAE5B,MAAM,mBAAmB,GAAG;;;;;;;;;;;;;;;;4gBAgBgf,CAAC;AAE7gB,MAAM,cAAc,GAClB,8EAA8E,CAAC;AAEjF,MAAM,uBAAuB,GAC3B,+HAA+H,CAAC;AAoBlI,SAAS,iBAAiB,CAAC,OAAwB;IACjD,IAAI,OAAO,OAAO,CAAC,OAAO,KAAK,QAAQ;QAAE,OAAO,OAAO,CAAC,OAAO,CAAC;IAChE,OAAO,OAAO,CAAC,OAAO;SACnB,MAAM,CAAC,CAAC,IAAI,EAA4B,EAAE,CAAC,MAAM,IAAI,IAAI,CAAC;SAC1D,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC;SACxB,IAAI,CAAC,EAAE,CAAC,CAAC;AACd,CAAC;AAED,SAAS,kBAAkB,CAAC,OAAgB;IAC1C,OAAO,CACL,OAAO,CAAC,IAAI,KAAK,MAAM;QACvB,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC;QAC9B,uBAAuB,CAAC,IAAI,CAAC,iBAAiB,CAAC,OAAO,CAAC,CAAC,CACzD,CAAC;AACJ,CAAC;AAED,SAAS,oBAAoB,CAAC,QAA4B;IACxD,KAAK,IAAI,KAAK,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,KAAK,IAAI,CAAC,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QAC7D,MAAM,OAAO,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC;QAChC,IAAI,OAAO,EAAE,IAAI,KAAK,MAAM,IAAI,CAAC,kBAAkB,CAAC,OAAO,CAAC;YAAE,OAAO,KAAK,CAAC;IAC7E,CAAC;IACD,OAAO,CAAC,CAAC,CAAC;AACZ,CAAC;AAOD,SAAS,kBAAkB,CAAC,MAAuB,EAAE,KAAa;IAChE,KAAK,MAAM,SAAS,IAAI,KAAK,EAAE,CAAC;QAC9B,IAAI,SAAS,CAAC,WAAW,CAAC,CAAC,CAAE,IAAI,IAAI;YAAE,MAAM,CAAC,KAAK,IAAI,CAAC,CAAC;;YACpD,MAAM,CAAC,QAAQ,IAAI,CAAC,CAAC;IAC5B,CAAC;AACH,CAAC;AAED,SAAS,uBAAuB,CAC9B,QAA4B,EAC5B,KAAgC;IAEhC,MAAM,MAAM,GAAoB,EAAE,KAAK,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC;IAC1D,MAAM,QAAQ,GAAG,CAAC,GAAW,EAAE,KAAc,EAAW,EAAE,CACxD,GAAG,KAAK,OAAO,IAAI,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC;IACnE,kBAAkB,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;IAC5C,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,QAAQ,CAAC,MAAM,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QACxD,IAAI,KAAK;YAAE,kBAAkB,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;QAC3C,MAAM,OAAO,GAAG,QAAQ,CAAC,KAAK,CAAE,CAAC;QACjC,MAAM,YAAY,GAChB,OAAO,CAAC,IAAI,KAAK,WAAW,IAAI,OAAO,CAAC,qBAAqB,EAAE,MAAM;YACnE,CAAC,CAAC;gBACE,IAAI,EAAE,OAAO,CAAC,IAAI;gBAClB,qBAAqB,EAAE,OAAO,CAAC,qBAAqB;aACrD;YACH,CAAC,CAAC,OAAO,CAAC;QACd,kBAAkB,CAAC,MAAM,EAAE,IAAI,CAAC,SAAS,CAAC,YAAY,EAAE,QAAQ,CAAC,CAAC,CAAC;IACrE,CAAC;IACD,kBAAkB,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;IACzC,kBAAkB,CAAC,MAAM,EAAE,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC,CAAC;IAC5D,kBAAkB,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAChC,OAAO,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,CAAC;AACvD,CAAC;AAED,SAAS,UAAU,CAAC,KAAc;IAChC,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;QACzB,OAAO,KAAK,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,KAAK,GAAG,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC;IACpE,CAAC;IACD,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,CAAC,CAAC;IAClD,MAAM,MAAM,GAAG,KAAgC,CAAC;IAChD,MAAM,QAAQ,GACZ,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,IAAI,KAAK,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1E,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC;QAAE,YAAY,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC;IAC3E,OAAO,QAAQ,GAAG,YAAY,CAAC;AACjC,CAAC;AAED,MAAM,UAAU,qBAAqB,CACnC,QAA4B,EAC5B,KAAgC,EAChC,kBAAkB,GAAG,IAAI;IAEzB,MAAM,oBAAoB,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CACpD,OAAO,CAAC,IAAI,KAAK,WAAW,IAAI,OAAO,CAAC,qBAAqB,EAAE,MAAM;QACnE,CAAC,CAAC;YACE,IAAI,EAAE,OAAO,CAAC,IAAI;YAClB,qBAAqB,EAAE,OAAO,CAAC,qBAAqB;SACrD;QACH,CAAC,CAAC,OAAO,CACZ,CAAC;IACF,MAAM,kBAAkB,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,GAAG,KAAK,CAAC,MAAM,GAAG,EAAE,CAAC;IACnE,OAAO,CACL,uBAAuB,CAAC,QAAQ,EAAE,KAAK,CAAC;QACxC,UAAU,CAAC,oBAAoB,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,kBAAkB,CAAC;QAClE,kBAAkB,CACnB,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAAC,KAAa;IAChC,OAAO,KAAK;SACT,IAAI,EAAE;SACN,OAAO,CAAC,0BAA0B,EAAE,8BAA8B,CAAC;SACnE,OAAO,CAAC,4BAA4B,EAAE,+BAA+B,CAAC,CAAC;AAC5E,CAAC;AAED,MAAM,UAAU,wBAAwB,CAAC,OAAe;IACtD,MAAM,WAAW,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC,OAAO,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;IAC/D,OAAO,CACL,0BAA0B;QAC1B,GAAG,oBAAoB,YAAY,WAAW,SAAS;QACvD,yBAAyB,CAC1B,CAAC;AACJ,CAAC;AAED,SAAS,cAAc,CACrB,OAAwB,EACxB,YAAoB;IAEpB,IAAI,OAAO,OAAO,CAAC,OAAO,KAAK,QAAQ,EAAE,CAAC;QACxC,OAAO;YACL,IAAI,EAAE,MAAM;YACZ,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,EAAE,EAAE,IAAI,EAAE,OAAO,CAAC,OAAO,EAAE,CAAC;SAC7B,CAAC;IACpC,CAAC;IACD,OAAO;QACL,IAAI,EAAE,MAAM;QACZ,OAAO,EAAE;YACP,EAAE,IAAI,EAAE,YAAY,EAAE;YACtB,GAAG,eAAe,CAAC,OAAO,CAAC,OAAO,CAAC;SACpC;KAC8B,CAAC;AACpC,CAAC;AAOD,KAAK,UAAU,eAAe,CAC5B,MAAmB,EACnB,KAAa,EACb,cAAkC,EAClC,SAAiB,EACjB,MAAoB;IAEpB,MAAM,OAAO,GAAiB;QAC5B,KAAK;QACL,QAAQ,EAAE;YACR,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,mBAAmB,EAAE;YAChD,GAAG,eAAe,CAAC,cAAc,CAAC;YAClC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,cAAc,EAAE;SAC1C;QACD,KAAK,EAAE,EAAE;QACT,UAAU,EAAE,MAAM;QAClB,WAAW,EAAE,CAAC;QACd,SAAS;QACT,cAAc,EAAE,KAAK;QACrB,cAAc,EAAE,CAAC;QACjB,gBAAgB,EAAE,KAAK;QACvB,MAAM;KACP,CAAC;IACF,MAAM,OAAO,GAAa,EAAE,CAAC;IAC7B,IAAI,KAA6B,CAAC;IAClC,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC;QACjD,IAAI,KAAK,CAAC,KAAK;YAAE,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC;QACrC,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC;QACjD,IAAI,KAAK;YAAE,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;IACzC,CAAC;IACD,OAAO,EAAE,OAAO,EAAE,WAAW,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC;AAC3D,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,uBAAuB,CAC3C,QAAmB,EACnB,KAAgC,EAChC,MAAmB,EACnB,KAAa,EACb,QAA4B,EAC5B,MAAoB;IAEpB,MAAM,WAAW,GAAG,QAAQ,CAAC,kBAAkB,IAAI,IAAI,CAAC;IACxD,MAAM,MAAM,GAAG,qBAAqB,CAAC,QAAQ,EAAE,KAAK,EAAE,WAAW,CAAC,CAAC;IACnE,MAAM,SAAS,GAAG,QAAQ,CAAC,yBAAyB,IAAI,CACtD,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,CAAC,aAAa,CAAC;QACnC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;IACnD,IAAI,MAAM,GAAG,SAAS,EAAE,CAAC;QACvB,OAAO;YACL,SAAS,EAAE,KAAK;YAChB,qBAAqB,EAAE,MAAM;YAC7B,oBAAoB,EAAE,MAAM;SAC7B,CAAC;IACJ,CAAC;IAED,MAAM,gBAAgB,GAAG,oBAAoB,CAAC,QAAQ,CAAC,CAAC;IACxD,IAAI,gBAAgB,IAAI,CAAC,EAAE,CAAC;QAC1B,OAAO;YACL,SAAS,EAAE,KAAK;YAChB,qBAAqB,EAAE,MAAM;YAC7B,oBAAoB,EAAE,MAAM;SAC7B,CAAC;IACJ,CAAC;IAED,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,GAAG,MAAM,eAAe,CAC9C,MAAM,EACN,KAAK,EACL,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,gBAAgB,CAAC,EACnC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,gBAAgB,CAAC,CAAC,EAClD,MAAM,CACP,CAAC;IACF,IAAI,CAAC,OAAO,EAAE,CAAC;QACb,OAAO;YACL,SAAS,EAAE,KAAK;YAChB,qBAAqB,EAAE,MAAM;YAC7B,oBAAoB,EAAE,MAAM;YAC5B,KAAK;SACN,CAAC;IACJ,CAAC;IAED,MAAM,WAAW,GAAG,QAAQ,CAAC,gBAAgB,CAAC,CAAC;IAC/C,IAAI,WAAW,EAAE,IAAI,KAAK,MAAM,EAAE,CAAC;QACjC,OAAO;YACL,SAAS,EAAE,KAAK;YAChB,qBAAqB,EAAE,MAAM;YAC7B,oBAAoB,EAAE,MAAM;YAC5B,KAAK;SACN,CAAC;IACJ,CAAC;IACD,MAAM,iBAAiB,GAAc;QACnC,QAAQ,CAAC,CAAC,CAAE;QACZ,cAAc,CAAC,WAAW,EAAE,wBAAwB,CAAC,OAAO,CAAC,CAAC;QAC9D,GAAG,QAAQ,CAAC,KAAK,CAAC,gBAAgB,GAAG,CAAC,CAAC;KACxC,CAAC;IACF,QAAQ,CAAC,MAAM,CAAC,CAAC,EAAE,QAAQ,CAAC,MAAM,EAAE,GAAG,iBAAiB,CAAC,CAAC;IAC1D,OAAO;QACL,SAAS,EAAE,IAAI;QACf,qBAAqB,EAAE,MAAM;QAC7B,oBAAoB,EAAE,qBAAqB,CAAC,QAAQ,EAAE,KAAK,EAAE,WAAW,CAAC;QACzE,KAAK;QACL,MAAM,EAAE,OAAO;QACf,eAAe,EAAE,gBAAgB;KAClC,CAAC;AACJ,CAAC"}