task.json

[
  {
    "id": "task-001",
    "title": "搭建 Agent Harness 基线",
    "goal": "创建可恢复、可审计的 AI 自主开发系统骨架。",
    "steps": [
      "建立 CLAUDE.md、task.json、progress.jsonl 和 runs 日志目录",
      "实现 pending/running/blocked/review/validated/deployed 任务状态机",
      "暴露 /api/agent/status 给前端读取任务与运行日志"
    ],
    "acceptance": [
      "任务可从 JSON 读取",
      "运行日志可跨刷新保留",
      "页面显示 Provider 能力检测结果"
    ],
    "status": "deployed",
    "provider": "claude",
    "artifact": "/agent",
    "updatedAt": "2026-05-07T02:20:00.000Z"
  },
  {
    "id": "task-002",
    "title": "实现 OpenClaw Skill 商店",
    "goal": "交付可浏览、搜索、筛选、查看详情和模拟安装的 Skill Marketplace。",
    "steps": [
      "建立 SkillManifest 数据结构与种子数据",
      "实现 /store、/store/[slug] 和安装记录 API",
      "实现发布表单与 SKILL.md 预览"
    ],
    "acceptance": [
      "至少 6 个 OpenClaw Skill 可浏览",
      "搜索和分类筛选可用",
      "安装按钮会写入运行日志"
    ],
    "status": "deployed",
    "provider": "claude",
    "artifact": "/store",
    "updatedAt": "2026-05-07T02:26:00.000Z"
  },
  {
    "id": "task-003",
    "title": "接入 Claude / Codex 双执行后端",
    "goal": "让 Demo 能检测并拉起本地 coding agent CLI，形成真实自循环入口。",
    "steps": [
      "检测 claude -p 非交互能力",
      "检测 npm 包 @openai/codex 或 Windows Codex CLI 能力",
      "实现 /api/agent/run，将 stdout/stderr 写入 runs 日志"
    ],
    "acceptance": [
      "Claude 可用时能执行 smoke prompt",
      "Codex 不可用时给出明确降级说明",
      "失败不会破坏 task.json"
    ],
    "status": "validated",
    "provider": "claude",
    "artifact": "/api/agent/run",
    "updatedAt": "2026-05-07T03:22:05.489Z"
  },
  {
    "id": "task-004",
    "title": "生成部署与证据页",
    "goal": "为面试评审提供公网 URL、测试结果和项目复现方式。",
    "steps": [
      "实现 /proof 展示构建、浏览器测试和公网访问信息",
      "提供 Vercel 与 Cloudflare Quick Tunnel 脚本",
      "记录最终访问 URL 和本地复现命令"
    ],
    "acceptance": [
      "/proof 能看到访问路径",
      "README 写明本地与公网部署方式",
      "Playwright 覆盖核心页面"
    ],
    "status": "deployed",
    "provider": "codex",
    "artifact": "/proof",
    "updatedAt": "2026-05-07T03:44:45.852Z"
  },
  {
    "id": "task-005",
    "title": "补强真实 AI 自主开发证据链",
    "goal": "让评审方能看到一次 Codex Agent 读取任务、修改代码、运行测试并留下完整日志的真实闭环。",
    "steps": [
      "读取原题 5 项要求和 REQUIREMENT_COVERAGE.md",
      "在证据页增加真实 development run 说明",
      "新增 SELF_DEVELOPMENT_RUN.md 记录输入、变更、测试、产物和人工协作点",
      "更新 Playwright 覆盖新增证据文档"
    ],
    "acceptance": [
      "/proof 能看到 Real development run 证据",
      "SELF_DEVELOPMENT_RUN.md 说明 Codex Agent 的实际改动和测试",
      "npm run pycheck 通过",
      ".agent-state/runs 中保留本次 Codex development run 日志"
    ],
    "status": "validated",
    "provider": "codex",
    "artifact": "/proof",
    "updatedAt": "2026-05-07T04:34:00.000Z"
  },
  {
    "id": "task-006",
    "title": "面试级 AgentOps 升级",
    "goal": "把基础 Demo 升级为可现场演示的 AgentOps 产品：工作流图、能力矩阵、证明资产、演示脚本和浏览器录屏。",
    "dependsOn": [
      "task-001",
      "task-002",
      "task-003",
      "task-004",
      "task-005"
    ],
    "steps": [
      "引入 @xyflow/react 构建交互式任务依赖图",
      "重做首页、Agent Console、Store 与 Proof 的中文内容和视觉层级",
      "补充可验证的截图、录屏、README 与 QA 说明",
      "运行 lint、build、Playwright 并记录结果"
    ],
    "acceptance": [
      "首页展示可交互 AgentOps 总览",
      "/agent 展示任务图、Provider 能力、运行日志与任务详情",
      "/store 保留搜索、筛选、安装和 SKILL.md 预览闭环",
      "/proof 展示验证命令、截图/录屏产物路径和复现方式",
      "Playwright 桌面与移动端测试通过"
    ],
    "status": "validated",
    "provider": "codex",
    "artifact": "/",
    "updatedAt": "2026-05-09T04:35:00.000Z"
  },
  {
    "id": "task-007",
    "title": "智能体调控与全流程录制强化",
    "goal": "补强 AI 使用与调控展示：安装相关 Codex 技能，使用子智能体协作，新增智能体调控中心，录制完整功能巡检视频，并更新交付证据链。",
    "dependsOn": [
      "task-006"
    ],
    "steps": [
      "安装并记录 openai-docs、vercel-deploy、cloudflare-deploy、pdf 等 Codex 技能",
      "使用子智能体分别检查代码落点和调研 AgentOps 前沿模式",
      "新增 /control 智能体调控中心，展示 Agent mesh、技能、插件、协议、治理与运行状态",
      "扩展 scripts/capture-proof.mjs，按 /、/control、/agent、/store、/proof 录制完整巡检视频",
      "更新 /proof、README、QA、交付包说明和 Playwright 测试，形成可审计证据链"
    ],
    "acceptance": [
      "/control 能展示智能体分工、技能插件、协议与前沿 AgentOps 模式",
      "/proof 能引用 task-007、development-task-007-codex.jsonl、control-center.png 和 full-flow-demo.webm",
      "public/proof/full-flow-demo.webm 记录完整功能巡检流程",
      "public/proof/control-center.png 记录智能体调控中心截图",
      "npm run pycheck 通过，交付包包含最新源码、文档、截图和视频"
    ],
    "status": "validated",
    "provider": "codex",
    "artifact": "/control",
    "updatedAt": "2026-05-09T07:16:00.000Z"
  },
  {
    "id": "task-008",
    "title": "真人式演示录制与技能强化",
    "goal": "把自动巡检视频升级为面试官能看懂的真人式慢速演示：加入字幕、鼠标移动、逐字输入、滚动停顿、具体 memory-system 安装实例，并补装演示、观测、语音转写、部署和协作类 Codex 技能。",
    "dependsOn": [
      "task-007"
    ],
    "steps": [
      "安装 sentry、speech、transcribe、netlify-deploy、render-deploy、notion、linear、figma 等 Codex 技能",
      "把 scripts/capture-proof.mjs 改成带字幕、慢节奏、模拟真人使用的 Playwright 录制流程",
      "在录制中演示具体实例：搜索 memory、筛选 Memory、安装 memory-system、生成 release-qa-copilot 的 SKILL.md 预览",
      "更新 /control、/proof、README、QA 和交付说明，记录新技能和真人式演示视频",
      "重新运行 bug 检查、质量门、视频 metadata 验证并刷新交付包"
    ],
    "acceptance": [
      "full-flow-demo.webm 时长明显长于旧版短巡检，并带有可读字幕",
      "视频可见模拟真人操作：鼠标移动、点击、逐字输入、滚动和停顿",
      "/proof 引用 task-008 和 development-task-008-codex.jsonl",
      "/control 展示新增 sentry、speech、transcribe、netlify/render、notion/linear/figma 技能",
      "npm run pycheck 通过，交付包包含最新长视频和证据"
    ],
    "status": "validated",
    "provider": "codex",
    "artifact": "public/proof/full-flow-demo.webm",
    "updatedAt": "2026-05-09T08:05:00.000Z"
  }
]