Files
proxycast/internal/exec-plans/plan-runtime-implementation-progress.md
2026-07-12 15:49:43 +08:00

68 KiB
Raw Permalink Blame History

Plan Runtime Implementation Progress

状态:active 更新时间:2026-06-23 主路线图:internal/roadmap/plan/README.md PRD:internal/roadmap/plan/prd.md

目标

把 Lime Plan 能力从路线图推进到 current 主链实现:Codex 风格 Plan Mode + update_plan + <proposed_plan> 驱动 GUI 计划轨,同时支持多模型 reasoning / thinking 过程展示。

当前原则

  1. 先完成骨架,再填细节。
  2. 不新增平行 planner;只扩展 App Server / RuntimeCore / RuntimeBackend / AgentUI current 主链。
  3. 不恢复 lime-rs/src/**、旧 Tauri wrapper、旧 agent_runtime_* 生产事实源或 mock fallback。
  4. Plan facts 与 reasoning facts 分层:plan.delta/final 驱动计划轨,reasoning.delta/final 只进过程层。
  5. 多模型能力走标准化 resolver / adapter,不让前端业务组件拼 Provider 私有字段。

分阶段计划

P0:骨架与写集边界

  • 新增 Plan roadmap 与 PRD。
  • 新增骨架优先的 implementation plan。
  • 放行 internal/roadmap/plan/*.md,保证路线图可版本化。
  • 新增本执行计划并放行。
  • 建立后端 skeleton:model capability、provider request options、reasoning event、plan event builder。
  • 建立前端 skeleton:PlanState selector、model reasoning state selector。
  • 骨架层最小单测通过。

退出条件:

  • 骨架文件存在且只做保守归一 / no-op adapter,不接复杂运行时细节。
  • 不触碰当前并行脏写集中的 right-surface 实现。
  • 后端骨架模块暂时允许 dead_code,因为 P0 不接主链;P1 接入事件主链时移除或收窄该标注。
  • ReasoningLevel::XHigh 序列化固定为 xhigh,与前端 reasoning 档位归一保持一致。

P1:后端事件主链

  • Agent <proposed_plan> 流式 parser 输出 plan.delta/final。
  • tool.result(update_plan) 统一投影为 plan.final 或等价 plan fact。
  • thinking_delta 标准化为 reasoning.delta。
  • reasoning start / end / final 映射为标准 runtime events。
  • model.effective 记录本轮 provider/model/reasoning 有效能力。

退出条件:

  • Rust 定向测试覆盖 plan delta/final、update_plan metadata、reasoning event、model effective。

P2:前端协议与状态

  • AgentEvent 一等支持 plan_delta / plan_final / reasoning_delta。
  • AgentEvent 一等支持 model_effective。
  • AgentEvent 一等支持 reasoning_final。
  • live plan_delta / plan_final 可进入 AgentThreadItem type="plan"。
  • 建立 live events 与 ThreadReadModel 共用的 PlanState / ReasoningState hydrator 纯函数。
  • 标准 PlanState 只接收 live plan event 和带 revisionId 的历史 plan item,并清退旧 per-step thread item / update_plan metadata 计划恢复路径。
  • PlanRail 通过 TaskRail projection 消费标准 PlanState。
  • Runtime strip 通过 harnessState.plan 消费标准 PlanState。
  • Runtime UI 消费标准 ReasoningState。

退出条件:

  • 前端纯单元测试覆盖 event parser、PlanState reducer、reasoning hydrator。

P3:UI 工作台闭环

  • 输入栏 Plan Mode 状态、模型选择、reasoning 档位显示。
  • 对话内 <proposed_plan> 去重展示。
  • 右侧计划轨 / 运行控制面显示 latest plan revision。
  • 右侧计划轨、Runtime Strip 与 Harness 摘要不再从旧 update_plan metadata 或无 revision plan thread item 直读计划。
  • Plan Mode ready 后出现实施确认,确认绑定 latest revision。
  • thinking / reasoning 过程行默认折叠,不抢 PlanRail 主对象。
  • latest revision / 实施确认绑定文案覆盖五语言。
  • thinking / reasoning 折叠相关文案覆盖五语言。

退出条件:

  • 用户可见回归测试覆盖计划轨、实施确认、模型/推理档位。
  • GUI smoke 覆盖显式 Plan Mode 和默认 update_plan。

P4:多模型验证

  • Codex/OpenAI fixture。
  • Anthropic fixture。
  • Gemini fixture。
  • OpenAI-compatible fixture。
  • 无 reasoning 模型降级 fixture。

退出条件:

  • 不支持 thinking 的模型不伪造 reasoning。
  • 支持 reasoning 的模型能保留过程 event,PlanRail 仍只消费 plan facts。

P5:provider-native reasoning metadata 保真

  • Agent ThinkingContent.signature 投影到 ItemRuntimePayload::Reasoning.metadata.provider_metadata.signature。
  • ItemRuntimePayload::Reasoning.metadata 透传到 AgentThreadItemPayload::Reasoning.metadata。
  • App Server thread read model 合并 reasoning item payload 中的 metadata.provider_metadata。
  • App Server thread read model 合并标准 reasoning.final payload 中的 providerMetadata / provider_metadata。
  • reasoning.final 进入历史 thread item 投影并标记为 completed,不再只依赖旧 reasoning.completed。

退出条件:

  • Provider 原生 metadata 只进入 reasoning metadata,不驱动 PlanState。
  • PlanState 仍只消费 plan.delta/final 与带 revisionId 的历史 plan item。
  • 不新增平行 planner、不恢复旧 agent_runtime_* 生产事实源。

P6:current read model / hydrator fixture 回归

  • AgentRuntimeThreadReadModel 显式声明 current thread_items 字段,避免 history hydrate 依赖未知字段透传。
  • App Server detail.thread_read.thread_items fixture 覆盖 reasoning metadata.provider_metadata 与 plan metadata.revisionId 保留。
  • PlanState 回归覆盖历史 reasoning provider metadata 不会恢复成计划;带 revision 的 plan item 仍是唯一历史计划事实。
  • ReasoningState 回归覆盖带 provider metadata 的历史 reasoning item 仍可水合完成态 / 运行态,metadata 不改变状态语义。

退出条件:

  • App Server read model 能把 current thread item history 带到前端,不把 provider 私有字段提升成 Plan 输入。
  • Plan / Reasoning hydrator 的职责继续分离:Plan 只读 plan facts,Reasoning 只读 reasoning facts。
  • 本阶段不新增生产 mock、不恢复旧 agent_runtime_*,fixture 只作为测试夹具。

P7:current fixture GUI evidence

  • web-tools-rendering external backend fixture 发出标准 reasoning.final,并携带 providerMetadata / provider_metadata。
  • 同一 fixture 的 historical reasoning item.updated / item.completed 携带 metadata.provider_metadata 与 native_reasoning_item_id。
  • current read model helper 纳入 thread_read.thread_items / threadItems,用于 fixture smoke 读取 history item。
  • WebSearch/WebFetch Electron GUI summary 断言 reasoning provider metadata 不会打开 PlanRail,也不会出现计划实施确认。
  • 真实 Electron 窄场景 web-tools-rendering 通过,证据落到 .lime/qc/gui-evidence/claw-chat-current-fixture/plan-reasoning-provider-metadata-regression-summary.json。

退出条件:

  • Provider 原生 reasoning metadata 能从 external fixture backend 进入 App Server read model,并被 GUI smoke 作为 current evidence 验证。
  • GUI 过程层能看到 reasoning/web process,但 PlanRail 仍只由 plan facts 驱动。
  • 不把 fixture backend、renderer mock 或 App Server mock backend 当作生产 fallback。

P8:aggregate current fixture regression

  • 聚合 npm run smoke:agent-runtime-current-fixture 覆盖 P7 新增 web-tools-rendering evidence,不破坏历史 / 缓存、流式收尾、Coding Workbench、cancel-then-continue、Skills Runtime、MCP structuredContent、Expert Skills Runtime / Plaza / Panel 等 current fixture。
  • 修正 claw-chat-current-fixture-smoke.mjs 成功路径截图采集为 best-effort,并设置短 timeout,避免 page.screenshot 超时覆盖已经通过的 read model / GUI / evidence pack 断言。
  • 保持失败路径截图同样有短 timeout;失败 evidence 仍会写入 summary.error / screenshotError,不会吞掉真实断言失败。

退出条件:

  • 聚合 current fixture smoke 通过,且输出仍明确 liveProviderUsed=false。
  • 截图失败只作为 evidence 缺口记录,不再把已完成场景误判为产品链路失败。
  • 不降低 scenario assertions:read model、GUI、backend ledger、evidence pack 断言仍必须通过。

P9:旧 update_plan 工具轨迹收口

  • 盘点 PlanRail、Runtime Strip、Harness 摘要和 MessageList inline process 的 plan 消费路径。
  • 确认 PlanState / TaskRail / Runtime Strip / Harness 摘要只消费标准 plan facts:live plan.delta/final、带 revisionId 的 type=plan thread item,以及 <proposed_plan> 消息级 fallback。
  • 将 isUpdatePlanToolName 收敛到共享 tool name family helper,避免 TaskRail 和 MessageList 各自维护旧工具名判断。
  • MessageList inline timeline 不再把 update_plan / UpdatePlanTool thread tool_call 渲染成工具过程卡;旧工具项只作为 provenance / retired guard,不驱动 UI 计划展示。
  • 历史 hydrate 不再把 update_plan / UpdatePlanTool thread tool_call 恢复到 message toolCalls / contentParts。
  • 补充回归:只有 update_plan 工具项时不生成 inline process;标准 plan item 存在时不重复展示旧工具卡;外置 timeline process 判断不被 update_plan 误触发;历史 hydrate 不带出旧工具卡。

退出条件:

  • update_plan 只能通过标准 plan.final / revisioned plan item 进入 Plan UI。
  • update_plan thread tool_call 不再作为 MessageList / TaskRail / Harness / history hydrate 的计划恢复入口。
  • 仍保留导入来源和历史 provenance,不删除 current Codex import evidence。

P10:plan item history hydrate current 闭环

  • 历史 hydrate 支持带 revisionId 的标准 type=plan item,将其恢复为 <proposed_plan> 内容块。
  • 无 revisionId 的历史 plan item 仍不进入 current hydrate,避免旧计划残留回流。
  • 同一 turn 内的旧 update_plan / UpdatePlanTool tool_call 仍被过滤,不恢复为消息工具卡。

退出条件:

  • plan.final -> revisioned plan item -> history hydrate -> GUI proposed_plan block 有稳定单测覆盖。
  • 旧 update_plan tool_call 只作为 provenance / retired guard,不参与 GUI 计划渲染。

P11:真实 Electron current fixture

  • claw-chat-current-fixture 的 plan 场景发出 current plan.final(source=proposed_plan),不依赖旧 update_plan tool_call。
  • App Server thread_item type="plan" metadata 保留 revisionId/source/plan,作为历史 hydrate 的 current 事实源。
  • plan 场景完成后 reload renderer,并从侧栏重新打开同一 session,验证 plan.final -> revisioned plan item -> history hydrate -> GUI proposed_plan。
  • GUI 负向断言覆盖 UpdatePlanTool / update_plan 旧工具卡不可见。
  • 聚合 npm run smoke:agent-runtime-current-fixture 纳入 --scenario plan,防止 plan fixture 只停留在手工入口。

退出条件:

  • 真实 Electron Desktop Host + App Server current JSON-RPC + external fixture backend 的 plan 场景通过。
  • read model summary 能看到 completed type=plan item、revisionId、source=proposed_plan 和全部 plan steps。
  • 历史重开后 GUI 仍显示计划步骤和实施确认,不显示旧 UpdatePlanTool / update_plan 工具过程卡。

本轮骨架写集

预计写集:

  • .gitignore
  • internal/roadmap/plan/README.md
  • internal/roadmap/plan/implementation-plan.md
  • internal/exec-plans/plan-runtime-implementation-progress.md
  • lime-rs/crates/app-server/src/runtime_backend.rs
  • lime-rs/crates/app-server/src/runtime_backend/model_capability.rs
  • lime-rs/crates/app-server/src/runtime_backend/plan_events.rs
  • lime-rs/crates/app-server/src/runtime_backend/reasoning_events.rs
  • src/components/agent/chat/utils/planState.ts
  • src/components/agent/chat/utils/planState.unit.test.ts
  • src/components/agent/chat/utils/modelReasoningState.ts
  • src/components/agent/chat/utils/modelReasoningState.unit.test.ts
  • src/lib/api/agentRuntime/types.ts
  • src/lib/api/agentRuntime/types.d.ts
  • src/lib/api/agentRuntime/appServerReadModelProjection.test.ts
  • scripts/agent-runtime/claw-chat-current-fixture-constants.mjs
  • scripts/agent-runtime/claw-chat-current-fixture-backend-file.mjs
  • scripts/agent-runtime/claw-chat-current-fixture-read-model-core.mjs
  • scripts/agent-runtime/claw-chat-current-fixture-gui-web-tools-waits.mjs
  • scripts/agent-runtime/claw-chat-current-fixture-plan-history.mjs
  • scripts/agent-runtime/claw-chat-current-fixture-scenario-flow.mjs
  • scripts/agent-runtime/claw-chat-current-fixture-scenario-assertions.mjs
  • scripts/agent-runtime/claw-chat-current-fixture-smoke.mjs

避让:

  • 当前 right-surface 相关脏写集。
  • Electron Host / App Server protocol 大范围生成文件。
  • GUI 组件深改和 i18n 文案扩展。

Plan 事实源治理分类

  • current:App Server / RuntimeCore 输出的 live plan.delta/final,前端标准 PlanState,以及带 revisionId 的 latest plan revision 投影。
  • current:Electron fixture plan 场景中的 plan.final(source=proposed_plan),只用于验证 current read model 与 GUI hydrate,不作为生产 mock fallback。
  • current:历史 thread_item type="plan" 只有带 revisionId 时才允许进入标准 PlanState。
  • current:<proposed_plan> 仍可作为当前消息级计划兜底,用于 plan fact 尚未持久化前的 UI 显示。
  • compat / test-only:isUpdatePlanToolName 只用于过滤 update_plan 工具活动,避免工具轨重复展示;不再负责解析计划步骤。
  • dead:PlanState、TaskRail、Runtime Strip 与 Harness 摘要不再从 message toolCall update_plan.result.metadata.plan、历史 tool_call.metadata.plan 或无 revisionId 的 thread_item type="plan" 直读计划。
  • fallback:todoItems 仍作为无 PlanState、无 <proposed_plan> 时的弱恢复来源,不作为 Plan Mode 主事实源。

验证记录

  • 2026-06-23:
    • npx vitest run "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,2 个文件 4 个测试;本轮新增 implementation plan 后回归。
    • rg -n "[ \t]+$" "internal/roadmap/plan" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:无匹配,新增 / 更新计划文档无尾随空白。
    • npx vitest run "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,2 个文件 4 个测试。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server model_capability
      • 结果:通过,3 个测试。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server plan_events
      • 结果:通过,1 个测试。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server reasoning_events
      • 结果:通过,1 个测试。
    • git diff --check -- ".gitignore" "lime-rs/crates/app-server/src/runtime_backend.rs"
      • 结果:通过,已跟踪改动无空白错误。
    • rg -n "[ \t]+$" <P0 新增文件>
      • 结果:无匹配,新增文件无尾随空白。
    • npx vitest run "src/lib/api/agentProtocol.test.ts" "src/lib/api/agentRuntime/threadClient.test.ts" "src/components/agent/chat/hooks/agentStreamPlanEventController.unit.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,3 个文件 63 个测试;覆盖 plan.delta/final、reasoning.delta parser / projection / live plan thread item。
    • cargo fmt --manifest-path "lime-rs/Cargo.toml" --all
      • 结果:通过。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server proposed_plan_parser
      • 结果:通过,5 个测试;覆盖 <proposed_plan> 闭合、跨 chunk 标签、未闭合计划、半截闭合标签和 RuntimeEvent 投影。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server plan_events
      • 结果:通过,3 个测试;覆盖 plan.delta skeleton、proposed_plan -> plan.final、update_plan -> plan.final。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server tool_events
      • 结果:通过,16 个相关测试;覆盖 thinking_delta -> reasoning.delta、工具事件和 update_plan plan fact 投影。
    • npx tsc --noEmit --pretty false --skipLibCheck false --project tsconfig.json
      • 结果:失败;该手工严检暴露 node_modules / vite-env.d.ts 全局声明冲突、right-surface 并行改动中的 null -> boolean,以及本轮新增 plan event 两处 TS 窄化问题。本轮相关 TS 问题已修复;该命令不是仓库默认类型入口。
    • npm run typecheck
      • 结果:失败;仅剩 src/components/agent/chat/workspace/useWorkspaceDebugRuntime.unit.test.ts(22,3): Type 'null' is not assignable to type 'boolean',归属当前 right-surface 并行写集,不属于本轮 Plan 主链改动。
    • npm run smoke:agent-runtime-current-fixture
      • 结果:失败在 Claw Expert Panel Skills Runtime override Electron fixture 的 Harness 证据导出入口,错误为未找到可打开 Harness 证据导出的入口;同次 smoke 中历史 / 缓存恢复、流式完成与运行态收尾、Electron/App Server fixture guard、Coding Workbench Electron fixture、Claw cancel-then-continue、Skills Runtime、MCP structuredContent、Expert Skills Runtime、Expert Plaza click-through 均已通过。失败点归属 Expert Panel / Harness UI 证据入口,不是本轮 plan.delta/final 事件链路。
    • cargo fmt --manifest-path "lime-rs/Cargo.toml" --all
      • 结果:通过。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server model_capability
      • 结果:通过,4 个测试;覆盖 reasoning level 归一、reasoning capability 降级和 xhigh 稳定序列化。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server model_effective
      • 结果:通过,1 个测试;覆盖 model.effective payload 中 provider/model/source/service slot/reasoning policy。
    • npx vitest run "src/lib/api/agentProtocol.test.ts" "src/lib/api/agentRuntime/threadClient.test.ts" "src/components/agent/chat/projection/agentUiEventProjection.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,3 个文件 83 个测试;覆盖 model.effective -> model_effective parser、App Server event stream projection、event sequence gate 和 Agent UI projection。
    • npm run test:contracts
      • 结果:通过;覆盖 app-server client contract、command contracts、harness contracts、modality contracts、scripts governance、Electron release workflow、harness cleanup contract 和 docs boundary。曾因 runtimeLifecycleProjection.ts 静态 guard 要求 model_change 直接委托 model: event.model / mode: event.mode 失败,已通过拆出 buildModelEffectiveEvent 修复。
    • git diff --check
      • 结果:通过。
    • rg -n "[ \t]+$" "internal/roadmap/plan" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:无匹配,Plan 路线图与执行计划无尾随空白。
    • npm run typecheck
      • 结果:本轮重跑超过多轮等待无输出,主动中断,退出码 130,未形成有效类型结论;上一轮已知失败归属 src/components/agent/chat/workspace/useWorkspaceDebugRuntime.unit.test.ts 的 right-surface 并行写集,不属于本轮 Plan 主链。
    • npm run smoke:agent-runtime-current-fixture
      • 结果:失败在 Claw Expert Panel Skills Runtime override Electron fixture 的 Harness panel evidence export;Playwright 点击 Harness 按钮时被窗口 drag region 覆盖并超时。同次 smoke 中历史 / 缓存恢复、流式收尾、Electron/App Server fixture guard、Coding Workbench、Claw cancel-then-continue、Skills Runtime、MCP structuredContent、Expert Skills Runtime、Expert Plaza click-through 均已通过。失败点归属 Expert Panel / Harness UI 点击层,不是本轮 Plan / model.effective 事件链路。
    • npx vitest run "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,2 个文件 9 个测试;覆盖结构化 update_plan plan 状态、历史 plan 初始态 + live plan 覆盖、model.effective 顶层 / modelRef 多模型归一、历史 reasoning item hydrate、reasoning.started/delta/final/ended lifecycle hydrate。
    • npx vitest run "src/lib/api/agentProtocol.test.ts" "src/lib/api/agentRuntime/threadClient.test.ts" "src/components/agent/chat/projection/agentUiEventProjection.test.ts" "src/components/agent/chat/hooks/agentStreamPlanEventController.unit.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,4 个文件 84 个测试;覆盖 plan.delta/final、reasoning.final、model.effective parser / App Server event stream projection / Agent UI projection / live plan thread item。输出中的 blocked invalid runtime event sequence 为既有负向测试断言。
    • npx eslint "src/components/agent/chat/utils/planState.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" --max-warnings 0
      • 结果:通过;覆盖本轮新增 / 修改的 PlanState 与 ReasoningState 纯状态文件。
    • git diff --check -- "src/components/agent/chat/utils/planState.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:通过;该命令仅覆盖已跟踪 diff,新文件尾随空白由下一条 rg 检查覆盖。
    • rg -n "[ \t]+$" "src/components/agent/chat/utils/planState.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:无匹配;本轮写集无尾随空白。
    • npm run typecheck
      • 结果:两次运行均超过长时间等待无输出,分别因用户打断和主动中断退出,退出码 130;未形成有效类型结论。上一轮已知有效失败仍归属 src/components/agent/chat/workspace/useWorkspaceDebugRuntime.unit.test.ts 的 right-surface 并行写集,不属于本轮 Plan 状态层。
    • npx vitest run "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,2 个文件 29 个测试;覆盖 TaskRail view model 消费标准 PlanState、带 revision 的标准 plan item 单步解析。
    • npx vitest run "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/lib/api/agentProtocol.test.ts" "src/lib/api/agentRuntime/threadClient.test.ts" "src/components/agent/chat/projection/agentUiEventProjection.test.ts" "src/components/agent/chat/hooks/agentStreamPlanEventController.unit.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,7 个文件 118 个测试;覆盖 Plan/Reasoning 状态 hydrator、TaskRail PlanState 消费、协议 parser、App Server event projection、Agent UI projection 与 live plan thread item。输出中的 blocked invalid runtime event sequence 为既有负向测试断言。
    • npx eslint "src/components/agent/chat/utils/planState.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" --max-warnings 0
      • 结果:通过;覆盖本轮 PlanState / ReasoningState 与 TaskRail view model 写集。
    • npx vitest run "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/components/AgentRuntimeStrip.test.tsx" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,2 个文件 15 个测试;覆盖 runtime strip 依赖的 harnessState.plan 从标准 PlanState 恢复 plan revision,并保持旧计划摘要和输出信号行为。
    • npx vitest run "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/components/AgentRuntimeStrip.test.tsx" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/lib/api/agentProtocol.test.ts" "src/lib/api/agentRuntime/threadClient.test.ts" "src/components/agent/chat/projection/agentUiEventProjection.test.ts" "src/components/agent/chat/hooks/agentStreamPlanEventController.unit.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,9 个文件 133 个测试;覆盖 Plan/Reasoning hydrator、TaskRail / PlanRail 间接消费、runtime strip 计划状态消费、协议和 projection 链路。
    • npx eslint "src/components/agent/chat/utils/planState.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "src/components/agent/chat/utils/harnessState.ts" "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/components/AgentRuntimeStrip.tsx" "src/components/agent/chat/components/AgentRuntimeStrip.test.tsx" --max-warnings 0
      • 结果:通过;覆盖本轮标准状态层、TaskRail/PlanRail 间接消费与 runtime strip 写集。
    • npx vitest run "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/components/AgentRuntimeStrip.test.tsx" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,2 个文件 18 个测试;覆盖 harnessState.reasoning 从标准 ReasoningState 恢复,以及 runtime strip 只在真实 reasoning lifecycle / text 信号下显示思考状态,单纯 model.effective 能力快照不伪装成运行状态。
    • npx vitest run "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/components/AgentRuntimeStrip.test.tsx" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/lib/api/agentProtocol.test.ts" "src/lib/api/agentRuntime/threadClient.test.ts" "src/components/agent/chat/projection/agentUiEventProjection.test.ts" "src/components/agent/chat/hooks/agentStreamPlanEventController.unit.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,9 个文件 136 个测试;覆盖 Plan/Reasoning hydrator、TaskRail / PlanRail 间接消费、runtime strip 标准 ReasoningState 消费、协议和 projection 链路。输出中的 blocked invalid runtime event sequence 为既有负向测试断言。
    • npx eslint "src/components/agent/chat/utils/planState.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "src/components/agent/chat/utils/harnessState.ts" "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/components/AgentRuntimeStrip.tsx" "src/components/agent/chat/components/AgentRuntimeStrip.test.tsx" --max-warnings 0
      • 结果:通过;覆盖本轮状态层、TaskRail / PlanRail 间接消费与 runtime strip reasoning UI 写集。
    • rg -n "[ \t]+$" "src/components/agent/chat/utils/planState.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "src/components/agent/chat/utils/harnessState.ts" "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/components/AgentRuntimeStrip.tsx" "src/components/agent/chat/components/AgentRuntimeStrip.test.tsx" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:无匹配;本轮写集无尾随空白。
    • npm run smoke:agent-runtime-current-fixture
      • 结果:失败在 Claw Expert Panel Skills Runtime override Electron fixture,错误为专家技能选择器候选未达到预期状态(candidateVisible=false、addButtonVisible=false)。同次 smoke 中历史 / 缓存恢复、流式完成与运行态收尾、Electron/App Server fixture guard、Coding Workbench Electron fixture、Claw cancel-then-continue、Skills Runtime、MCP structuredContent、Expert Skills Runtime、Expert Plaza click-through 均已通过。失败点归属 Expert Panel Skills Runtime / picker 候选状态,不是本轮 Plan / Reasoning UI 状态链路。
    • npx vitest run "src/components/agent/chat/workspace/planImplementationDecision.unit.test.ts" "src/components/agent/chat/utils/harnessState.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,2 个文件 16 个测试;覆盖 HarnessPlanState 从标准 PlanState 恢复 revisionId / turnId / source,以及本地实施确认请求携带 plan_revision_id / source_item_id / turn_id / plan_source。
    • npx eslint "src/components/agent/chat/workspace/planImplementationDecision.ts" "src/components/agent/chat/workspace/planImplementationDecision.unit.test.ts" "src/components/agent/chat/utils/harnessState.ts" "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/AgentChatWorkspace.tsx" --max-warnings 0
      • 结果:通过;覆盖 latest plan revision metadata builder、harness state 投影和 AgentChatWorkspace 本地实施确认发送接线。
    • npx vitest run "src/components/agent/chat/workspace/planComposerDecision.unit.test.ts" "src/components/agent/chat/workspace/PlanComposerDecisionPanel.test.tsx" "src/components/agent/chat/workspace/planImplementationDecision.unit.test.ts" "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/components/AgentRuntimeStrip.test.tsx" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,5 个文件 46 个测试;覆盖计划确认识别、实施确认 latest revision metadata、确认抽屉回归、runtime strip plan / reasoning 状态消费。
    • npx eslint "src/components/agent/chat/workspace/PlanComposerDecisionPanel.tsx" "src/components/agent/chat/workspace/PlanComposerDecisionPanel.test.tsx" "src/components/agent/chat/workspace/planComposerDecision.ts" "src/components/agent/chat/workspace/planComposerDecision.unit.test.ts" "src/components/agent/chat/workspace/planImplementationDecision.ts" "src/components/agent/chat/workspace/planImplementationDecision.unit.test.ts" --max-warnings 0
      • 结果:通过;覆盖计划确认抽屉、计划确认识别和实施确认 metadata 写集。
    • npx vitest run "src/components/agent/chat/workspace/PlanComposerDecisionPanel.test.tsx" "src/components/agent/chat/workspace/planComposerDecision.unit.test.ts" "src/components/agent/chat/workspace/planImplementationDecision.unit.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,3 个文件 21 个测试;覆盖确认抽屉在存在 latest plan revision 时显示绑定状态,缺少 revision metadata 时不显示噪声状态,并保持提交 / 调整 / 忽略行为。
    • npx eslint "src/components/agent/chat/workspace/PlanComposerDecisionPanel.tsx" "src/components/agent/chat/workspace/PlanComposerDecisionPanel.test.tsx" "src/components/agent/chat/workspace/planComposerDecision.ts" "src/components/agent/chat/workspace/planComposerDecision.unit.test.ts" "src/components/agent/chat/workspace/planImplementationDecision.ts" "src/components/agent/chat/workspace/planImplementationDecision.unit.test.ts" --max-warnings 0
      • 结果:通过;覆盖确认抽屉 latest revision UI、五语言 key 接入和计划实施确认纯逻辑。
    • git diff --check -- "src/components/agent/chat/workspace/PlanComposerDecisionPanel.tsx" "src/components/agent/chat/workspace/PlanComposerDecisionPanel.test.tsx" "src/i18n/resources/zh-CN/agent.json" "src/i18n/resources/zh-TW/agent.json" "src/i18n/resources/en-US/agent.json" "src/i18n/resources/ja-JP/agent.json" "src/i18n/resources/ko-KR/agent.json"
      • 结果:通过;本轮确认抽屉与五语言资源写集无 diff 空白错误。
    • rg -n "[ \t]+$" "src/components/agent/chat/workspace/PlanComposerDecisionPanel.tsx" "src/components/agent/chat/workspace/PlanComposerDecisionPanel.test.tsx" "src/i18n/resources/zh-CN/agent.json" "src/i18n/resources/zh-TW/agent.json" "src/i18n/resources/en-US/agent.json" "src/i18n/resources/ja-JP/agent.json" "src/i18n/resources/ko-KR/agent.json" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:无匹配;本轮确认抽屉、五语言资源和执行计划无尾随空白。
    • npx vitest run "src/components/agent/chat/utils/proposedPlan.test.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/components/TaskCenterUtilityToolbar.integration.test.tsx" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,3 个文件 65 个测试;覆盖 Markdown checklist 状态解析、标准 PlanState latest revision 投影、右侧计划轨 / 运行控制面 revision badge 与完整 data-* metadata。
    • npx eslint "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/components/generalWorkbenchRunControlSurfaceViewModel.ts" "src/components/agent/chat/components/TaskCenterTaskRail.tsx" "src/components/agent/chat/components/TaskCenterRunControlSurface.tsx" "src/components/agent/chat/components/TaskCenterUtilityToolbar.tsx" "src/components/agent/chat/components/TaskCenterUtilityToolbar.integration.test.tsx" "src/components/agent/chat/components/generalWorkbenchTaskRailText.ts" "src/components/agent/chat/utils/proposedPlan.ts" "src/components/agent/chat/utils/proposedPlan.test.ts" --max-warnings 0
      • 结果:通过;覆盖本轮 right rail / run control projection、latest revision UI 接线和共享 proposed plan parser。
    • git diff --check -- "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/components/generalWorkbenchRunControlSurfaceViewModel.ts" "src/components/agent/chat/components/TaskCenterTaskRail.tsx" "src/components/agent/chat/components/TaskCenterRunControlSurface.tsx" "src/components/agent/chat/components/TaskCenterUtilityToolbar.tsx" "src/components/agent/chat/components/TaskCenterUtilityToolbar.integration.test.tsx" "src/components/agent/chat/components/generalWorkbenchTaskRailText.ts" "src/components/agent/chat/utils/proposedPlan.ts" "src/components/agent/chat/utils/proposedPlan.test.ts" "src/i18n/resources/zh-CN/agent.json" "src/i18n/resources/zh-TW/agent.json" "src/i18n/resources/en-US/agent.json" "src/i18n/resources/ja-JP/agent.json" "src/i18n/resources/ko-KR/agent.json" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:通过;本轮 Plan UI / i18n / 执行计划写集无 diff 空白错误。
    • rg -n "[ \t]+$" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/components/generalWorkbenchRunControlSurfaceViewModel.ts" "src/components/agent/chat/components/TaskCenterTaskRail.tsx" "src/components/agent/chat/components/TaskCenterRunControlSurface.tsx" "src/components/agent/chat/components/TaskCenterUtilityToolbar.tsx" "src/components/agent/chat/components/TaskCenterUtilityToolbar.integration.test.tsx" "src/components/agent/chat/components/generalWorkbenchTaskRailText.ts" "src/components/agent/chat/utils/proposedPlan.ts" "src/components/agent/chat/utils/proposedPlan.test.ts" "src/i18n/resources/zh-CN/agent.json" "src/i18n/resources/zh-TW/agent.json" "src/i18n/resources/en-US/agent.json" "src/i18n/resources/ja-JP/agent.json" "src/i18n/resources/ko-KR/agent.json" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:无匹配;本轮 Plan UI / i18n / 执行计划写集无尾随空白。
    • npx vitest run "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/utils/proposedPlan.test.ts" "src/components/agent/chat/components/TaskCenterUtilityToolbar.integration.test.tsx" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,3 个文件 65 个测试;覆盖旧 Plan 恢复路径负向守卫、Markdown checklist 状态解析、标准 PlanState latest revision 投影、右侧计划轨 / 运行控制面 revision badge 与完整 data-* metadata。
    • npx eslint "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/components/planToolProjection.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailThreadItems.ts" --max-warnings 0
      • 结果:通过;覆盖 TaskRail view model、旧 update_plan 工具活动过滤、<proposed_plan> 投影和对应负向守卫。
    • git diff --check -- "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/components/planToolProjection.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailThreadItems.ts"
      • 结果:通过;本轮 Plan 旧恢复清理写集无 diff 空白错误。
    • rg -n "[ \t]+$" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/components/planToolProjection.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailThreadItems.ts" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:无匹配;本轮 Plan 旧恢复清理写集与执行计划无尾随空白。
    • npx vitest run "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/components/AgentRuntimeStrip.test.tsx" "src/components/agent/chat/utils/proposedPlan.test.ts" "src/components/agent/chat/components/TaskCenterUtilityToolbar.integration.test.tsx" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,6 个文件 88 个测试;覆盖 PlanState 源头只接收带 revision 的历史 plan item、无 revision 历史 plan 负向守卫、TaskRail、Runtime Strip、Harness 摘要和右侧计划轨消费面。
    • npx eslint "src/components/agent/chat/utils/planState.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/harnessState.ts" "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/components/planToolProjection.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailThreadItems.ts" "src/components/agent/chat/components/AgentRuntimeStrip.tsx" "src/components/agent/chat/components/AgentRuntimeStrip.test.tsx" --max-warnings 0
      • 结果:通过;覆盖 PlanState、HarnessState、TaskRail、旧 update_plan 工具活动过滤和 Runtime Strip 写集。
    • git diff --check -- "src/components/agent/chat/utils/planState.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/harnessState.ts" "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/components/planToolProjection.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailThreadItems.ts" "src/components/agent/chat/components/AgentRuntimeStrip.tsx" "src/components/agent/chat/components/AgentRuntimeStrip.test.tsx" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:通过;本轮 Plan 事实源治理写集无 diff 空白错误。
    • rg -n "[ \t]+$" "src/components/agent/chat/utils/planState.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/harnessState.ts" "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/components/planToolProjection.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailThreadItems.ts" "src/components/agent/chat/components/AgentRuntimeStrip.tsx" "src/components/agent/chat/components/AgentRuntimeStrip.test.tsx" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:无匹配;本轮 Plan 事实源治理写集和执行计划无尾随空白。
    • npx vitest run "src/components/agent/chat/components/Inputbar/components/InputbarComposerSection.planStatus.test.tsx" "src/components/agent/chat/components/Inputbar/index.test.tsx" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,2 个文件 77 个测试;覆盖 Plan Mode 左侧状态 chip、模型 / reasoning 只读上下文、右侧模型选择器位置不回流,以及既有输入栏发送 / 工具状态回归。
    • npx eslint "src/components/agent/chat/components/Inputbar/components/InputbarComposerSection.tsx" "src/components/agent/chat/components/Inputbar/components/inputbarComposerSectionCopy.ts" "src/components/agent/chat/components/Inputbar/components/InputbarComposerSection.planStatus.test.tsx" "src/components/agent/chat/components/Inputbar/styles.ts" --max-warnings 0
      • 结果:通过;覆盖输入栏 Plan Mode 上下文展示、copy 构造和样式写集。
    • npx vitest run "src/i18n/__tests__/loadNamespace.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,1 个文件 7 个测试;覆盖新增 agentInputbar 五语言 key 能合并进 agent namespace。
    • git diff --check -- "src/components/agent/chat/components/Inputbar/components/InputbarComposerSection.tsx" "src/components/agent/chat/components/Inputbar/components/inputbarComposerSectionCopy.ts" "src/components/agent/chat/components/Inputbar/components/InputbarComposerSection.planStatus.test.tsx" "src/components/agent/chat/components/Inputbar/styles.ts" "src/i18n/resources/zh-CN/agentInputbar.json" "src/i18n/resources/zh-TW/agentInputbar.json" "src/i18n/resources/en-US/agentInputbar.json" "src/i18n/resources/ja-JP/agentInputbar.json" "src/i18n/resources/ko-KR/agentInputbar.json"
      • 结果:通过;本轮输入栏 Plan Mode UI 与五语言资源写集无 diff 空白错误。
    • rg -n "[ \t]+$" "src/components/agent/chat/components/Inputbar/components/InputbarComposerSection.tsx" "src/components/agent/chat/components/Inputbar/components/inputbarComposerSectionCopy.ts" "src/components/agent/chat/components/Inputbar/components/InputbarComposerSection.planStatus.test.tsx" "src/components/agent/chat/components/Inputbar/styles.ts" "src/i18n/resources/zh-CN/agentInputbar.json" "src/i18n/resources/zh-TW/agentInputbar.json" "src/i18n/resources/en-US/agentInputbar.json" "src/i18n/resources/ja-JP/agentInputbar.json" "src/i18n/resources/ko-KR/agentInputbar.json"
      • 结果:无匹配;本轮输入栏 Plan Mode UI 与五语言资源写集无尾随空白。
    • npx vitest run "src/components/agent/chat/components/MessageList.webProcess.test.tsx" "src/components/agent/chat/components/messageListInlineProcess.test.ts" "src/components/agent/chat/components/StreamingRenderer.structuredContent.test.tsx" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,3 个文件 22 个测试;覆盖标准 plan timeline fact 已存在时聊天正文不再重复渲染 AgentPlanBlock,同时没有 plan fact 的 <proposed_plan> 仍保留消息级 fallback。
    • npx eslint "src/components/agent/chat/components/messageListItemProjection.ts" "src/components/agent/chat/components/MessageList.webProcess.test.tsx" --max-warnings 0
      • 结果:通过;覆盖本轮消息投影去重判断与回归测试写集。
    • git diff --check -- "src/components/agent/chat/components/messageListItemProjection.ts" "src/components/agent/chat/components/MessageList.webProcess.test.tsx" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:通过;本轮 <proposed_plan> 去重写集无 diff 空白错误。
    • rg -n "[ \t]+$" "src/components/agent/chat/components/messageListItemProjection.ts" "src/components/agent/chat/components/MessageList.webProcess.test.tsx" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:无匹配;本轮 <proposed_plan> 去重写集和执行计划无尾随空白。
    • npx vitest run "src/components/agent/chat/components/thinkingBlockDisplay.test.ts" "src/components/agent/chat/components/StreamingProcessGroupModel.unit.test.ts" "src/components/agent/chat/components/StreamingRenderer.processGroups.test.tsx" "src/components/agent/chat/components/StreamingRenderer.test.tsx" "src/i18n/__tests__/loadNamespace.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,5 个文件 48 个测试;覆盖流式纯 thinking / reasoning 默认折叠、展开后可查看详情、thinking 状态文案可注入本地化,以及新增 agentChat.thinkingBlock.* 五语言 key 能加载。
    • npx vitest run "src/components/agent/chat/components/MessageList.webProcess.test.tsx" "src/components/agent/chat/components/messageListInlineProcess.test.ts" "src/components/agent/chat/components/StreamingRenderer.structuredContent.test.tsx" "src/components/agent/chat/components/MessageList.reasoningFlow.test.tsx" "src/components/agent/chat/components/MessageList.reasoningPersistence.test.tsx" "src/components/agent/chat/components/StreamingRenderer.webSearch.sequence.test.tsx" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,6 个文件 46 个测试;覆盖 <proposed_plan> 去重、reasoning / thinking 从 MessageList timeline 到 StreamingRenderer 的顺序与持久化回归、Web 检索过程顺序回归。
    • npx eslint "src/components/agent/chat/components/messageListItemProjection.ts" "src/components/agent/chat/components/MessageList.webProcess.test.tsx" "src/components/agent/chat/components/StreamingProcessGroupModel.ts" "src/components/agent/chat/components/StreamingRenderer.tsx" "src/components/agent/chat/components/StreamingProcessGroup.tsx" "src/components/agent/chat/components/ThinkingBlock.tsx" "src/components/agent/chat/components/thinkingBlockDisplay.ts" "src/components/agent/chat/components/thinkingBlockDisplay.test.ts" "src/components/agent/chat/components/StreamingProcessGroupModel.unit.test.ts" "src/components/agent/chat/components/StreamingRenderer.processGroups.test.tsx" "src/components/agent/chat/components/StreamingRenderer.testMocks.tsx" --max-warnings 0
      • 结果:通过;覆盖本轮消息投影、过程组默认折叠、ThinkingBlock i18n 和测试 mock 写集。
    • git diff --check -- <本轮 Plan UI tracked 写集>
      • 结果:通过;覆盖本轮已跟踪 TS / TSX / i18n 写集无 diff 空白错误。internal/exec-plans/plan-runtime-implementation-progress.md 当前仍是 untracked 文件,由下一条 rg 覆盖尾随空白。
    • rg -n "[ \t]+$" <本轮 Plan UI 写集 + internal/exec-plans/plan-runtime-implementation-progress.md>
      • 结果:无匹配;本轮 Plan UI 写集与执行计划无尾随空白。
    • cargo fmt --manifest-path "lime-rs/Cargo.toml" --all
      • 结果:通过;格式化本轮 Rust model capability 写集。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server model_capability
      • 结果:通过,8 个测试;覆盖 Codex/OpenAI、Anthropic、Gemini、OpenAI-compatible、多模型 reasoning 能力矩阵,OpenAI-compatible 普通聊天模型降级,none 不误升 reasoning,默认档位为 medium,以及 xhigh 序列化稳定。
    • npx vitest run "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "src/components/agent/chat/utils/planState.unit.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,2 个文件 13 个测试;覆盖多模型 model.effective payload 归一、无 reasoning 模型不伪造运行态、minimal 档位归一,以及 PlanState 只消费 plan.delta/final facts、不从 model / reasoning 事件恢复计划。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server model_effective
      • 结果:通过,1 个测试;覆盖运行时 model.effective payload 仍记录选中模型、service slot 与 reasoning policy。
    • npx vitest run "src/lib/api/agentProtocol.test.ts" "src/lib/api/agentRuntime/threadClient.test.ts" "src/components/agent/chat/projection/agentUiEventProjection.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,3 个文件 83 个测试;覆盖 Plan / Reasoning / model.effective 事件 parser、App Server event stream projection 与 Agent UI projection。输出中的 blocked invalid runtime event sequence 为既有负向测试断言。
    • npx eslint "src/components/agent/chat/utils/modelReasoningState.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "src/components/agent/chat/utils/planState.ts" "src/components/agent/chat/utils/planState.unit.test.ts" --max-warnings 0
      • 结果:通过;覆盖本轮 P4 前端状态层写集。
    • npm run smoke:agent-runtime-current-fixture
      • 结果:通过;覆盖 history/cache hydration、turn.completed 工具收尾(legacy final_done 仅负向 guard)、Claw 终态 UI、Electron fixture guard、Coding Workbench Electron fixture、Claw cancel-then-continue、Skills Runtime natural / 显式 $skill / 工作台试用入口、MCP structuredContent、Expert Skills Runtime、Expert Plaza、ExpertInfoPanel skillRefs override 与 Harness panel Evidence Pack 导出;liveProviderUsed=false。
    • npm run verify:gui-smoke
      • 结果:通过;完成 renderer smoke build、Electron host build、App Server sidecar asset build,Electron smoke 中 renderer loaded、App Server initialized、Claw workbench shell ready、memory settings ready。
    • cargo fmt --manifest-path "lime-rs/Cargo.toml" --all
      • 结果:通过;格式化本轮 reasoning metadata Rust 写集。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p lime-agent test_convert_item_started_reasoning_runtime_item_preserves_summary
      • 结果:通过,1 个测试;覆盖 ItemRuntimePayload::Reasoning.metadata 透传到 AgentThreadItemPayload::Reasoning.metadata。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server reasoning_item_payload_metadata_is_preserved_in_thread_items
      • 结果:通过,1 个测试;覆盖 App Server thread read model 保留 reasoning item payload 中的 metadata.provider_metadata 与 provider 原生字段。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server reasoning_final_provider_metadata_is_projected_to_thread_item_metadata
      • 结果:通过,1 个测试;覆盖标准 reasoning.final payload 中的 providerMetadata 进入历史 thread item metadata,并标记 completed。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server runtime_thinking_delta_emits_reasoning_lifecycle_events
      • 结果:通过,1 个测试;确认 reasoning.started/delta/final/ended lifecycle 事件序列保持稳定。
    • CARGO_TARGET_DIR="/Users/coso/Documents/dev/ai/aiclientproxy/lime/lime-rs/target" cargo test --manifest-path "lime-rs/crates/agent-rust/Cargo.toml" -p agent-core test_project_message_emits_reasoning_summary_runtime_item
      • 结果:通过,1 个测试;覆盖 Agent projector 将 thinking signature 保真投影到 reasoning metadata。显式设置 CARGO_TARGET_DIR,未生成 lime-rs/crates/agent-rust/target。
    • npx vitest run "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "src/lib/api/agentRuntime/appServerReadModelProjection.test.ts" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,3 个文件 21 个测试;覆盖 current read model thread_items metadata 保留、历史 reasoning metadata 不污染 PlanState、ReasoningState 历史水合完成态 / 运行态。
    • npx eslint "src/lib/api/agentRuntime/types.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "src/lib/api/agentRuntime/appServerReadModelProjection.test.ts" --max-warnings 0
      • 结果:通过;覆盖本轮 TS 源码与测试写集。
    • npx eslint "src/lib/api/agentRuntime/types.d.ts" --max-warnings 0 --no-warn-ignored
      • 结果:通过;.d.ts 按仓库 ignore 规则不进入 ESLint 解析,本命令确认不会因 ignored warning 干扰本轮门禁。
    • git diff --check -- "src/lib/api/agentRuntime/types.ts" "src/lib/api/agentRuntime/types.d.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "src/lib/api/agentRuntime/appServerReadModelProjection.test.ts"
      • 结果:通过;本轮 touched 写集无 diff 空白错误。
    • rg -n "[ \t]+$" "src/lib/api/agentRuntime/types.ts" "src/lib/api/agentRuntime/types.d.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/modelReasoningState.unit.test.ts" "src/lib/api/agentRuntime/appServerReadModelProjection.test.ts" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:无匹配;本轮 touched 写集与执行计划无尾随空白。
    • node --check "scripts/agent-runtime/claw-chat-current-fixture-constants.mjs"
      • 结果:通过;覆盖 fixture constants 语法。
    • node --check "scripts/agent-runtime/claw-chat-current-fixture-backend-file.mjs" && node --check "scripts/agent-runtime/claw-chat-current-fixture-read-model-core.mjs" && node --check "scripts/agent-runtime/claw-chat-current-fixture-gui-web-tools-waits.mjs" && node --check "scripts/agent-runtime/claw-chat-current-fixture-scenario-flow.mjs" && node --check "scripts/agent-runtime/claw-chat-current-fixture-scenario-assertions.mjs"
      • 结果:通过;覆盖本轮 fixture mjs 写集语法。
    • npx vitest run "scripts/agent-runtime/claw-chat-current-fixture-smoke.test.mjs" "scripts/agent-runtime/current-fixture-regression-smoke.test.mjs" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,2 个文件 23 个测试;覆盖 current fixture guard、非 live provider / 非 mock backend 边界、WebSearch/WebFetch scenario guard。
    • npx eslint "scripts/agent-runtime/claw-chat-current-fixture-constants.mjs" "scripts/agent-runtime/claw-chat-current-fixture-backend-file.mjs" "scripts/agent-runtime/claw-chat-current-fixture-read-model-core.mjs" "scripts/agent-runtime/claw-chat-current-fixture-gui-web-tools-waits.mjs" "scripts/agent-runtime/claw-chat-current-fixture-scenario-flow.mjs" "scripts/agent-runtime/claw-chat-current-fixture-scenario-assertions.mjs" --max-warnings 0
      • 结果:通过;覆盖本轮 fixture mjs 写集 lint。
    • node "scripts/agent-runtime/claw-chat-current-fixture-smoke.mjs" --scenario web-tools-rendering --prefix plan-reasoning-provider-metadata-regression --timeout-ms 180000
      • 结果:通过;真实 Electron Desktop Host + App Server current JSON-RPC + external fixture backend。Summary 关键证据:includesReasoningFinal=true、includesReasoningFinalProviderMetadata=true、includesReasoningItem=true、includesReasoningItemProviderMetadata=true、hasAllPlanSteps=false、planDecisionVisible=false、assertions 中 readModelWebToolsReasoningProviderMetadataPreserved=true、guiWebToolsReasoningDidNotOpenPlanRail=true。
    • git diff --check -- "scripts/agent-runtime/claw-chat-current-fixture-constants.mjs" "scripts/agent-runtime/claw-chat-current-fixture-backend-file.mjs" "scripts/agent-runtime/claw-chat-current-fixture-read-model-core.mjs" "scripts/agent-runtime/claw-chat-current-fixture-gui-web-tools-waits.mjs" "scripts/agent-runtime/claw-chat-current-fixture-scenario-flow.mjs" "scripts/agent-runtime/claw-chat-current-fixture-scenario-assertions.mjs"
      • 结果:通过;本轮 fixture 写集无 diff 空白错误。
    • rg -n "[ \t]+$" "scripts/agent-runtime/claw-chat-current-fixture-constants.mjs" "scripts/agent-runtime/claw-chat-current-fixture-backend-file.mjs" "scripts/agent-runtime/claw-chat-current-fixture-read-model-core.mjs" "scripts/agent-runtime/claw-chat-current-fixture-gui-web-tools-waits.mjs" "scripts/agent-runtime/claw-chat-current-fixture-scenario-flow.mjs" "scripts/agent-runtime/claw-chat-current-fixture-scenario-assertions.mjs" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:无匹配;本轮 fixture 写集与执行计划无尾随空白。
    • node --check "scripts/agent-runtime/claw-chat-current-fixture-smoke.mjs"
      • 结果:通过;覆盖截图 best-effort 修正后的 smoke 主脚本语法。
    • npx eslint "scripts/agent-runtime/claw-chat-current-fixture-smoke.mjs" --max-warnings 0
      • 结果:通过;覆盖 smoke 主脚本 lint。
    • npm run smoke:agent-runtime-current-fixture
      • 第一次结果:失败在 Claw Expert Skills Runtime declared + selected + invoked Electron fixture 成功路径截图采集,page.screenshot: Timeout 180000ms exceeded。failure summary 证明该场景 read model、evidence pack 与 GUI 已完成:latestTurnStatus=completed、readModelTurnTerminal=true、skillSearchToolStatus=completed、skillToolStatus=completed、skillGateObserved=true、guiExpertSkillsRuntimeCompleted.hasAssistantSummary=true。
      • 修正:成功路径截图改为 best-effort,并设置 timeout: 15000;失败路径截图同样设置短 timeout。
      • 第二次结果:通过;聚合入口完成 history/cache hydration、turn.completed 工具收尾(legacy final_done 仅负向 guard)、Claw 终态 UI、Electron fixture guard、Coding Workbench Electron fixture、cancel-then-continue、Skills Runtime 三入口、MCP structuredContent、Expert Skills Runtime / Plaza / Panel Electron fixture;输出 liveProviderUsed=false。
    • npx vitest run "src/components/agent/chat/hooks/agentChatHistory.timeline.test.ts" "src/components/agent/chat/components/messageListTimelineContentParts.unit.test.ts" "src/components/agent/chat/components/messageListInlineProcess.test.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.unit.test.ts" "src/components/agent/chat/utils/toolNameFamily.unit.test.ts" "src/components/agent/chat/utils/planState.unit.test.ts" "src/components/agent/chat/utils/harnessState.test.ts" "src/components/agent/chat/components/AgentRuntimeStrip.test.tsx" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,8 个文件 81 个测试;覆盖 MessageList 不再渲染旧 update_plan 工具过程、外置 timeline process 判断不被 update_plan 误触发、历史 hydrate 不带出旧工具卡、TaskRail 计划恢复负向守卫、PlanState / Harness / Runtime Strip 仍只走标准 plan facts,以及共享工具名 helper。
    • npx eslint "src/components/agent/chat/utils/toolNameFamily.ts" "src/components/agent/chat/utils/toolNameFamily.unit.test.ts" "src/components/agent/chat/components/planToolProjection.ts" "src/components/agent/chat/components/messageListTimelineContentParts.ts" "src/components/agent/chat/components/messageListTimelineContentParts.unit.test.ts" "src/components/agent/chat/components/messageListInlineProcess.ts" "src/components/agent/chat/components/messageListInlineProcess.test.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailViewModel.ts" "src/components/agent/chat/components/generalWorkbenchTaskRailThreadItems.ts" "src/components/agent/chat/hooks/agentChatHistoryThreadItems.ts" "src/components/agent/chat/hooks/agentChatHistory.timeline.test.ts" --max-warnings 0
      • 结果:通过;覆盖本轮共享 helper、MessageList 过滤、历史 hydrate 过滤、TaskRail re-export 和回归测试写集。
    • git diff --check -- "src/components/agent/chat/utils/toolNameFamily.ts" "src/components/agent/chat/utils/toolNameFamily.unit.test.ts" "src/components/agent/chat/components/planToolProjection.ts" "src/components/agent/chat/components/messageListTimelineContentParts.ts" "src/components/agent/chat/components/messageListTimelineContentParts.unit.test.ts" "src/components/agent/chat/components/messageListInlineProcess.ts" "src/components/agent/chat/components/messageListInlineProcess.test.ts" "src/components/agent/chat/hooks/agentChatHistoryThreadItems.ts" "src/components/agent/chat/hooks/agentChatHistory.timeline.test.ts"
      • 结果:通过;本轮 P9 tracked 写集无 diff 空白错误。
    • rg -n "[ \t]+$" "src/components/agent/chat/utils/toolNameFamily.ts" "src/components/agent/chat/utils/toolNameFamily.unit.test.ts" "src/components/agent/chat/components/planToolProjection.ts" "src/components/agent/chat/components/messageListTimelineContentParts.ts" "src/components/agent/chat/components/messageListTimelineContentParts.unit.test.ts" "src/components/agent/chat/components/messageListInlineProcess.ts" "src/components/agent/chat/components/messageListInlineProcess.test.ts" "src/components/agent/chat/hooks/agentChatHistoryThreadItems.ts" "src/components/agent/chat/hooks/agentChatHistory.timeline.test.ts" "internal/exec-plans/plan-runtime-implementation-progress.md"
      • 结果:无匹配;本轮 P9 写集与执行计划无尾随空白。
    • node --check "scripts/agent-runtime/claw-chat-current-fixture-plan-history.mjs"
      • 结果:通过;覆盖新增 plan history hydrate helper 语法。
    • node --check "scripts/agent-runtime/claw-chat-current-fixture-scenario-flow.mjs"
      • 结果:通过;覆盖 plan 场景接入 history hydrate helper 后的语法。
    • node --check "scripts/agent-runtime/current-fixture-regression-smoke.mjs"
      • 结果:通过;覆盖聚合 current fixture 增加 --scenario plan 后的语法。
    • cargo fmt --manifest-path "lime-rs/Cargo.toml" --all --check
      • 结果:通过;本轮 App Server plan thread item metadata 投影符合格式。
    • cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server plan_item_preserves_revision_source_and_plan_metadata
      • 结果:通过,1 个测试;覆盖 plan.final 投影到 thread_item type=plan 时保留 metadata.revisionId/source/plan。
    • npx vitest run "scripts/agent-runtime/claw-chat-current-fixture-smoke.test.mjs" "scripts/agent-runtime/current-fixture-regression-smoke.test.mjs" --silent=passed-only --disableConsoleIntercept
      • 结果:通过,2 个文件 26 个测试;覆盖 Plan revisioned history hydrate guard、非 live provider / 非 mock backend 边界和聚合 plan scenario 入口。
    • npx eslint "scripts/agent-runtime/claw-chat-current-fixture-backend-file.mjs" "scripts/agent-runtime/claw-chat-current-fixture-plan-history.mjs" "scripts/agent-runtime/claw-chat-current-fixture-scenario-flow.mjs" "scripts/agent-runtime/claw-chat-current-fixture-scenario-assertions.mjs" --max-warnings 0
      • 结果:通过;覆盖本轮 Plan fixture 脚本写集 lint。
    • node "scripts/agent-runtime/claw-chat-current-fixture-smoke.mjs" --scenario plan --prefix plan-current-history-hydrate --timeout-ms 180000
      • 第一次结果:失败,暴露 PLAN_STEPS 未导入到 claw-chat-current-fixture-scenario-flow.mjs。修正:补齐导入。
      • 第二次结果:失败,暴露 external backend 没有显式 current plan.final,read model 只有 GUI <proposed_plan> fallback。修正:plan 分支补 plan.final(source=proposed_plan, revisionId=proposed_plan:fixture-1)。
      • 第三次结果:通过;真实 Electron Desktop Host + App Server current JSON-RPC + external fixture backend。Summary 关键证据:readModelPlanThreadItem.planThreadItemCount=1、hasRevisionId=true、source=proposed_plan、includesAllPlanSteps=true、readModelPlanHistoryHydrate.hasRevisionId=true、guiPlanHistoryHydrateCompleted.hasAllPlanSteps=true、legacyUpdatePlanToolVisible=false、scenario assertions 中 readModelPlanThreadItemRevisioned=true、guiPlanHistoryHydrateCompleted=true、readModelPlanHistoryHydratePreserved=true、legacyUpdatePlanToolHidden=true。

拆分说明:

  • src/components/agent/chat/utils/harnessState.ts 已超过 1000 行,本轮只删除旧 Plan 恢复路径并调整既有状态投影,没有继续追加新业务逻辑。后续若继续扩展 Harness 状态,应优先拆出 harnessPlanState / harnessReasoningState 子模块,避免在中心文件继续膨胀。
  • src/components/agent/chat/components/messageListItemProjection.ts 已超过 800 行预警线。本轮只把既有 shouldRenderProposedPlanBlocks 判断收敛到标准 timeline plan fact,没有追加新的消息投影业务分支;后续若继续扩展 MessageList 投影,应优先拆出 plan / reasoning projection helper。
  • src/components/agent/chat/components/StreamingRenderer.test.tsx 已超过 800 行预警线。本轮只保留既有回归并把新增 thinking / reasoning 折叠覆盖放到 StreamingRenderer.processGroups.test.tsx,没有继续扩展该大测试文件;后续 StreamingRenderer 过程流回归应优先进入分域测试文件。
  • src/lib/api/agentRuntime/types.ts 已超过 1000 行。本轮只补 AgentRuntimeThreadReadModel.thread_items 类型字段并同步 .d.ts 镜像,没有追加新业务逻辑;后续若继续扩展 agent runtime read model 类型,应优先拆出 session read / thread read / runtime summary 分域类型文件,再由当前 types.ts 做 facade re-export。
  • scripts/agent-runtime/claw-chat-current-fixture-backend-file.mjs 已超过 800 行预警线。本轮追加的 plan.final 是 current fixture 最小事实源事件;后续继续扩展 external backend fixture 时,应拆出 plan / web-tools event renderer 子模块。
  • scripts/agent-runtime/claw-chat-current-fixture-scenario-flow.mjs 已超过 800 行预警线。本轮只接入 verifyPlanHistoryHydrate helper,新增逻辑放到小模块;后续新增 scenario 或继续扩展 read model summary 时,应拆出 scenario-specific flow helpers。

GUI 交付状态:

  • npm run smoke:agent-runtime-current-fixture 已通过,先前 Expert Panel / Harness blocker 当前不再复现。
  • npm run verify:gui-smoke 已通过,Plan 主线本轮具备最小 GUI smoke 交付证据。
  • node scripts/agent-runtime/claw-chat-current-fixture-smoke.mjs --scenario web-tools-rendering --prefix plan-reasoning-provider-metadata-regression --timeout-ms 180000 已通过,补齐 P7 provider metadata reasoning current fixture GUI evidence。
  • node scripts/agent-runtime/claw-chat-current-fixture-smoke.mjs --scenario plan --prefix plan-current-history-hydrate --timeout-ms 180000 已通过,补齐 P11 Plan revisioned thread item + history hydrate current fixture GUI evidence。
  • npm run smoke:agent-runtime-current-fixture 在 P8 已通过;本轮 P11 新增 plan 场景后未重跑全量聚合,下一刀优先补该回归。

下一刀

P11 真实 Electron current fixture 已完成。下一刀若继续增强 Plan 主链,优先补一次聚合 npm run smoke:agent-runtime-current-fixture 全量回归;如果继续做工程收口,则优先拆分已登记的大文件风险,不再向 harnessState.ts、types.ts 或 agent-runtime fixture 巨型脚本追加新业务逻辑。