chore(release): prepare ARTEMIS 2.4.3

AGENTS.md

@@ -21,7 +21,7 @@
 | 探索、示范、用例提炼与复跑 | [探索说明](docs/exploration-mode.md) |
 | 插件协议、适配或市场协同 | [插件开发规范](docs/plugin-development.md)及插件仓库自身的 AGENTS.md、DESIGN.md |
 | 远程 Python SDK | [SDK 设计](packages/artemis-client/DESIGN.md)，保持标准库薄客户端边界 |
-| 继续历史开发、确认版本或发布 | [迁入时的开发快照](docs/development-handoff.md)、[2.1.0 发行说明](docs/releases/2.1.0.md)，再核对实时 Git 状态；`docs/plans/` 保留各自阶段的状态 |
+| 继续历史开发、确认版本或发布 | [README 当前发行入口](README.md)、对应发行说明及实时 Git 状态；[开发快照](docs/development-handoff.md)与 `docs/plans/` 仅作为历史线索，不替代当前源码与发布包核验 |
 
 ## Quick Commands
 
@@ -60,7 +60,7 @@ Windows 首次安装使用 `start.bat`，国内镜像使用 `start-cn.bat`。启
 ## Conventions
 
 - Python 使用绝对导入、4 空格缩进和双引号，由 Ruff 检查；不要放宽质量基线来掩盖新增问题。
-- 新增可调参数集中进入配置层，并更新环境示例、网页/MCP 入口和必要文档；默认值同步 `config/artemis.jsonc` 与 `artemis/resources/config/artemis.jsonc`。
+- 新增可调参数集中进入配置层，按作用域更新环境示例、网页/MCP 入口和文档；明确仅供环境配置的参数不增加前端入口。JSON 默认值同步 `config/artemis.jsonc` 与 `artemis/resources/config/artemis.jsonc`。
 - MCP 工具增删时同步工具注册、CLI 安装白名单、测试、Skill 和能力文档。
 - 同步 Codex MCP/Skill 时保留已有工具白名单、禁用状态和审批策略；异常时保留原配置并告警，不生成非法 TOML 或阻断网页启动。
 - Windows PowerShell 启动脚本保持兼容 Windows PowerShell 5.1，避免编码导致解析失败。升级只做 Git 快进，保护已有配置和本地修改。
@@ -75,6 +75,16 @@ Windows 首次安装使用 `start.bat`，国内镜像使用 `start-cn.bat`。启
 - 设备验证沿用用户指定设备；多设备且未指定时先确认。编写应用交互测试前，先通过 ARTEMIS 探索实际页面和操作路径。
 - 提交前检查 `git diff --check` 和暂存范围，只暂存本次具体路径；报告实际执行的验证及未完成项，不将未运行的检查写成通过。
 
+## Release Gates
+
+- 主程序、SDK、插件和第三方依赖独立版本化。主程序升版只修改 `pyproject.toml` 的项目版本、`uv.lock` 的 artemis 条目及前端应用的根版本字段，发行文档另行更新。禁止对仓库或整个锁文件全局替换版本字符串。
+- 单独审查锁文件差异。第三方条目的版本、下载地址、完整性校验和依赖声明必须相互匹配；主程序升版不应改变这些条目。真正升级依赖时由包管理器生成并核对上游声明；发生 EINTEGRITY 先核验来源，禁止关闭校验或只照错误提示改哈希。
+- 发布前必须在隔离目录解压最终 ZIP，使用独立空 npm 缓存完成 `npm ci` 和生产构建，覆盖本次发布的官方/国内安装路径。包内安装或构建输入变化后必须重跑；不得清空用户全局缓存或改动用户运行目录来制造干净环境。
+- 记录实际 Node/npm 版本、镜像、源码提交、ZIP SHA256 和验证范围。已有 node_modules、构建指纹、缓存命中或大量单元测试通过，均不能证明全新安装成功；某个 Node 版本通过也不能冒充其他版本实测。测试和构建命令见前端局部指南。
+- 发布顺序：固定提交并打包 → 验证最终包 → 推送标签 → 创建草稿并上传包及校验文件 → 回读校验 → 正式发布 → 无认证下载核验。标签必须指向打包提交；推送标签后检查 Release 状态，本仓库曾出现 Gitea 将已有草稿自动发布的情况。失败时修复并重新验证，不提前宣布完成。
+- 候选包与正式包目录明确区分。候选包失效后移出可分发路径并标明替代版本，不能留下同名旧 ZIP 供用户误取；对外只给已核验的正式发布链接或准确文件路径。正式版本不覆盖原标签或附件，后续修复发补丁版。
+- 安装或市场版本不符时，先核对用户实际 ZIP 的 SHA256、运行源码提交及生效市场地址；不凭界面版本号判断，也不把未经验证的缓存假设当成原因。主程序与插件版本、索引兼容性分别验证。
+
 ## Boundaries
 
 - 始终保护用户任务、历史库、环境配置和凭据。不要把个人设备序列号、绝对目录或 API Key 写入源码。

README.md

@@ -2,7 +2,7 @@
 
 基于 [ARTEMIS](https://github.com/google/artemis) 扩展的 Android AI 自动化测试平台。用自然语言描述任务，即可通过简体中文网页或 MCP 驱动真机、模拟器执行操作并收集结果。模型统一使用 **OpenAI Responses API**。
 
-**2.4.1**：修复全新安装时的 npm 校验失败，准确检查 Node.js 版本要求。人工评审插件继续使用 1.1.0，市场使用兼容旧宿主的独立索引。见 [2.4.1 发行说明](docs/releases/2.4.1.md)；功能更新见 [2.4.0](docs/releases/2.4.0.md)。
+**2.4.3**：异常模型输出耗尽重试后只失败当前任务，避免触发全局暂停；补充请求关联和有界参数摘要，便于定位流式异常。见 [2.4.3 发行说明](docs/releases/2.4.3.md)。人工评审插件保持 1.1.0；列表优化见 [2.4.2](docs/releases/2.4.2.md)，安装修复见 [2.4.1](docs/releases/2.4.1.md)。
 
 ## 主要功能
 

apps/admin_console/routers/automation.py

@@ -12,7 +12,7 @@ from apps.admin_console.services.exploration_listing import EXPLORATION_RUNS_CTE
 from apps.admin_console.services.automation_store import identifier, store
 from apps.admin_console.services import automation_service as service
 from apps.admin_console.services import batch_archive as archive_service
-from apps.admin_console.services import report_export, task_import
+from apps.admin_console.services import report_export, task_extensions, task_import
 from artemis.config.settings import settings
 from apps.admin_console.services.case_review import read_reviewed_result
 
@@ -131,7 +131,9 @@ def batch_by_request(request_id: str):
 @router.get("/batches/{batch_id}")
 def batch_detail(batch_id: str):
     jobs = store.jobs(batch_id)
+    reviewed = task_extensions.reviewed_batch_sessions(batch_id, store)
     for job in jobs:
+        job["human_reviewed"] = job["session_id"] in reviewed
         job["test_result"] = read_reviewed_result(
             job["session_id"], job["status"], job.get("error") or "", storage=store
         )

apps/admin_console/services/task_extensions.py

@@ -36,6 +36,32 @@ def installation_signature() -> str:
     )
 
 
+def reviewed_batch_sessions(batch_id: str, store: AutomationStore) -> set[str]:
+    """Read current human review values once per batch, without loading history."""
+    manifest = next((item for item in active_manifests() if item.id == "task-review"), None)
+    if manifest is None or manifest.workspace is None:
+        return set()
+    with store.connection() as conn:
+        rows = conn.execute(
+            """SELECT j.session_id, json_extract(r.payload, '$."task-review".values') AS review_values
+               FROM automation_jobs j JOIN automation_records r
+                 ON r.kind='task_extensions' AND r.id=j.session_id
+               WHERE j.batch_id=?""",
+            (batch_id,),
+        ).fetchall()
+    reviewed = set()
+    for row in rows:
+        values = json.loads(row["review_values"] or "{}")
+        if any(
+            isinstance(value := values.get(field.id), str)
+            and value.strip()
+            and (field.kind == "text" or value in field.options)
+            for field in manifest.workspace.fields
+        ):
+            reviewed.add(row["session_id"])
+    return reviewed
+
+
 def task_context(session_id: str, store: AutomationStore) -> dict:
     identifier(session_id)
     session = SessionRepository(store.path).get_session_by_id(session_id) or {}

apps/showcase_ui/AGENTS.md

@@ -30,6 +30,9 @@ npm run build
 
 ## Local Conventions
 
+- 发版遵守根指南的 Release Gates。前端应用版本只改 package.json 顶层 `version` 和 package-lock.json 顶层及 `packages[""].version`；禁止把相同数字的第三方依赖一并替换。
+- Node 支持范围以锁定的 Angular CLI/构建依赖实际 `engines` 为准，核对启动脚本的主、次、补丁版本边界和提示；不要只判断“Node 22/24/26”。
+- 安装或依赖变更需在隔离目录、独立空 npm 缓存下运行 `npm ci`，再运行上述测试和构建；国内入口同时核验镜像安装。不得以关闭完整性校验、强制安装或删除锁文件掩盖不一致，也不得用已装依赖的构建结果宣称全新安装通过。
 - 页面组件在 src/app/components，服务在 src/app/services；先复用已有组件、样式及 API 服务。
 - 保持现有简体中文文案、颜色、间距和控件样式，菜单、弹窗和表单必须清晰可读。
 - 修改任务或配置交互时覆盖加载、保存、失败和空状态；不要将执行中误标成录屏中或把完成误标成验证通过。

apps/showcase_ui/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "artemis-showcase-ui",
-  "version": "2.4.1",
+  "version": "2.4.3",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "artemis-showcase-ui",
-      "version": "2.4.1",
+      "version": "2.4.3",
       "dependencies": {
         "@angular/cdk": "^22.2.4",
         "@angular/common": "^22.2.3",

apps/showcase_ui/package.json

@@ -1,6 +1,6 @@
 {
   "name": "artemis-showcase-ui",
-  "version": "2.4.1",
+  "version": "2.4.3",
   "scripts": {
     "ng": "ng",
     "start": "ng serve --proxy-config proxy.conf.json",

apps/showcase_ui/src/app/components/automation-panel/automation-panel.component.html

@@ -108,6 +108,19 @@
        </div>
      }
     @if(b.report_error){<p class="error">报告生成失败：{{b.report_error}}</p>}
-      @for(j of b.jobs; track j.session_id){<div class="job" [class.selected]="agent.currentSessionId() === j.session_id"><app-favorite-button [sourceId]="j.session_id" /><button (click)="openBatchTask(j.session_id)" [title]="j.case.goal" [attr.aria-pressed]="agent.currentSessionId() === j.session_id">{{j.case.case_id ? j.case.case_id + ' ' : ''}}{{j.case.name}}@if(j.repeat_count > 1){ · 第 {{j.iteration}} / {{j.repeat_count}} 轮}</button><span>{{label(j.status)}}</span><app-test-result-badge [result]="j.test_result" [executionStatus]="j.status" [compact]="true" />@if(j.error){<small>{{j.error}}</small>}</div>}
+      @for(j of b.jobs; track j.session_id){
+        <div class="job" [class.selected]="agent.currentSessionId() === j.session_id">
+          <div class="job-heading">
+            <app-favorite-button [sourceId]="j.session_id" />
+            <button (click)="openBatchTask(j.session_id)" [title]="j.case.goal" [attr.aria-pressed]="agent.currentSessionId() === j.session_id">{{j.case.case_id ? j.case.case_id + ' ' : ''}}{{j.case.name}}@if(j.repeat_count > 1){ · 第 {{j.iteration}} / {{j.repeat_count}} 轮}</button>
+          </div>
+          <div class="job-statuses">
+            <span>{{label(j.status)}}</span>
+            <app-test-result-badge [result]="j.test_result" [executionStatus]="j.status" [compact]="true" />
+            @if(j.human_reviewed){<span class="human-review-badge" title="人工评审已有保存内容，不代表测试通过">人工已审</span>}
+          </div>
+          @if(j.error){<small>{{j.error}}</small>}
+        </div>
+      }
   </section>}
 }

apps/showcase_ui/src/app/components/automation-panel/automation-panel.component.scss

@@ -37,7 +37,15 @@ input[type=checkbox]{width:15px;height:15px;accent-color:var(--app-primary-text,
 @keyframes archive-spin{to{transform:rotate(360deg);}}
 .code{font-family:Consolas,monospace;font-size:12px;line-height:1.6;resize:vertical;background:var(--app-surface-soft, #f8fafc);}.table-wrap{max-height:260px;overflow:auto;border:1px solid var(--app-border, #e2e8f0);border-radius:8px;margin-bottom:14px;}table{width:100%;border-collapse:collapse;}th,td{padding:9px 10px;border-bottom:1px solid var(--app-border, #e2e8f0);text-align:left;overflow-wrap:anywhere;}th{background:var(--app-surface-soft, #f8fafc);color:var(--app-text-secondary, #475569);font-weight:500;}
 .preview{white-space:pre-wrap;max-height:400px;overflow:auto;font:inherit;color:var(--app-text-control, #334155);background:var(--app-surface-soft, #f8fafc);border-radius:8px;padding:12px;border:1px solid var(--app-border, #e2e8f0);}
-.job{display:flex;flex-wrap:wrap;gap:8px;align-items:center;padding:10px 0;border-bottom:1px solid var(--app-border, #f1f5f9);}.job:last-child{border-bottom:0;}.job button{flex:1;justify-content:flex-start;text-align:left;min-width:150px;}.job span{font-size:12px;background:var(--app-surface-soft, #f1f5f9);color:var(--app-text-secondary, #475569);border-radius:5px;padding:3px 7px;}.job small{width:100%;}details{margin:14px 0;}summary{color:var(--app-text-secondary, #475569);cursor:pointer;padding:8px 0;}
+.job{display:flex;flex-direction:column;gap:8px;padding:12px 0;border-bottom:1px solid var(--app-border, #f1f5f9);}
+.job:last-child{border-bottom:0;}
+.job-heading{position:relative;min-width:0;}
+.job-heading>app-favorite-button{position:absolute;left:6px;top:4px;z-index:1;}
+.job-heading>button{box-sizing:border-box;width:100%;min-width:0;padding-left:40px;justify-content:flex-start;text-align:left;overflow-wrap:anywhere;}
+.job-statuses{display:flex;flex-wrap:wrap;align-items:center;gap:8px;}
+.job-statuses>span{display:inline-flex;align-items:center;box-sizing:border-box;height:20px;padding:0 6px;border:1px solid transparent;border-radius:5px;font-size:11px;line-height:14px;background:var(--app-surface-soft, #f1f5f9);color:var(--app-text-secondary, #475569);white-space:nowrap;}
+.job small{overflow-wrap:anywhere;}
+details{margin:14px 0;}summary{color:var(--app-text-secondary, #475569);cursor:pointer;padding:8px 0;}
 .task-guidance{margin-top:16px;padding:14px;border:1px solid var(--app-primary-border, #bfdbfe);border-radius:10px;background:var(--app-primary-soft, #f8fbff);}
 .task-guidance label{margin:8px 0 4px;}
 .report-overview{padding:14px;border:1px solid var(--app-primary-border, #dbeafe);border-radius:10px;background:var(--app-primary-soft, #f8fbff);}
@@ -58,3 +66,5 @@ input[type=checkbox]{width:15px;height:15px;accent-color:var(--app-primary-text,
   table{width:100%;border-collapse:collapse;font-size:12px;}th,td{padding:7px 9px;border:1px solid var(--app-border, #e2e8f0);vertical-align:top;}th{background:var(--app-primary-soft, #eff6ff);color:var(--app-primary-text, #1e3a8a);}
   .md-para-gap{height:8px;}
 }
+
+.job .human-review-badge{white-space:nowrap;border:1px solid var(--app-primary-border, #bfdbfe);background:var(--app-primary-soft, #eff6ff);color:var(--app-primary-text, #1d4ed8);font-weight:500;}

apps/showcase_ui/src/app/components/automation-panel/automation-panel.component.spec.ts

@@ -90,6 +90,30 @@ describe('AutomationPanelComponent batch deletion', () => {
     expect(http.post).not.toHaveBeenCalled();
   });
 
+  it('shows only saved human review badges and removes them on a batch refresh', async () => {
+    const fixture = TestBed.createComponent(AutomationPanelComponent);
+    const component = fixture.componentInstance;
+    await new Promise<void>(resolve => setTimeout(resolve, 0));
+    component.tab.set('batches');
+    const batch = { id: 'review-batch', name: '评审批次', status: 'completed', jobs: [
+      { session_id: 'reviewed', status: 'completed', human_reviewed: true, case: { name: '已评用例' } },
+      { session_id: 'empty', status: 'completed', human_reviewed: false, case: { name: '空白用例' } },
+      { session_id: 'legacy', status: 'completed', case: { name: '旧版任务' } }
+    ] };
+    component.batch.set(batch);
+    fixture.detectChanges();
+    const rows = fixture.nativeElement.querySelectorAll('.job');
+    expect(rows[0].querySelector('.human-review-badge').textContent).toBe('人工已审');
+    expect(rows[1].querySelector('.human-review-badge')).toBeNull();
+    expect(rows[2].querySelector('.human-review-badge')).toBeNull();
+    expect(fixture.nativeElement.textContent).not.toContain('人工未审');
+    (http.get.and as any).callFake((url: string) => of(url.endsWith('/review-batch')
+      ? { ...batch, jobs: batch.jobs.map(job => ({ ...job, human_reviewed: false })) } : []));
+    await component.refresh();
+    fixture.detectChanges();
+    expect(fixture.nativeElement.querySelector('.human-review-badge')).toBeNull();
+  });
+
   it('confirms deletion and removes the batch from the visible list', async () => {
     const fixture = TestBed.createComponent(AutomationPanelComponent);
     fixture.componentInstance.tab.set('batches');
@@ -129,7 +153,7 @@ describe('AutomationPanelComponent batch deletion', () => {
       {session_id:'task-b',status:'completed',case:{name:'用例 B',goal:'观察主页'}}
     ]});
     fixture.detectChanges();
-    const jobs: HTMLButtonElement[] = [...fixture.nativeElement.querySelectorAll('.job > button')];
+    const jobs: HTMLButtonElement[] = [...fixture.nativeElement.querySelectorAll('.job-heading > button')];
     jobs[0].click(); jobs[1].click();
     expect(agent.selectSession.calls.allArgs()).toEqual([['task-a',true],['task-b',true]]);
     expect(component.tab()).toBe('batches');

artemis/llm/argument_fingerprint.py

@@ -0,0 +1,53 @@
+"""Bounded fingerprints of received argument deltas, without retaining text."""
+
+import hashlib
+
+
+PREFIX_BYTES = 4096
+MAX_PREFIXES = 16
+MAX_TRACKED_TOOLS = 64
+
+
+class ArgumentFingerprint:
+    def __init__(self):
+        self.characters = 0
+        self.bytes = 0
+        self.deltas = 0
+        self.encoding_valid = True
+        self.digest = hashlib.sha256()
+        self.prefixes: list[dict] = []
+
+    def observe(self, arguments: str) -> None:
+        self.characters += len(arguments)
+        if not arguments:
+            return
+        self.deltas += 1
+        if not self.encoding_valid:
+            return
+        try:
+            encoded = arguments.encode("utf-8")
+        except UnicodeEncodeError:
+            # Diagnostics must not repair, reject or otherwise alter the stream.
+            self.encoding_valid = False
+            return
+        offset = 0
+        while offset < len(encoded) and len(self.prefixes) < MAX_PREFIXES:
+            size = min(PREFIX_BYTES - self.bytes % PREFIX_BYTES, len(encoded) - offset)
+            self.digest.update(encoded[offset : offset + size])
+            self.bytes += size
+            offset += size
+            if self.bytes % PREFIX_BYTES == 0:
+                self.prefixes.append({"bytes": self.bytes, "sha256": self.digest.hexdigest()})
+        if offset < len(encoded):
+            self.digest.update(encoded[offset:])
+            self.bytes += len(encoded) - offset
+
+    def snapshot(self) -> dict:
+        return {
+            "argument_chars": self.characters,
+            "argument_bytes": self.bytes if self.encoding_valid else None,
+            "argument_sha256": self.digest.hexdigest() if self.encoding_valid else None,
+            "argument_deltas": self.deltas,
+            "argument_prefixes": [dict(prefix) for prefix in self.prefixes],
+            "argument_encoding_valid": self.encoding_valid,
+        }

artemis/llm/diagnostics.py

@@ -13,6 +13,7 @@ from uuid import uuid4
 from PIL import Image
 
 from artemis.data_engine.context_vars import CURRENT_NODE_NAME, CURRENT_TRACE_ID
+from artemis.llm.argument_fingerprint import ArgumentFingerprint, MAX_TRACKED_TOOLS
 from artemis.llm.final_review import ACTIVE_FINAL_REVIEW, FinalReviewBudgetExceeded
 from artemis.llm.reliability import LLMExhaustedError
 from artemis.llm.usage import usage_from_message
@@ -107,6 +108,7 @@ class RequestAttempt:
         self.clock = time.monotonic()
         self.last_publish = self.clock
         self.tools: dict[int | str, dict] = {}
+        self.tool_fingerprints: dict[int | str, ArgumentFingerprint] = {}
         self.review = ACTIVE_FINAL_REVIEW.get()
         if self.review:
             self.review.request_started(request_id, attempt)
@@ -126,6 +128,9 @@ class RequestAttempt:
             "usage": None,
             "output_chunks": 0,
             "tool_argument_chars": 0,
+            "tool_observation_layer": "langchain_tool_call_chunks",
+            "argument_fingerprint_version": 1,
+            "tool_stats_truncated": False,
             "tools": [],
             "step_id": str(self.step_id) if self.step_id else None,
             "parent_trace_id": str(self.parent_id) if self.parent_id else None,
@@ -167,9 +172,10 @@ class RequestAttempt:
         self.data.update(
             headers_at=time.time(),
             http_status=response.status_code,
-            server_request_id=_identifier(
+            server_request_id=_identifier(response.headers.get("x-oneapi-request-id"))
-                response.headers.get("x-request-id") or response.headers.get("request-id")
+            or _identifier(response.headers.get("x-request-id"))
-            ),
+            or _identifier(response.headers.get("request-id")),
+            response_x_request_id=_identifier(response.headers.get("x-request-id")),
             server_trace_id=_identifier(response.headers.get("x-trace-id")),
             phase="waiting_output",
         )
@@ -196,18 +202,31 @@ class RequestAttempt:
             self.data["usage"] = usage
         tool_chunks = getattr(chunk, "tool_call_chunks", None) or []
         # Responses exposes the same argument delta in content and tool_call_chunks.
-        # Count only the latter; retain lengths/names, never argument bodies.
+        # Count only the latter; retain bounded fingerprints, never argument bodies.
         for tool in tool_chunks:
+            arguments = tool.get("args")
+            if isinstance(arguments, str):
+                self.data["tool_argument_chars"] += len(arguments)
             index = tool.get("index")
-            key = index if index is not None else tool.get("id") or "unknown"
+            call_id = _identifier(tool.get("id"))
-            stats = self.tools.setdefault(key, {"name": None, "argument_chars": 0})
+            key = index if index is not None else call_id or "unknown"
+            if key not in self.tools and len(self.tools) >= MAX_TRACKED_TOOLS:
+                self.data["tool_stats_truncated"] = True
+                continue
+            stats = self.tools.setdefault(key, {"name": None})
+            if call_id:
+                stats["call_id"] = call_id
+            if index is not None:
+                stats["chunk_index"] = index
             name = tool.get("name")
             if isinstance(name, str) and re.fullmatch(r"[\w.:-]{1,128}", name):
                 stats["name"] = name
-            arguments = tool.get("args")
+            fingerprint = self.tool_fingerprints.get(key)
+            if fingerprint is None:
+                fingerprint = self.tool_fingerprints[key] = ArgumentFingerprint()
             if isinstance(arguments, str):
-                stats["argument_chars"] += len(arguments)
+                fingerprint.observe(arguments)
-                self.data["tool_argument_chars"] += len(arguments)
+            stats.update(fingerprint.snapshot())
         self.data["tools"] = list(self.tools.values())
         content = getattr(chunk, "content", None)
         substantive = isinstance(content, str) and bool(content)
@@ -252,7 +271,9 @@ class RequestAttempt:
             if isinstance(status, int):
                 self.data["http_status"] = status
             if request_id := _identifier(getattr(error, "request_id", None)):
-                self.data["server_request_id"] = request_id
+                self.data["error_request_id"] = request_id
+                if not self.data.get("server_request_id"):
+                    self.data["server_request_id"] = request_id
             if category == "timeout":
                 self.data.setdefault("timeout_scope", "transport")
         if response is not None:
@@ -301,6 +322,7 @@ async def observe_request_headers(request):
     if attempt is None:
         return
     request.headers.setdefault("X-Client-Request-Id", attempt.data["request_id"])
+    request.headers.setdefault("X-Client-Attempt-Id", attempt.data["attempt_id"])
     routing = {}
     for name in ("thread-id", "thread_id", "session-id", "session_id"):
         value = request.headers.get(name)
@@ -309,9 +331,12 @@ async def observe_request_headers(request):
     attempt.data.update(
         request_sent_at=time.time(),
         client_request_id=_identifier(request.headers.get("X-Client-Request-Id")),
+        client_attempt_id=_identifier(request.headers.get("X-Client-Attempt-Id")),
         routing_header_fingerprints=routing,
         client_request_id_matches=request.headers.get("X-Client-Request-Id")
         == attempt.data["request_id"],
+        client_attempt_id_matches=request.headers.get("X-Client-Attempt-Id")
+        == attempt.data["attempt_id"],
     )
     attempt.publish()
 

artemis/llm/reliability.py

@@ -153,7 +153,7 @@ class LLMPermanentError(LLMCallError):
 
 
 class LLMExhaustedError(LLMCallError):
-    """Retryable failures exhausted every attempt and the pause deadline."""
+    """Retryable failures exhausted attempts and any applicable pause deadline."""
 
 
 # ---------------------------------------------------------------------------

artemis/llm/stream_guard.py

@@ -23,7 +23,7 @@ class ToolArgumentOverflow(ValueError):
             "limit_chars": SMALL_ACTION_ARGUMENT_LIMIT,
         }
         super().__init__(
-            f"模型生成的 {name} 参数异常膨胀（{characters} 字符，"
+            f"收到的 {name} 参数流超出安全限制（{characters} 字符，"
             f"保护上限 {SMALL_ACTION_ARGUMENT_LIMIT}）；已关闭输出流，未执行本次动作。"
         )
 

artemis/services/llm.py

@@ -381,7 +381,7 @@ async def _run_with_recovery[T](
       immediately: retrying cannot help and pausing would hang the task.
     - When retryable attempts are exhausted: if a fallback model is waiting
       (with_fallback), raise LLMExhaustedError so it takes over immediately;
-      final acceptance also raises instead of pausing an already-finished task.
+      final acceptance and invalid output also raise instead of globally pausing.
       Other calls pause (bounded by settings.LLM_PAUSE_TIMEOUT_SECONDS) and
       retry the same request on resume.
     """
@@ -470,7 +470,7 @@ async def _run_with_recovery[T](
                     )
                     if failure.category is FailureCategory.INVALID_OUTPUT:
                         # One fresh response can recover a generation loop. Repeated
-                        # runaway output must reach the existing fallback/pause path,
+                        # runaway output must reach fallback or fail this request,
                         # even when ordinary network retries are configured higher.
                         max_attempts = min(
                             max_attempts, total_attempts + max(0, policy.max_attempts - attempt)
@@ -505,7 +505,11 @@ async def _run_with_recovery[T](
                     ACTIVE_ATTEMPT.reset(observation_token)
 
             # Retryable attempts exhausted.
-            if _FALLBACK_AVAILABLE.get() or ACTIVE_FINAL_REVIEW.get() is not None:
+            if (
+                _FALLBACK_AVAILABLE.get()
+                or ACTIVE_FINAL_REVIEW.get() is not None
+                or last_failure.category is FailureCategory.INVALID_OUTPUT
+            ):
                 if failed_endpoints is not None and endpoint_key:
                     failed_endpoints[endpoint_key] = last_error
                 _record_llm_event(
@@ -514,7 +518,11 @@ async def _run_with_recovery[T](
                         "error": str(last_error)[:1000],
                         "category": last_failure.category.value,
                         "retryable": True,
-                        "handover": "fallback" if _FALLBACK_AVAILABLE.get() else "final_review_end",
+                        "handover": "fallback"
+                        if _FALLBACK_AVAILABLE.get()
+                        else "final_review_end"
+                        if ACTIVE_FINAL_REVIEW.get() is not None
+                        else "task_failed",
                     },
                     status="failed",
                 )
@@ -1253,7 +1261,7 @@ async def _with_fallback[T](
         )
 
     # The contextvar must be reset BEFORE fallback_call runs: the fallback has
-    # no further fallback behind it, so its own exhaustion should pause.
+    # no further fallback behind it, so apply its category's terminal policy.
     fallback_token = _FALLBACK_AVAILABLE.set(True)
     try:
         result = await main_call()

docs/llm-stream-diagnostics.md

@@ -0,0 +1,39 @@
+# 模型流式参数诊断
+
+小型内置动作的参数流仍受 16384 字符保护。超限时关闭输出流、丢弃不完整参数，不执行该次动作；一次新响应重试和可用备用端点耗尽后，只结束当前任务，不进入共享的模型暂停。批次仍遵循“失败即停止”、人工暂停和取消设置。网络故障、认证错误和最终验收保留各自的恢复规则。
+
+## 请求关联
+
+每个模型请求尝试沿用现有 `llm_diagnostic` 记录，不另建日志数据库。
+
+| 字段 / 请求头 | 含义 |
+| --- | --- |
+| `request_id` / `X-Client-Request-Id` | 同一次恢复循环内保持稳定的逻辑请求编号；切换备用调用或恢复暂停后可能建立新的逻辑请求 |
+| `attempt_id` / `X-Client-Attempt-Id` | 本次客户端尝试编号；重试时变化，`previous_attempt_id` 关联前次尝试 |
+| `client_*_id_matches` | 实际发送的请求头是否与本地编号一致；调用方已指定的头保留原值 |
+| `server_request_id` | 首个有效的 `X-Oneapi-Request-Id`、`X-Request-Id` 或 `Request-Id`，按此前后顺序选择 |
+| `response_x_request_id` | 响应中的原始 `X-Request-Id`，不推定它来自模型上游 |
+| `error_request_id` | SDK 异常带回的编号，不覆盖已经从响应头读取的编号 |
+| `response_id` | SDK 实际返回的 Responses 响应编号 |
+
+编号只接收有长度限制的可打印白名单字符。请求头不控制会话亲和性，不替换 `thread-id` / `session-id`。网关自身的重试应使用独立编号或序号，不能假装是一次新的客户端尝试。
+
+## 参数指纹 v1
+
+`tool_observation_layer=langchain_tool_call_chunks` 明确表示客户端观测的是 SDK 转换后的参数片段，不是原始 SSE 帧。每工具记录实际可得的 `call_id`、`chunk_index`、工具名，以及：
+
+- `argument_chars`：参数片段的 Unicode 字符数之和。
+- `argument_bytes` / `argument_sha256`：按原顺序连接片段后，其 UTF-8 字节长度和 SHA256；不解析、重排参数 JSON，也不规范化空白或 Unicode。
+- `argument_deltas`：非空参数片段数，不能直接与网关原始事件数比较。
+- `argument_prefixes`：每 4096 字节的累计 SHA256，最多保留前 16 个。可以在分片大小不同、上游预读或中途取消时比较相同字节位置。
+- `argument_encoding_valid`：遇到不能编码为 UTF-8 的片段时为 false，当前总字节数和总哈希留空，避免伪造可比结果；诊断本身不改变执行流程。
+
+最多记录 64 个工具；超过时设置 `tool_stats_truncated=true`，总参数字符计数仍继续。诊断不保留参数正文。哈希仅证明已收到的字节前缀，不能证明输出完整或动作成功；完整性仍由响应终止状态和解析校验决定。
+
+跨三端优先使用 `response_id + call_id` 关联。LangChain 的 `chunk_index` 可能因文本或推理子块不同于原始 `output_index`，缺失编号时不能按数组位置猜测。正常的 `arguments.done`、`output_item.done`、`response.completed` 快照不重复计入参数；自定义工具转换、非空初始参数或协议异常需要先核对 SDK 映射，不能仅凭跨层哈希不同认定某一端重复发送。
+
+历史日志没有保存的原始帧无法由新增诊断恢复。HTTP 200 只说明响应头已到达；客户端保护后出现的 499 或 Broken pipe 也不能证明最初原因。网关写入本机下游 writer 成功不等于客户端已消费或执行。
+
+## 离线验证
+
+`tests/unit/test_llm_stream_guard.py` 验证保护、有限重试、相同/不同备用端点，以及不会创建或清除其他请求的全局暂停。`tests/unit/test_request_diagnostics.py` 验证请求编号、超时重试、响应头及单次参数计数。`tests/unit/test_argument_fingerprint.py` 验证不同分片、中文和 emoji、固定字节前缀、有界统计与异常编码。这些测试不调用真实模型或设备。

docs/releases/2.4.2.md

@@ -0,0 +1,24 @@
+# ARTEMIS 2.4.2
+
+本版为批次任务增加人工评审标记，并调整任务列表布局。人工评审插件保持 **1.1.0**，第三方依赖版本不变。
+
+## 更新
+
+- 安装并启用人工评审插件后，任务存在已保存的有效评审内容时，批次列表显示“人工已审”。评审结论或非空备注均计入；清空全部内容后隐藏，禁用或卸载插件后也不显示。
+- 标记仅说明已有人工评审内容，不代表测试通过，不覆盖 AI 判定或改变任务执行流程。没有安装插件的用户不显示该标记。
+- 批次任务标题与状态分行显示：收藏星标位于标题框内，执行状态、AI 判定和人工已审标签在下方等高、左对齐排列。此布局适用于所有用户。
+- 按批次一次读取当前评审内容，沿用列表自动刷新，不逐条增加网络请求；重跑任务不继承旧任务的已审标记。
+- 补充发版约束：应用与第三方依赖独立版本化，检查锁文件差异，并验证最终源码包的全新安装。
+
+## 验证
+
+- 默认 Python 回归：3269 通过，2 跳过，8 按标记排除；Ruff、质量基线和核心类型检查通过。
+- 前端 ChromeHeadless：348 项通过；生产构建通过。
+- 内置浏览器使用隔离数据检查收藏星标、两行布局及有评审/无评审标签显示。未写入真实任务评审数据。
+- 最终源码包另行执行独立目录、空 npm 缓存的官方源和国内镜像安装及生产构建，结果随发布附件提供。
+
+## 升级
+
+已有安装请等待任务结束，运行 `update-cn.bat` 或 `update.bat` 后重启。新安装下载 `artemis-2.4.2-source.zip`，按 `SHA256SUMS.txt` 校验后启动。请勿直接覆盖旧安装目录。
+
+人工评审插件仍需单独安装并启用；本次无需升级已安装的 1.1.0。发布过程不构建 Docker，不重启用户服务，不执行设备或真实业务任务。

docs/releases/2.4.3.md

@@ -0,0 +1,25 @@
+# ARTEMIS 2.4.3
+
+本版修复异常模型输出影响整个批次的问题，并增加流式参数诊断。人工评审插件保持 **1.1.0**，SDK、其他插件及第三方依赖版本不变。
+
+## 更新
+
+- 已识别的异常输出（包括工具参数溢出、响应完整性校验失败）在有限重试和可用备用端点耗尽后，只结束当前任务，不再创建全局模型暂停。批次未开启“失败即停止”时可继续下一条；人工暂停、失败即停止、网络恢复及最终验收的原有规则保留。
+- 保留 16384 字符工具参数保护，关闭异常输出流，丢弃未完成内容，不执行本次异常动作。报错改为描述“收到的参数流”，避免在证据不足时直接归责模型。
+- 增加单次尝试请求头，区分逻辑请求、客户端尝试、网关响应与 SDK 异常中的请求编号，便于跨服务核对。
+- 对收到的工具参数片段记录字符数、UTF-8 字节数及有界 SHA256 前缀摘要，默认不新增参数正文记录。最多跟踪 64 个工具，每工具最多保留 16 个固定位置前缀。SDK 的 chunk_index 与原始协议 output_index 分开命名。
+
+诊断字段及边界见 [流式参数诊断](../llm-stream-diagnostics.md)。本次主程序安装不会部署或升级 NEWAPI、CodexManager；服务端修复独立交付。
+
+## 验证
+
+- 默认 Python 回归：3284 通过，6 跳过，8 按标记排除；Ruff 格式和静态检查、质量基线、核心类型检查通过。
+- 五类合成输入经过实际 CodexManager、NEWAPI 转发函数及原 OpenAI/LangChain SDK 的离线文件串接：正常输入有效；超长、累计全文、重复及半关闭输入被拒绝；各层参数摘要一致，设备动作执行数为 0。
+- 最终源码 ZIP 的隔离解压、独立空 npm 缓存、官方源与国内镜像安装和生产构建结果随发布附件 `INSTALL-VERIFICATION.md` 提供。
+- 未重跑用户原业务批次、未做高并发压力测试或真实模型端到端验证。历史原始参数帧未保存，不能据此认定原事故参数膨胀首先发生在哪一层。
+
+## 升级
+
+已有安装请等待任务结束，运行 `update-cn.bat` 或 `update.bat` 后重启。新安装下载 `artemis-2.4.3-source.zip`，按 `SHA256SUMS.txt` 校验后启动。请勿直接覆盖旧安装目录。
+
+人工评审插件 1.1.0 无需更新。本次发布不构建 Docker，不重启用户服务，不执行设备或真实业务任务。

pyproject.toml

@@ -14,7 +14,7 @@
 
 [project]
 name = "artemis"
-version = "2.4.1"
+version = "2.4.3"
 description = "Autonomous AI Assistant for Mobile Automation"
 
 authors = [

tests/unit/admin_console/test_task_extensions.py

@@ -57,15 +57,16 @@ def manifest_payload():
     }
 
 
-def install(plugins, original_case=False):
+def install(plugins, original_case=False, plugin_id="review"):
     package = BytesIO()
     manifest = manifest_payload()
+    manifest["id"] = plugin_id
     if original_case:
         manifest["workspace"]["panels"][0]["kind"] = "original_case"
     with zipfile.ZipFile(package, "w") as archive:
         archive.writestr("plugin.json", json.dumps(manifest))
     install_archive(plugins, package.getvalue())
-    plugins.set_enabled("review", True)
+    plugins.set_enabled(plugin_id, True)
 
 
 @pytest.fixture
@@ -108,6 +109,71 @@ def save(client, values, revision=0, request_id=None):
     )
 
 
+def test_batch_review_badge_tracks_current_values_and_plugin_lifecycle(context, monkeypatch):
+    from apps.admin_console.routers import automation
+
+    store, plugins, client = context
+    monkeypatch.setattr(automation, "store", store)
+    result = {"status": "inconclusive", "source": "execution"}
+    monkeypatch.setattr(automation, "read_reviewed_result", lambda *args, **kwargs: result)
+    client.app.include_router(automation.router)
+    store.create_batch({"id": "review-batch", "status": "completed"}, [], {})
+    store.save_job(
+        {"session_id": "batch-task", "batch_id": "review-batch", "status": "completed", "case": {}}
+    )
+    store.save_job(
+        {"session_id": "new-run", "batch_id": "review-batch", "status": "completed", "case": {}}
+    )
+
+    def badge():
+        response = client.get("/api/automation/batches/review-batch")
+        assert response.status_code == 200
+        jobs = {job["session_id"]: job for job in response.json()["jobs"]}
+        assert "batch-task" in jobs, response.json()
+        assert jobs["batch-task"]["test_result"] == result
+        assert jobs["new-run"]["human_reviewed"] is False
+        return jobs["batch-task"]["human_reviewed"]
+
+    assert badge() is False
+    install(plugins)
+    assert (
+        client.put(
+            "/api/task-extensions/batch-task/review",
+            json={
+                "values": {"accuracy": "Pass"},
+                "expected_revision": 0,
+                "request_id": str(uuid4()),
+            },
+        ).status_code
+        == 200
+    )
+    assert badge() is False  # Other workspace plugins are not the human review plugin.
+    install(plugins, plugin_id="task-review")
+    assert badge() is False
+    for revision, (values, expected) in enumerate(
+        [
+            ({}, False),
+            ({"accuracy_note": "  \n"}, False),
+            ({"accuracy": "Pass"}, True),
+            ({"accuracy": "Failed"}, True),
+            ({"accuracy": "", "accuracy_note": ""}, False),
+            ({"accuracy_note": "人工核对了截图"}, True),
+        ]
+    ):
+        response = client.put(
+            "/api/task-extensions/batch-task/task-review",
+            json={"values": values, "expected_revision": revision, "request_id": str(uuid4())},
+        )
+        assert response.status_code == 200
+        assert badge() is expected
+    plugins.set_enabled("task-review", False)
+    assert badge() is False
+    plugins.set_enabled("task-review", True)
+    assert badge() is True
+    plugins.uninstall("task-review")
+    assert badge() is False
+
+
 def test_original_case_import_requires_enabled_plugin_and_keeps_execution_prompt(context):
     from apps.admin_console.services import task_import
 

tests/unit/test_argument_fingerprint.py

@@ -0,0 +1,59 @@
+import hashlib
+import json
+
+import pytest
+from langchain_core.messages import AIMessageChunk
+
+from artemis.llm.argument_fingerprint import ArgumentFingerprint, MAX_PREFIXES, PREFIX_BYTES
+from artemis.llm.diagnostics import RequestAttempt
+
+
+@pytest.mark.parametrize("fragment_size", [1, 37, 4096, 100000])
+def test_fingerprints_match_at_byte_offsets_despite_unicode_and_different_fragmentation(
+    fragment_size,
+):
+    arguments = '{"target":"' + "私人内容🙂" * 8000 + '"}'
+    expected = arguments.encode("utf-8")
+    fingerprint = ArgumentFingerprint()
+    for offset in range(0, len(arguments), fragment_size):
+        fingerprint.observe(arguments[offset : offset + fragment_size])
+    result = fingerprint.snapshot()
+    assert result["argument_chars"] == len(arguments)
+    assert result["argument_bytes"] == len(expected)
+    assert result["argument_sha256"] == hashlib.sha256(expected).hexdigest()
+    assert result["argument_encoding_valid"] is True
+    assert result["argument_prefixes"] == [
+        {"bytes": size, "sha256": hashlib.sha256(expected[:size]).hexdigest()}
+        for size in range(PREFIX_BYTES, PREFIX_BYTES * MAX_PREFIXES + 1, PREFIX_BYTES)
+    ]
+    assert "私人" not in json.dumps(result, ensure_ascii=False)
+
+
+def test_invalid_utf8_is_marked_without_failing_or_forging_a_digest():
+    fingerprint = ArgumentFingerprint()
+    for value in ["safe", "\ud800", "later"]:
+        fingerprint.observe(value)
+    assert fingerprint.snapshot() == {
+        "argument_chars": 10,
+        "argument_bytes": None,
+        "argument_sha256": None,
+        "argument_deltas": 3,
+        "argument_prefixes": [],
+        "argument_encoding_valid": False,
+    }
+
+
+def test_tool_tracking_is_bounded_but_total_count_keeps_all_received_arguments():
+    attempt = RequestAttempt(
+        None, request_id="bounded", attempt=1, timeout=300, max_retries=0, metadata={}
+    )
+    for index in range(65):
+        attempt.chunk(
+            AIMessageChunk(
+                content=[],
+                tool_call_chunks=[{"name": "click", "args": "{}", "index": index}],
+            )
+        )
+    assert len(attempt.data["tools"]) == len(attempt.tool_fingerprints) == 64
+    assert attempt.data["tool_stats_truncated"] is True
+    assert attempt.data["tool_argument_chars"] == 130

tests/unit/test_llm_response_integrity.py

@@ -2,6 +2,7 @@
 
 import asyncio
 import json
+import hashlib
 from types import SimpleNamespace
 from unittest.mock import AsyncMock
 
@@ -15,6 +16,7 @@ import pytest_asyncio
 from artemis.llm.reliability import FailureCategory, LLMExhaustedError
 from artemis.llm.response_guard import ResponseIntegrityError, ensure_response_complete
 from artemis.llm.router import ModelEndpoint, ModelFactory
+from artemis.llm.diagnostics import RequestAttempt
 from artemis.services import llm as llm_service
 from artemis.services.llm import RobustChatModelWrapper, acomplete
 
@@ -195,6 +197,33 @@ async def test_completed_stream_returns_original_valid_arguments(transport_case)
     assert result.tool_calls[0]["args"] == {"passed": True}
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize("transport_case", ["fragmented"], indirect=True)
+async def test_real_sdk_fingerprint_counts_received_arguments_once(transport_case):
+    case = transport_case
+    arguments = json.dumps({"passed": True, "evidence": "中文🙂" * 1000}, ensure_ascii=False)
+    case.outputs.append(("completed", arguments))
+    observation = RequestAttempt(
+        None, request_id="sdk-test", attempt=1, timeout=300, max_retries=0, metadata={}
+    )
+    merged = None
+    async for chunk in case.model.astream([]):
+        observation.chunk(chunk)
+        merged = chunk if merged is None else merged + chunk
+    observation.finish("completed", response=merged)
+    assert len(observation.data["tools"]) == 1
+    tool = observation.data["tools"][0]
+    assert tool["call_id"] == "call_test"
+    assert tool["argument_chars"] == len(arguments)
+    assert tool["argument_bytes"] == len(arguments.encode())
+    assert tool["argument_sha256"] == hashlib.sha256(arguments.encode()).hexdigest()
+    assert tool["argument_prefixes"][0] == {
+        "bytes": 4096,
+        "sha256": hashlib.sha256(arguments.encode()[:4096]).hexdigest(),
+    }
+    assert tool["argument_deltas"] == 1
+
+
 @pytest.mark.asyncio
 @pytest.mark.parametrize("transport_case", ["fragmented"], indirect=True)
 @pytest.mark.parametrize("status", ["completed", None])

tests/unit/test_llm_stream_guard.py

@@ -124,7 +124,9 @@ async def test_gateway_closes_bad_stream_discards_partial_and_bounds_retry(
         with pytest.raises(LLMExhaustedError) as caught:
             await wrapper.complete([])
         assert caught.value.failure.category is FailureCategory.INVALID_OUTPUT
-        llm._wait_for_resume.assert_awaited_once()
+        llm._wait_for_resume.assert_not_awaited()
+        assert not llm.PAUSE_FILE.exists()
+        assert any(name == "llm_gave_up" and p["handover"] == "task_failed" for name, p in resets)
     assert model.calls == 2
     assert closed == [1, 2]
     failure = next(
@@ -139,3 +141,47 @@ async def test_gateway_closes_bad_stream_discards_partial_and_bounds_retry(
     assert failure["usage"] is None
     assert "private" not in json.dumps(records + resets, default=str)
     assert any(name == "llm_stream_reset" and p["action"] == "discard" for name, p in resets)
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("fallback_kind", ["same", "failing", "working"])
+@pytest.mark.parametrize("already_paused", [False, True])
+async def test_invalid_output_fallback_does_not_pause_or_clear_another_request(
+    fallback_kind, already_paused
+):
+    calls = []
+    if already_paused:
+        llm.PAUSE_FILE.write_text("another request or user pause", encoding="utf-8")
+
+    async def main():
+        calls.append("main")
+        raise ToolArgumentOverflow("click", SMALL_ACTION_ARGUMENT_LIMIT + 1)
+
+    async def fallback():
+        calls.append("fallback")
+        if fallback_kind == "working":
+            return "valid response"
+        raise ToolArgumentOverflow("click", SMALL_ACTION_ARGUMENT_LIMIT + 2)
+
+    async def run():
+        return await llm.with_fallback(
+            lambda: llm._run_with_recovery(main, endpoint_key="main", max_retries=5),
+            lambda: llm._run_with_recovery(
+                fallback,
+                endpoint_key="main" if fallback_kind == "same" else "fallback",
+                max_retries=5,
+            ),
+        )
+
+    if fallback_kind == "working":
+        assert await run() == "valid response"
+    else:
+        with pytest.raises(LLMExhaustedError) as caught:
+            await run()
+        assert caught.value.failure.category is FailureCategory.INVALID_OUTPUT
+    assert calls.count("main") == 2
+    assert calls.count("fallback") == {"same": 0, "failing": 2, "working": 1}[fallback_kind]
+    llm._wait_for_resume.assert_not_awaited()
+    assert llm.PAUSE_FILE.exists() is already_paused
+    if already_paused:
+        assert llm.PAUSE_FILE.read_text(encoding="utf-8") == "another request or user pause"

tests/unit/test_request_diagnostics.py

@@ -1,6 +1,7 @@
 import asyncio
 from copy import deepcopy
 import json
+import hashlib
 from types import SimpleNamespace
 from uuid import uuid4
 
@@ -36,6 +37,46 @@ class Engine:
         pass
 
 
+@pytest.mark.parametrize(
+    ("headers", "expected"),
+    [
+        ({"X-Oneapi-Request-Id": "gateway-request"}, "gateway-request"),
+        (
+            {"X-Oneapi-Request-Id": "gateway-request", "X-Request-Id": "provider-request"},
+            "gateway-request",
+        ),
+        ({"X-Request-Id": "provider-request"}, "provider-request"),
+        ({"Request-Id": "legacy-request"}, "legacy-request"),
+        (
+            {"X-Oneapi-Request-Id": "invalid header value", "X-Request-Id": "standard-request"},
+            "standard-request",
+        ),
+    ],
+)
+def test_gateway_request_id_preserves_the_standard_header_without_assuming_its_source(
+    headers, expected
+):
+    attempt = RequestAttempt(
+        Engine(), request_id="local-request", attempt=1, timeout=300, max_retries=1, metadata={}
+    )
+    attempt.headers(httpx.Response(200, headers=headers))
+    assert attempt.data["server_request_id"] == expected
+    assert attempt.data["response_x_request_id"] == headers.get("X-Request-Id")
+
+
+def test_sdk_error_cannot_overwrite_the_observed_gateway_request_id():
+    class RequestError(Exception):
+        request_id = "sdk-request"
+
+    attempt = RequestAttempt(
+        Engine(), request_id="local-request", attempt=1, timeout=300, max_retries=1, metadata={}
+    )
+    attempt.headers(httpx.Response(400, headers={"X-Oneapi-Request-Id": "gateway-request"}))
+    attempt.finish("failed", error=RequestError("failed"))
+    assert attempt.data["server_request_id"] == "gateway-request"
+    assert attempt.data["error_request_id"] == "sdk-request"
+
+
 @pytest.mark.asyncio
 async def test_request_hook_correlates_api_calls_without_logging_header_values():
     engine = Engine()
@@ -55,13 +96,18 @@ async def test_request_hook_correlates_api_calls_without_logging_header_values()
         )
         await observe_request_headers(request)
         assert request.headers["x-client-request-id"] == "local-request"
+        assert request.headers["x-client-attempt-id"] == str(attempt.trace_id)
+        assert attempt.data["client_attempt_id_matches"] is True
         assert attempt.data["client_request_id_matches"] is True
         assert len(attempt.data["routing_header_fingerprints"]) == 2
         assert "private" not in json.dumps(engine.records, default=str)
         request.headers["X-Client-Request-Id"] = "caller-id"
+        request.headers["X-Client-Attempt-Id"] = "caller-attempt"
         await observe_request_headers(request)
         assert request.headers["x-client-request-id"] == "caller-id"
         assert attempt.data["client_request_id_matches"] is False
+        assert request.headers["X-Client-Attempt-Id"] == "caller-attempt"
+        assert attempt.data["client_attempt_id_matches"] is False
     finally:
         ACTIVE_ATTEMPT.reset(token)
 
@@ -75,6 +121,7 @@ async def test_headers_without_body_timeout_then_retry_keeps_evidence(monkeypatc
     )
     llm._ENDPOINT_BREAKER._states.clear()
     calls = 0
+    wire_ids = []
 
     class StalledBody(httpx.AsyncByteStream):
         async def __aiter__(self):
@@ -84,6 +131,9 @@ async def test_headers_without_body_timeout_then_retry_keeps_evidence(monkeypatc
     async def transport(request):
         nonlocal calls
         calls += 1
+        wire_ids.append(
+            (request.headers["X-Client-Request-Id"], request.headers["X-Client-Attempt-Id"])
+        )
         headers = {"x-request-id": f"server-{calls}", "set-cookie": "secret-cookie"}
         return httpx.Response(
             200, headers=headers, stream=StalledBody() if calls == 1 else httpx.ByteStream(b"ok")
@@ -91,7 +141,7 @@ async def test_headers_without_body_timeout_then_retry_keeps_evidence(monkeypatc
 
     async with httpx.AsyncClient(
         transport=httpx.MockTransport(transport),
-        event_hooks={"response": [observe_response_headers]},
+        event_hooks={"request": [observe_request_headers], "response": [observe_response_headers]},
     ) as client:
 
         class Model:
@@ -123,6 +173,10 @@ async def test_headers_without_body_timeout_then_retry_keeps_evidence(monkeypatc
     assert final["request_id"] == failed["request_id"]
     assert final["attempt_id"] != failed["attempt_id"] and final["attempt"] == 2
     assert final["previous_attempt_id"] == failed["attempt_id"]
+    assert wire_ids == [
+        (failed["request_id"], failed["attempt_id"]),
+        (final["request_id"], final["attempt_id"]),
+    ]
     retry = next(r["payload"] for r in engine.records if r["name"] == "llm_retry")
     assert retry["request_id"] == failed["request_id"]
     assert retry["attempt_id"] == failed["attempt_id"]
@@ -163,10 +217,26 @@ def test_tool_deltas_are_counted_once_without_retaining_arguments():
     attempt.finish("completed", response=combined)
     assert attempt.data["output_chunks"] == 4
     assert attempt.data["tool_argument_chars"] == len('{"secret":"private"}{"seconds":1}')
-    assert attempt.data["tools"] == [
+    for tool, name, arguments, call_id, index, count in zip(
-        {"name": "click", "argument_chars": len('{"secret":"private"}')},
+        attempt.data["tools"],
-        {"name": "wait", "argument_chars": len('{"seconds":1}')},
+        ["click", "wait"],
-    ]
+        ['{"secret":"private"}', '{"seconds":1}'],
+        ["c1", "c2"],
+        [0, 1],
+        [2, 1],
+        strict=True,
+    ):
+        assert tool == {
+            "name": name,
+            "call_id": call_id,
+            "chunk_index": index,
+            "argument_chars": len(arguments),
+            "argument_bytes": len(arguments.encode()),
+            "argument_sha256": hashlib.sha256(arguments.encode()).hexdigest(),
+            "argument_deltas": count,
+            "argument_prefixes": [],
+            "argument_encoding_valid": True,
+        }
     assert attempt.data["response_status"] == "completed"
     assert "private" not in json.dumps(engine.records, default=str)
     assert "secret" not in json.dumps(engine.records, default=str)
@@ -205,7 +275,10 @@ async def test_timeout_during_tool_generation_keeps_last_output_and_tool_name(mo
     )
     assert failed["failure_phase"] == "generating_tool_arguments"
     assert failed["last_output_at"] >= failed["first_output_at"]
-    assert failed["tools"] == [{"name": "click", "argument_chars": 5}]
+    assert len(failed["tools"]) == 1
+    tool = failed["tools"][0]
+    assert tool["name"] == "click" and tool["argument_chars"] == tool["argument_bytes"] == 5
+    assert tool["argument_sha256"] == hashlib.sha256(b'{"x":').hexdigest()
     assert failed["usage"] is None
     assert failed["max_output_tokens"] == 12000
 

uv.lock

@@ -248,7 +248,7 @@ wheels = [
 
 [[package]]
 name = "artemis"
-version = "2.4.1"
+version = "2.4.3"
 source = { editable = "." }
 dependencies = [
     { name = "adbutils" },