merge: candidate refresh-task-state-recovery-2026-09-28 into dev

2026-09-28 06:18:08 +00:00
parent 8d5e782079 5bc88faf4f
commit e5c2519fea
40 changed files with 2371 additions and 83 deletions
@@ -8,13 +8,15 @@
 |---|---|---|---|---|---|---|---|---|
 | ISSUE-20260928-quality-heading-drift | squad / quality_gate / 缺少必填章节 / State-of-the-Art / H3 | 调研报告章节在，只因 H3 和少空格被判缺章，受控文档不投影 | high | verifying | Cursor | 2026-09-28；dev `research_1790566833205199368` manifest score 85；单测已覆盖，dev 重跑未做 | [主记录](../ops/short-term/2026-09-28-quality-gate-heading-mismatch.md#issue-20260928-quality-heading-drift) | [LESSON-20260928-quality-heading-normalize](../ops/lessons/2026-09-28-quality-heading-normalize.md) |
 | ISSUE-20260928-data-dir-doubled | workspace / data/dev/data/dev / DataDir | 出现双层 `data/dev/data/dev/workspaces` 且只有孤立 output.md，写入点未证实 | medium | open | — | 2026-09-28；`intake_1790565513115586191` 双路径对照 | [主记录](../ops/short-term/2026-09-28-quality-gate-heading-mismatch.md#issue-20260928-data-dir-doubled) | 本轮不修 |
 | ISSUE-20260928-progress-query-trigger | intent / matchTriggerTask / QUERY_PROGRESS / 编写进度 | 询问调研报告编写进度时，写动词加任务名词被识别成启动新任务，进度回复不读真实 run | medium | verifying | Cursor | 2026-09-28；「调研报告编写进度怎么样了」单测为 QUERY_PROGRESS；「开始做调研报告」仍为 research。浏览器复验见测试报告 | [主记录](../ops/short-term/2026-09-28-refresh-task-state-recovery.md) | 疑问进度句必须让行查询；显式「开始/重新」仍启动任务 |
 | ISSUE-20260928-step-empty-output | squad / stepOutputAccepted / output_digest / e3b0c442 / MaterializeTargets | 步骤已把正文写到 output.md，判定仍因 runner Success=false 或空 patch 记空输出失败 | high | verifying | Cursor | 2026-09-28；11:27 `research_1790566028724181453` 的 06_draft 盘上 41KB 但 digest 为空；单测覆盖缺交付物但盘上有正文，以及连接错误仍失败关闭 | [主记录](../ops/short-term/2026-09-28-refresh-task-state-recovery.md) | 传输/可重试失败留下的文件可能是截断的，不能当成功；缺交付物且盘上非空才接受 |
 | ISSUE-20260928-task-write-verb | intent / hasTaskWriteVerb / 帮我定 / 制定 / AMBIGUOUS_FALLBACK | 「帮我定/制定」设计开发计划任务书未进入任务触发，对话只有掂量阶段模板 | medium | resolved | Cursor | 2026-09-28；test sess-p8-u33 msg 89/90；意图单测 | [主记录](../ops/short-term/2026-09-28-test-user-doc-issues.md) | [LESSON-20260928-task-write-verb-visible-refusal](../ops/lessons/2026-09-28-task-write-verb-visible-refusal.md) |
 | ISSUE-20260928-workflow-scope-reply | coordinator / ClarificationQuestion / ErrPrerequisiteNotMet / rd_design_plan | 跨工作流澄清与研发前置未满足时，原因没有进入对话 | medium | resolved | Cursor | 2026-09-28；协调器单测。test 原例停在动词未命中，未走到本分支 | [主记录](../ops/short-term/2026-09-28-test-user-doc-issues.md) | [LESSON-20260928-task-write-verb-visible-refusal](../ops/lessons/2026-09-28-task-write-verb-visible-refusal.md) |
 | ISSUE-20260928-quality-failed-success-copy | bizdev / quality_failed / 正文已保存至 / output.md | 质量未通过仍声称正文已保存，并写出工作区绝对路径 | medium | resolved | Cursor | 2026-09-28；test cdo_quote_1790560155851280170；同步回复单测 | [主记录](../ops/short-term/2026-09-28-test-user-doc-issues.md) | [LESSON-20260928-task-write-verb-visible-refusal](../ops/lessons/2026-09-28-task-write-verb-visible-refusal.md) |
 | ISSUE-20260928-pi-runner-data-dir | agent / PI runner / DATA_DIR / MISSING_TARGET / 18866 | test 启动时根 .env 的 dev runner 地址盖过本环境 sidecar，工作区被重写到空目录 | high | resolved | Cursor | 2026-09-28；dev `4f581306`。当时 test 二进制仍是修复前制品 | [验收报告](../test-reports/2026-09-28-pi-runner-env-mismatch-test-report.md) | — |
 | ISSUE-20260928-material-chunk-embed | knowledge / chunk index failed / 8192 / 34635 tokens | 材料分块批超过嵌入模型上下文，索引失败 | medium | open | — | 2026-09-28；test 日志 project 8，文件 W020251117353305579317.docx | [主记录](../ops/short-term/2026-09-28-test-user-doc-issues.md) | 本轮不修 |
 | ISSUE-20260928-attachment-only-turn | chat / chat/stream / message cannot be empty / HTTP_400 / attachments | 只上传文档、不输入文字时 turn 被空消息 400 拒绝，前端不读信封，用户看不到任何回复 | high | resolved | Cursor | 2026-09-28；dev :8080 重载后浏览器仅附件得到梳理审阅，无关文字先回答再追问；pending 分支 API 200 | [主记录](../ops/short-term/2026-09-28-attachment-only-turn.md) | 空消息门禁要放在附件判定之后；4xx 必须读信封。见[测试报告](../test-reports/2026-09-28-attachment-only-turn-test-report.md) |
-| ISSUE-20260928-task-run-replay-502 | web / useRecoveredTaskPhases / tasks/runs / ERR_HTTP2_PROTOCOL_ERROR / 502 | 历史失败气泡恢复订阅 task run 事件时出现 HTTP/2 协议错误和 502，与附件空消息 400 无因果关系 | medium | investigating | Cursor | 2026-09-28；用户控制台 `run_1790558010622744831`；本轮未改订阅逻辑 | [主记录](../ops/short-term/2026-09-28-attachment-only-turn.md#范围外) | 未修，保持观察 |
+| ISSUE-20260928-task-run-replay-502 | web / useRecoveredTaskPhases / tasks/runs / ERR_HTTP2_PROTOCOL_ERROR / 502 | 历史失败气泡恢复订阅 task run 事件时出现 HTTP/2 协议错误和 502，与附件空消息 400 无因果关系 | medium | verifying | Cursor | 2026-09-28；终态重放改为最多 4 次退避，401/404 立即停止；进行中订阅外层只再补 1 次，避免和内部 40 次重试相乘。尚未在真实 502 上复验 | [主记录](../ops/short-term/2026-09-28-refresh-task-state-recovery.md) | 重放失败要可重试；鉴权与不存在不要空转 |
 | ISSUE-20260928-activation-preview-precheck | web / projectSettings / activation-preview / NO_GOVERNED_DOCUMENT / clientDutyIssues | 全部文档未纳入人工审批时前端仍 POST activation-preview，控制台 400 | medium | resolved | Cursor | 2026-09-28；dev 项目 6023 默认计划无 activation-preview 请求，兼任后 preview 200 | [主记录](../ops/short-term/2026-09-28-activation-preview-precheck.md#issue-20260928-activation-preview-precheck) | [LESSON-20260928-activation-preview-client-gate](../ops/lessons/2026-09-28-activation-preview-client-gate.md) |
 | ISSUE-20260928-admin-secret-password-form | web / EnvSettingsSection / password / form / secrets | 管理台 5 个密钥输入不在 form 内，Chromium 报 password DOM 警告 | low | resolved | Cursor | 2026-09-28；dev 模型与密钥页 passwordInputs=5 outsideForm=0 | [主记录](../ops/short-term/2026-09-28-activation-preview-precheck.md#issue-20260928-admin-secret-password-form) | 密码框放进 form；autocomplete 不能单独消除该 DOM 警告 |
 | ISSUE-20260928-squad-commit-authority | bizdev / docflow / dutyAuthorizer / uow permission denied / OnBehalfOf / bd_intake | 非编制人触发受控小队任务时，账本提交用触发者身份，被编制人职责门拒绝，正文无法落盘 | high | fixing | Cursor | 2026-09-28；dev task_runs user 74 ≠ author 4（doc_proj6022_bd_intake）；单测已绿，dev 重跑未做 | [主记录](../ops/short-term/2026-09-28-squad-commit-authority.md#issue-20260928-squad-commit-authority) | [LESSON-20260928-squad-commit-as-author](../ops/lessons/2026-09-28-squad-commit-as-author.md) |
@@ -0,0 +1,57 @@
 ---
 title: "刷新后恢复任务进度并纠正进度问询"
 task_id: refresh-task-state-recovery-2026-09-28
 status: in_progress
 created: 2026-09-28
 updated: 2026-09-28
 owner: Cursor
 task_type: fix
 policy_revision: "AGENTS.md + agent-worktree-workflow @ 2026-09-21"
 branch: fix/refresh-task-state-recovery-2026-09-28
 worktree: .worktrees/refresh-task-state-recovery-2026-09-28
 base_dev_sha: "4c42d1bf74e262930bcf127e517ab6b76a0dda71"
 write_scope:
  - internal
  - web
  - .doc
 mutex: refresh-task-state-recovery
 review_state: pending
 integration_state: branch_only
 deployment: not_loaded
 cleanup: pending
 remote_sync: not_requested
 ---
 # 刷新后恢复任务进度并纠正进度问询
 ## 用户目标与授权
 用户发起调研报告后，聊天下方的任务进度在刷新页面时消失。随后两次询问「调研报告编写进度怎么样了」，没有看到调研报告生成。用户要求浏览器和服务器同时记下进行中的任务：刷新或重连后先显示本地缓存，再按会话和任务标识向服务器收敛。授权范围同时包含进度问句被当成新任务，以及 11:27 / 11:38 两次步骤被判空输出失败。
 ## 检索
 - 问题台账命中 ISSUE-20260928-task-run-replay-502（历史失败气泡重放 502）。本轮给重放补了退避，并保持终态 `replay_only`。
 - catalog overlap 探测无硬重叠。mutex 为 `refresh-task-state-recovery`，不占用 `chat+coordinator` 与 `bizdev+docflow`。
 ## 根因
 1. `run_id` 与阶段快照只在任务终态写回助手消息。窗口缓存只保存服务端窗口，刷新后没有锚点，进度卡片无法重放。
 2. 「调研报告编写进度怎么样了」含写动词「编写」和任务名词「调研报告」。`matchTriggerTask` 排在进度查询之前，于是又启动了 research。进度回复只读模板，不读真实 run。
 3. 步骤判定把 runner 的 `Success=false` 当成失败，即使盘上已经有正文。连接错误留下的文件仍可能是截断的，这条不能放宽。
 ## 修复
 - run 启动和绑定助手消息时写入 `{run_id, task_status: running}`。
 - `GET /api/v1/tasks/runs?session_id=&active=1` 只返回当前用户、当前会话里仍为 pending/running 的 run。
 - 浏览器 `smdf_agent_workbench_v6_active_tasks` 在窗口就绪后先画出进行中卡片，再用活跃列表收敛；终态删除缓存。
 - 有 run 锚点但还没有 `task_status` 和阶段快照时，消息保持 `streaming_status`。终态重放对 502 退避，401/404 不再试。
 - 疑问进度句让行 `QUERY_PROGRESS`。回复读取该会话最近一次 run 的真实状态。
 - 盘上已有非空正文，且失败原因不是传输/可重试错误时，步骤接受该文件。空 patch 不覆盖已有 `output.md`。
 ## 验证
 见 [测试报告](../../test-reports/2026-09-28-refresh-task-state-recovery-test-report.md)。
 ## 五态
 合入与 dev 重载完成后回写。当前仍是任务分支上的实现。
@@ -0,0 +1,80 @@
 # 刷新后任务状态恢复 测试报告
 ## 一、 测试背景 (Background)
 刷新页面后，进行中的调研报告进度卡片消失。询问「调研报告编写进度怎么样了」又启动了新的 research 工作区。11:27 的 `step_06_draft` 与 11:38 的并行步骤被记成空输出失败，盘上的 `output.md` 已经有正文。
 ## 二、 测试目标 (Test Objectives)
 - 进行中的 run 在刷新后仍能回到进度卡片。
 - 疑问进度句查询真实 run，不新建工作区。
 - 盘上已有非空正文且不是传输失败时，步骤不再被空结果否决。
 - 连接错误留下的文件仍然失败关闭。
 ## 三、 测试方案与环境 (Test Plan & Environment)
 - OS：Linux。工作树：`.worktrees/refresh-task-state-recovery-2026-09-28`，基线 `4c42d1bf`。
 - 层级：Go/接口短测、前端单测与构建、架构边界检查。产品 PI 浏览器链路在 dev 重载后另记，本文件先记录已跑命令。
 - 证据目录：`.doc/test-reports/assets/2026-09-28-refresh-task-state-recovery/`。
 ## 四、 预期结果 (Expected Results)
 - `go vet ./...` 无输出。
 - 本改动涉及的包 `go test -short` 通过。
 - `web` 的 `npm test` 与 `npm run build` 通过。
 - 「调研报告编写进度怎么样了」不是写意图；「开始做调研报告」解析为 research。
 - 缺交付物但盘上有正文时步骤接受；`Connection error` 且可重试时拒绝。
 ## 五、 实际测试结果与判定 (Actual Results & Evaluation)
 ### 用例 1：意图与进度回复
 - 输入：`go test -short -count=1 ./internal/intent/ ./internal/coordinator/`
 - 实际输出：两包 `ok`（intent 0.179s，coordinator 0.034s，见 `go-test-short.txt`）。新增用例「调研报告编写进度怎么样了」为 `QUERY_PROGRESS`，「开始做调研报告」为 `research`。
 - 判定：PASS
 - 耗时：见短测日志
 ### 用例 2：空输出判定与空 patch
 - 输入：`go test -short -count=1 ./internal/squad/ ./internal/agent/`
 - 实际输出：两包 `ok`（squad 0.041s，agent 2.252s）。连接错误仍失败关闭的既有用例通过。
 - 判定：PASS
 ### 用例 3：锚点、活跃列表、消息映射、重放退避、本地缓存
 - 输入：`go test -short ./internal/taskrun/ ./internal/server/`；`npm test` 中的 `activeTaskCache`、`useRecoveredTaskPhases`、`useChatWindow`、`templateMatch`
 - 实际输出：taskrun `ok` 0.171s，server `ok` 20.914s。`npm test` 532 通过、vitest 34 通过，`NPM_TEST:0`。
 - 判定：PASS
 ### 用例 4：全量短测与架构边界
 - 输入：`go vet ./...`；`go test -short -count=1 ./...`；`./scripts/arch-check.sh --mode block`
 - 实际输出：`VET:0`。短测唯一失败是 `internal/archtest`：`NEW import coordinator internal/coordinator/engine_task.go projectworkflow`。该文件相对基线 `4c42d1bf` 的 diff 为 0，导入来自已合入的 `agent-reply-llm-narrative`，不是本任务新增。`ARCH:1` 与之相同。
 - 判定：本任务包 PASS；架构边界 FAIL 记为基线既有项，本任务未改该导入
 ### 用例 5：前端构建
 - 输入：`npm run build`
 - 实际输出：`NPM_BUILD:0`，precompress 写出 66 个 br/gz。
 - 判定：PASS
 ### 用例 6：dev 浏览器与真实 PI Runner
 - 输入：发起调研报告，执行中刷新，完成后刷新，再问编写进度。
 - 实际输出：本段写入时尚未在已加载新代码的 dev 上运行。
 - 判定：未运行
 ## 六、 自动化测试套件汇总 (Test Suite Execution)
 | 检查 | 结果 |
 |---|---|
 | go vet ./... | PASS（VET:0） |
 | go test -short ./... | 除既有 archtest 外 PASS |
 | arch-check --mode block | FAIL（基线既有 coordinator→projectworkflow） |
 | npm test | PASS（532 + 34） |
 | npm run build | PASS |
 | dev 浏览器 + 真实 PI | 未运行 |
 ## 七、 测试结论与后续优化建议 (Conclusion & Next Steps)
 实现与单测、前端构建已通过。交付终点还差：把候选合入 local dev、重载服务后，用真实 PI Runner 走刷新恢复和进度问句。架构边界失败不由本 diff 引入，不在本 mutex 里改 `engine_task.go`。
@@ -0,0 +1,16 @@
 arch-check mode=block
 boundaries=/home/weichen/projects/Agent-System/.worktrees/refresh-task-state-recovery-2026-09-28/.doc/system-design/module-boundaries.yaml
 --- FAIL: TestArchitectureBoundaries (0.19s)
    arch_test.go:32: archtest findings=1 excepted=0 unmatched=1
          import=1
          sql_write=0
          path=0
          event=0
        REGRESSION import: 1 > baseline 0
        NEW import coordinator internal/coordinator/engine_task.go projectworkflow (production import coordinator -> projectworkflow)
 FAIL
 FAIL	chat-agent-system/internal/archtest	0.195s
 FAIL
 arch-check: new, expired, or unowned findings failed go test ./internal/archtest
 ARCH:1
@@ -0,0 +1,99 @@
 ?   	chat-agent-system/cmd/artifact-vault	[no test files]
 ok  	chat-agent-system/cmd/bizdev-cli	2.985s
 ?   	chat-agent-system/cmd/dataset	[no test files]
 ok  	chat-agent-system/cmd/document-content-reconcile	0.003s
 ?   	chat-agent-system/cmd/document-revision-backfill	[no test files]
 ?   	chat-agent-system/cmd/eventmode	[no test files]
 ?   	chat-agent-system/cmd/events	[no test files]
 ?   	chat-agent-system/cmd/maintpage	[no test files]
 ?   	chat-agent-system/cmd/recategorize	[no test files]
 ok  	chat-agent-system/cmd/server	0.012s
 ok  	chat-agent-system/internal/agent	2.252s
 ok  	chat-agent-system/internal/agentx	0.023s
 ok  	chat-agent-system/internal/agentx/abtest	0.004s
 ok  	chat-agent-system/internal/agentx/llm	0.024s
 ok  	chat-agent-system/internal/agentx/session	0.008s
 ok  	chat-agent-system/internal/agentx/tools	0.006s
 ok  	chat-agent-system/internal/app	0.046s
 ok  	chat-agent-system/internal/app/chatflow	0.018s
 ok  	chat-agent-system/internal/app/docadvance	0.015s
 ok  	chat-agent-system/internal/app/docflow	0.009s
 ok  	chat-agent-system/internal/app/docnotify	0.005s
 ok  	chat-agent-system/internal/app/docproject	0.003s
 ?   	chat-agent-system/internal/app/eventsconsumer	[no test files]
 ok  	chat-agent-system/internal/app/fav	0.007s
 ok  	chat-agent-system/internal/app/intentflow	0.088s
 ok  	chat-agent-system/internal/app/stageflow	0.010s
 ok  	chat-agent-system/internal/app/uow	0.005s
 --- FAIL: TestArchitectureBoundaries (0.50s)
    arch_test.go:32: archtest findings=1 excepted=0 unmatched=1
          import=1
          sql_write=0
          path=0
          event=0
        REGRESSION import: 1 > baseline 0
        NEW import coordinator internal/coordinator/engine_task.go projectworkflow (production import coordinator -> projectworkflow)
 FAIL
 FAIL	chat-agent-system/internal/archtest	0.507s
 ok  	chat-agent-system/internal/artifactvault	0.005s
 ok  	chat-agent-system/internal/auth	1.494s
 ok  	chat-agent-system/internal/authz	0.007s
 ok  	chat-agent-system/internal/bizdev	0.418s
 ok  	chat-agent-system/internal/chat	0.067s
 ok  	chat-agent-system/internal/config	0.009s
 ok  	chat-agent-system/internal/conflictresolve	0.006s
 ?   	chat-agent-system/internal/contracts/vocabulary	[no test files]
 ok  	chat-agent-system/internal/coordinator	0.034s
 ok  	chat-agent-system/internal/ctxmgr	0.174s
 ok  	chat-agent-system/internal/dataset	0.010s
 ok  	chat-agent-system/internal/db	0.007s
 ok  	chat-agent-system/internal/dialoguectx	0.007s
 ok  	chat-agent-system/internal/doccontent	0.004s
 ok  	chat-agent-system/internal/docduty	0.004s
 ok  	chat-agent-system/internal/docgov	0.006s
 ok  	chat-agent-system/internal/docquery	0.004s
 ok  	chat-agent-system/internal/docrevision	0.017s
 ok  	chat-agent-system/internal/errcode	0.007s
 ok  	chat-agent-system/internal/errorevents	0.008s
 ok  	chat-agent-system/internal/events	0.012s
 ok  	chat-agent-system/internal/freedoc	0.035s
 ok  	chat-agent-system/internal/fsutil	0.005s
 ok  	chat-agent-system/internal/handoff	0.008s
 ok  	chat-agent-system/internal/httpx	0.004s
 ok  	chat-agent-system/internal/intent	0.179s
 ok  	chat-agent-system/internal/jevdecide	0.274s
 ok  	chat-agent-system/internal/kb	0.013s
 ok  	chat-agent-system/internal/localpg	0.598s
 ok  	chat-agent-system/internal/maintenance	0.005s
 ok  	chat-agent-system/internal/materialindex	0.007s
 ok  	chat-agent-system/internal/maturity	0.007s
 ok  	chat-agent-system/internal/mineru	3.008s
 ok  	chat-agent-system/internal/notification	0.005s
 ?   	chat-agent-system/internal/observability	[no test files]
 ok  	chat-agent-system/internal/projectcontext	0.005s
 ok  	chat-agent-system/internal/projectinstance	0.010s
 ok  	chat-agent-system/internal/projectworkflow	0.003s
 ok  	chat-agent-system/internal/quote	0.012s
 ok  	chat-agent-system/internal/replyexport	0.002s
 ok  	chat-agent-system/internal/server	20.914s
 ok  	chat-agent-system/internal/squad	0.041s
 ok  	chat-agent-system/internal/storage	0.003s
 ok  	chat-agent-system/internal/stream	0.003s
 ok  	chat-agent-system/internal/stream/transport	0.003s
 ok  	chat-agent-system/internal/supplementimpact	0.003s
 ok  	chat-agent-system/internal/swboot	0.070s
 ok  	chat-agent-system/internal/sysadmin	0.003s
 ok  	chat-agent-system/internal/taskrun	0.171s
 ok  	chat-agent-system/internal/template	1.059s
 ok  	chat-agent-system/internal/traceimpact	0.003s
 ok  	chat-agent-system/internal/upload	0.002s
 ?   	chat-agent-system/internal/workflow	[no test files]
 ok  	chat-agent-system/internal/workflow/runtime	0.004s
 ok  	chat-agent-system/internal/workflowdef	0.004s
 ok  	chat-agent-system/internal/workflowscope	0.002s
 ok  	chat-agent-system/internal/workspace	0.003s
 ?   	chat-agent-system/templates	[no test files]
 ?   	chat-agent-system/web	[no test files]
 FAIL
 TEST:1
@@ -0,0 +1 @@
 VET:0
@@ -0,0 +1,291 @@
 > shuimu-workbench@1.0.0 build
 > tsc && vite build && node scripts/precompress-dist.mjs
 vite v7.3.6 building client environment for production...
 transforming...
 ✓ 4169 modules transformed.
 rendering chunks...
 computing gzip size...
 dist/index.html                                         0.72 kB │ gzip:   0.45 kB
 dist/assets/KaTeX_Size3-Regular-CTq5MqoE.woff           4.42 kB
 dist/assets/KaTeX_Size4-Regular-Dl5lxZxV.woff2          4.93 kB
 dist/assets/KaTeX_Size2-Regular-Dy4dx90m.woff2          5.21 kB
 dist/assets/KaTeX_Size1-Regular-mCD8mA8B.woff2          5.47 kB
 dist/assets/KaTeX_Size4-Regular-BF-4gkZK.woff           5.98 kB
 dist/assets/KaTeX_Size2-Regular-oD1tc_U0.woff           6.19 kB
 dist/assets/KaTeX_Size1-Regular-C195tn64.woff           6.50 kB
 dist/assets/KaTeX_Caligraphic-Regular-Di6jR-x-.woff2    6.91 kB
 dist/assets/KaTeX_Caligraphic-Bold-Dq_IR9rO.woff2       6.91 kB
 dist/assets/KaTeX_Size3-Regular-DgpXs0kz.ttf            7.59 kB
 dist/assets/KaTeX_Caligraphic-Regular-CTRA-rTL.woff     7.66 kB
 dist/assets/KaTeX_Caligraphic-Bold-BEiXGLvX.woff        7.72 kB
 dist/assets/KaTeX_Script-Regular-D3wIWfF6.woff2         9.64 kB
 dist/assets/KaTeX_SansSerif-Regular-DDBCnlJ7.woff2     10.34 kB
 dist/assets/KaTeX_Size4-Regular-DWFBv043.ttf           10.36 kB
 dist/assets/KaTeX_Script-Regular-D5yQViql.woff         10.59 kB
 dist/assets/KaTeX_Fraktur-Regular-CTYiF6lA.woff2       11.32 kB
 dist/assets/KaTeX_Fraktur-Bold-CL6g_b3V.woff2          11.35 kB
 dist/assets/KaTeX_Size2-Regular-B7gKUWhC.ttf           11.51 kB
 dist/assets/KaTeX_SansSerif-Italic-C3H0VqGB.woff2      12.03 kB
 dist/assets/KaTeX_SansSerif-Bold-D1sUS0GD.woff2        12.22 kB
 dist/assets/KaTeX_Size1-Regular-Dbsnue_I.ttf           12.23 kB
 dist/assets/KaTeX_SansSerif-Regular-CS6fqUqJ.woff      12.32 kB
 dist/assets/KaTeX_Caligraphic-Regular-wX97UBjC.ttf     12.34 kB
 dist/assets/KaTeX_Caligraphic-Bold-ATXxdsX0.ttf        12.37 kB
 dist/assets/KaTeX_Fraktur-Regular-Dxdc4cR9.woff        13.21 kB
 dist/assets/KaTeX_Fraktur-Bold-BsDP51OF.woff           13.30 kB
 dist/assets/KaTeX_Typewriter-Regular-CO6r4hn1.woff2    13.57 kB
 dist/assets/KaTeX_SansSerif-Italic-DN2j7dab.woff       14.11 kB
 dist/assets/KaTeX_SansSerif-Bold-DbIhKOiC.woff         14.41 kB
 dist/assets/KaTeX_Typewriter-Regular-C0xS9mPB.woff     16.03 kB
 dist/assets/KaTeX_Math-BoldItalic-CZnvNsCZ.woff2       16.40 kB
 dist/assets/KaTeX_Math-Italic-t53AETM-.woff2           16.44 kB
 dist/assets/KaTeX_Script-Regular-C5JkGWo-.ttf          16.65 kB
 dist/assets/KaTeX_Main-BoldItalic-DxDJ3AOS.woff2       16.78 kB
 dist/assets/KaTeX_Main-Italic-NWA7e6Wa.woff2           16.99 kB
 dist/assets/KaTeX_Math-BoldItalic-iY-2wyZ7.woff        18.67 kB
 dist/assets/KaTeX_Math-Italic-DA0__PXp.woff            18.75 kB
 dist/assets/KaTeX_Main-BoldItalic-SpSLRI95.woff        19.41 kB
 dist/assets/KaTeX_SansSerif-Regular-BNo7hRIc.ttf       19.44 kB
 dist/assets/KaTeX_Fraktur-Regular-CB_wures.ttf         19.57 kB
 dist/assets/KaTeX_Fraktur-Bold-BdnERNNW.ttf            19.58 kB
 dist/assets/KaTeX_Main-Italic-BMLOBm91.woff            19.68 kB
 dist/assets/KaTeX_SansSerif-Italic-YYjJ1zSn.ttf        22.36 kB
 dist/assets/KaTeX_SansSerif-Bold-CFMepnvq.ttf          24.50 kB
 dist/assets/KaTeX_Main-Bold-Cx986IdX.woff2             25.32 kB
 dist/assets/KaTeX_Main-Regular-B22Nviop.woff2          26.27 kB
 dist/assets/KaTeX_Typewriter-Regular-D3Ib7_Hf.ttf      27.56 kB
 dist/assets/KaTeX_AMS-Regular-BQhdFMY1.woff2           28.08 kB
 dist/assets/KaTeX_Main-Bold-Jm3AIy58.woff              29.91 kB
 dist/assets/KaTeX_Main-Regular-Dr94JaBh.woff           30.77 kB
 dist/assets/KaTeX_Math-BoldItalic-B3XSjfu4.ttf         31.20 kB
 dist/assets/KaTeX_Math-Italic-flOr_0UB.ttf             31.31 kB
 dist/assets/KaTeX_Main-BoldItalic-DzxPMmG6.ttf         32.97 kB
 dist/assets/KaTeX_AMS-Regular-DMm9YOAa.woff            33.52 kB
 dist/assets/KaTeX_Main-Italic-3WenGoN9.ttf             33.58 kB
 dist/assets/KaTeX_Main-Bold-waoOVXN0.ttf               51.34 kB
 dist/assets/KaTeX_Main-Regular-ypZvNtVU.ttf            53.58 kB
 dist/assets/KaTeX_AMS-Regular-DRggAlZN.ttf             63.63 kB
 dist/assets/index-Bq7tVxen.css                          6.58 kB │ gzip:   1.99 kB
 dist/assets/katex-BOvhZl6_.css                         30.25 kB │ gzip:   8.09 kB
 dist/assets/channel-DHR9WBLM.js                         0.12 kB │ gzip:   0.13 kB
 dist/assets/katexRender-DyvI-2y4.js                     0.15 kB │ gzip:   0.15 kB
 dist/assets/init-Gi6I4Gst.js                            0.15 kB │ gzip:   0.13 kB
 dist/assets/chunk-2Q5K7J3B-CHsI7zem.js                  0.19 kB │ gzip:   0.16 kB
 dist/assets/chunk-JWPE2WC7-CeOXuZlw.js                  0.23 kB │ gzip:   0.17 kB
 dist/assets/chunk-XXDRQBXY-Cu_mBr7x.js                  0.23 kB │ gzip:   0.21 kB
 dist/assets/git-branch-D2R7ViCv.js                      0.26 kB │ gzip:   0.21 kB
 dist/assets/chunk-5VM5RSS4-lzTX0tjp.js                  0.37 kB │ gzip:   0.27 kB
 dist/assets/stateDiagram-v2-MP3YSRHH-dJKbyi-1.js        0.43 kB │ gzip:   0.30 kB
 dist/assets/classDiagram-ZZMXUADV-CTPa2PBn.js           0.47 kB │ gzip:   0.32 kB
 dist/assets/classDiagram-v2-VYDZK3BY-CTPa2PBn.js        0.47 kB │ gzip:   0.32 kB
 dist/assets/chunk-POPQ4Y6H-B2jpZUkp.js                  0.51 kB │ gzip:   0.36 kB
 dist/assets/swimlanesDiagram-VR7AAH4N-jP-BQxXC.js       0.55 kB │ gzip:   0.37 kB
 dist/assets/calculator-K0r_LA6K.js                      0.56 kB │ gzip:   0.30 kB
 dist/assets/infoDiagram-27XIBGKW-BO8R-0BH.js            0.59 kB │ gzip:   0.40 kB
 dist/assets/sizeCapture-INFHLROL-KBPVYQON.js            0.94 kB │ gzip:   0.53 kB
 dist/assets/downloadDocx-Ep6Y-mWJ.js                    1.13 kB │ gzip:   0.77 kB
 dist/assets/ordinal-Cboi1Yqb.js                         1.19 kB │ gzip:   0.57 kB
 dist/assets/railroadDiagram-O6MQD6OU-3o3NI4V-.js        1.61 kB │ gzip:   0.76 kB
 dist/assets/MathBlockView-CP3gIAfz.js                   1.67 kB │ gzip:   0.97 kB
 dist/assets/abnfDiagram-VCTEODGH-Bv_S-VQO.js            1.81 kB │ gzip:   0.89 kB
 dist/assets/chunk-F27PBJKO-DwG2Zkg1.js                  1.88 kB │ gzip:   0.83 kB
 dist/assets/pegDiagram-XKGWAZYB-BRKb5jpA.js             1.96 kB │ gzip:   0.92 kB
 dist/assets/ebnfDiagram-PWID7BFC-CAlTqs1O.js            2.01 kB │ gzip:   0.88 kB
 dist/assets/xlsxRenderer-BQPATLFB.js                    2.79 kB │ gzip:   1.69 kB
 dist/assets/arc-DOxBXeCg.js                             3.44 kB │ gzip:   1.48 kB
 dist/assets/MermaidBlockView-BNZ12-FU.js                3.69 kB │ gzip:   1.80 kB
 dist/assets/diagram-Z3DM3KII-iO8-m9Hl.js                4.21 kB │ gzip:   1.83 kB
 dist/assets/defaultLocale-DX6XiGOO.js                   4.69 kB │ gzip:   2.18 kB
 dist/assets/renderClipboardHtml-BXoFMW-3.js             5.18 kB │ gzip:   1.78 kB
 dist/assets/ExcelViewer-BZSJZn5Y.js                     5.55 kB │ gzip:   2.21 kB
 dist/assets/linear-BDZQyaJG.js                          5.66 kB │ gzip:   2.31 kB
 dist/assets/diagram-UQ7AKVKN-BnBJ_2R3.js                6.08 kB │ gzip:   2.62 kB
 dist/assets/pieDiagram-E7YTZNPT-nrtuGrNr.js             6.29 kB │ gzip:   2.62 kB
 dist/assets/ChatExportModal-Bfjz4958.js                 7.58 kB │ gzip:   3.29 kB
 dist/assets/diagram-S7CK7UJ4-T-Bxl4t1.js                8.22 kB │ gzip:   3.67 kB
 dist/assets/cynefinDiagram-5FMLGOSQ-CnH3Z2JB.js        10.10 kB │ gzip:   3.63 kB
 dist/assets/stateDiagram-D77RDMKH-DVKtcg-v.js          10.36 kB │ gzip:   3.62 kB
 dist/assets/dagre-GXQ25YYZ-4R6bcum3.js                 10.68 kB │ gzip:   4.03 kB
 dist/assets/diagram-VSXAHHWV-DLnsWaIP.js               10.72 kB │ gzip:   4.13 kB
 dist/assets/chunk-SVP7TREG-BhgGIDFD.js                 15.74 kB │ gzip:   4.41 kB
 dist/assets/diagram-VX7I27RA-Duotn1-e.js               15.96 kB │ gzip:   5.70 kB
 dist/assets/ishikawaDiagram-5VMMS53U-Cg4LfJyo.js       17.47 kB │ gzip:   6.64 kB
 dist/assets/kanban-definition-UXKFOSKX-BALPAjTC.js     20.29 kB │ gzip:   7.20 kB
 dist/assets/BlockEditorModal-BOLlSDTH.js               22.24 kB │ gzip:   6.47 kB
 dist/assets/mindmap-definition-YA3MSWOX-pgZMQ8uL.js    23.18 kB │ gzip:   7.83 kB
 dist/assets/sankeyDiagram-P5KCCOFB-cHAs8eQM.js         23.20 kB │ gzip:   8.56 kB
 dist/assets/journeyDiagram-3NMN7TZE-DXSneWK0.js        23.57 kB │ gzip:   8.35 kB
 dist/assets/wardleyDiagram-VM6X3IG4-83YMC1gW.js        25.50 kB │ gzip:   6.79 kB
 dist/assets/gitGraphDiagram-WWUBYQGX-3ajohVe1.js       29.42 kB │ gzip:   8.70 kB
 dist/assets/FilePreviewPanel-BaAl4sya.js               30.12 kB │ gzip:   8.97 kB
 dist/assets/erDiagram-RLTQ6QDP-BDpOdOIc.js             30.79 kB │ gzip:  10.59 kB
 dist/assets/timeline-definition-24CTP7MA-d7EudEZG.js   30.83 kB │ gzip:  10.15 kB
 dist/assets/requirementDiagram-BXWQKSXE-2YVxwHhy.js    31.00 kB │ gzip:   9.73 kB
 dist/assets/quadrantDiagram-AXDQQJYC-DHNugfXZ.js       34.01 kB │ gzip:   9.98 kB
 dist/assets/layout-BrexQxSk.js                         35.44 kB │ gzip:  12.75 kB
 dist/assets/chunk-IMKFNOWR-DL3J6pk5.js                 38.08 kB │ gzip:  12.51 kB
 dist/assets/docxRenderer-Ilu9VYYG.js                   40.90 kB │ gzip:  12.04 kB
 dist/assets/blockDiagram-I7D4REHJ-ozi4Xk9Q.js          41.76 kB │ gzip:  13.58 kB
 dist/assets/vennDiagram-4TSXK5OY-Dswu0LDK.js           41.96 kB │ gzip:  15.70 kB
 dist/assets/xychartDiagram-S5SC5T6Z-Do0p5dm5.js        44.51 kB │ gzip:  12.60 kB
 dist/assets/chunk-TICWLB2K-CsHC9dlc.js                 49.18 kB │ gzip:  15.74 kB
 dist/assets/flowDiagram-HODETNUW-BGwlzTt0.js           61.85 kB │ gzip:  19.87 kB
 dist/assets/c4Diagram-7LVT6UL2-MxYk0iqL.js             65.47 kB │ gzip:  18.79 kB
 dist/assets/ganttDiagram-EL5Y4UJY-ijkjYohu.js          69.70 kB │ gzip:  23.56 kB
 dist/assets/AdminConsoleView-C6f6_5zf.js               74.34 kB │ gzip:  18.18 kB
 dist/assets/cose-bilkent-JH36ORCC-DW6Qeju_.js          81.70 kB │ gzip:  22.47 kB
 dist/assets/swimlanes-42K2YHIH-BaJ44xBg.js            115.79 kB │ gzip:  42.51 kB
 dist/assets/sequenceDiagram-WJ2MYXX4-BqEzakUS.js      116.96 kB │ gzip:  31.13 kB
 dist/assets/architectureDiagram-5GKGNRK7-BW5RJe7u.js  151.83 kB │ gzip:  43.14 kB
 dist/assets/DocxViewer-thvChHTS.js                    175.87 kB │ gzip:  52.31 kB
 dist/assets/katex-HP8lGamR.js                         261.31 kB │ gzip:  77.57 kB
 dist/assets/katex-BJxdM42i.js                         261.44 kB │ gzip:  77.80 kB
 dist/assets/index-BXtlAAcH.js                         365.75 kB │ gzip: 104.72 kB
 dist/assets/xlsx-Cig0qvpi.js                          424.39 kB │ gzip: 141.60 kB
 dist/assets/cytoscape.esm-BB4DxJjf.js                 443.04 kB │ gzip: 141.98 kB
 dist/assets/mermaid.core-C3o1vFDn.js                  682.57 kB │ gzip: 168.90 kB
 dist/assets/cynefin-OW5HDTMX-Ca8YX88f.js              687.91 kB │ gzip: 154.14 kB
 dist/assets/index-cbAtdfhj.js                         814.04 kB │ gzip: 239.32 kB
 (!) Some chunks are larger than 500 kB after minification. Consider:
 - Using dynamic import() to code-split the application
 - Use build.rollupOptions.output.manualChunks to improve chunking: https://rollupjs.org/configuration-options/#output-manualchunks
 - Adjust chunk size limit for this warning via build.chunkSizeWarningLimit.
 ✓ built in 8.74s
 precompress: 66 candidate(s) under /home/weichen/projects/Agent-System/.worktrees/refresh-task-state-recovery-2026-09-28/web/dist
  br   assets/AdminConsoleView-C6f6_5zf.js  74340 → 14847 (−80.0%)
  gzip assets/AdminConsoleView-C6f6_5zf.js  74340 → 17742 (−76.1%)
  br   assets/BlockEditorModal-BOLlSDTH.js  22241 → 5389 (−75.8%)
  gzip assets/BlockEditorModal-BOLlSDTH.js  22241 → 6375 (−71.3%)
  br   assets/ChatExportModal-Bfjz4958.js  7581 → 2732 (−64.0%)
  gzip assets/ChatExportModal-Bfjz4958.js  7581 → 3242 (−57.2%)
  br   assets/DocxViewer-thvChHTS.js  175867 → 46191 (−73.7%)
  gzip assets/DocxViewer-thvChHTS.js  175867 → 52082 (−70.4%)
  br   assets/ExcelViewer-BZSJZn5Y.js  5552 → 1885 (−66.0%)
  gzip assets/ExcelViewer-BZSJZn5Y.js  5552 → 2194 (−60.5%)
  br   assets/FilePreviewPanel-BaAl4sya.js  30116 → 7564 (−74.9%)
  gzip assets/FilePreviewPanel-BaAl4sya.js  30116 → 8813 (−70.7%)
  br   assets/MathBlockView-CP3gIAfz.js  1669 → 800 (−52.1%)
  gzip assets/MathBlockView-CP3gIAfz.js  1669 → 962 (−42.4%)
  br   assets/MermaidBlockView-BNZ12-FU.js  3690 → 1499 (−59.4%)
  gzip assets/MermaidBlockView-BNZ12-FU.js  3690 → 1794 (−51.4%)
  br   assets/abnfDiagram-VCTEODGH-Bv_S-VQO.js  1809 → 785 (−56.6%)
  gzip assets/abnfDiagram-VCTEODGH-Bv_S-VQO.js  1809 → 895 (−50.5%)
  br   assets/arc-DOxBXeCg.js  3441 → 1296 (−62.3%)
  gzip assets/arc-DOxBXeCg.js  3441 → 1455 (−57.7%)
  br   assets/architectureDiagram-5GKGNRK7-BW5RJe7u.js  151827 → 36975 (−75.6%)
  gzip assets/architectureDiagram-5GKGNRK7-BW5RJe7u.js  151827 → 42841 (−71.8%)
  br   assets/blockDiagram-I7D4REHJ-ozi4Xk9Q.js  41757 → 11969 (−71.3%)
  gzip assets/blockDiagram-I7D4REHJ-ozi4Xk9Q.js  41757 → 13469 (−67.7%)
  br   assets/c4Diagram-7LVT6UL2-MxYk0iqL.js  65473 → 15133 (−76.9%)
  gzip assets/c4Diagram-7LVT6UL2-MxYk0iqL.js  65473 → 18679 (−71.5%)
  br   assets/chunk-F27PBJKO-DwG2Zkg1.js  1883 → 742 (−60.6%)
  gzip assets/chunk-F27PBJKO-DwG2Zkg1.js  1883 → 831 (−55.9%)
  br   assets/chunk-IMKFNOWR-DL3J6pk5.js  38076 → 10974 (−71.2%)
  gzip assets/chunk-IMKFNOWR-DL3J6pk5.js  38076 → 12434 (−67.3%)
  br   assets/chunk-SVP7TREG-BhgGIDFD.js  15735 → 3966 (−74.8%)
  gzip assets/chunk-SVP7TREG-BhgGIDFD.js  15735 → 4418 (−71.9%)
  br   assets/chunk-TICWLB2K-CsHC9dlc.js  49178 → 13033 (−73.5%)
  gzip assets/chunk-TICWLB2K-CsHC9dlc.js  49178 → 15645 (−68.2%)
  br   assets/cose-bilkent-JH36ORCC-DW6Qeju_.js  81701 → 19574 (−76.0%)
  gzip assets/cose-bilkent-JH36ORCC-DW6Qeju_.js  81701 → 22391 (−72.6%)
  br   assets/cynefin-OW5HDTMX-Ca8YX88f.js  687913 → 124540 (−81.9%)
  gzip assets/cynefin-OW5HDTMX-Ca8YX88f.js  687913 → 153068 (−77.7%)
  br   assets/cynefinDiagram-5FMLGOSQ-CnH3Z2JB.js  10097 → 3248 (−67.8%)
  gzip assets/cynefinDiagram-5FMLGOSQ-CnH3Z2JB.js  10097 → 3620 (−64.1%)
  br   assets/cytoscape.esm-BB4DxJjf.js  443036 → 119602 (−73.0%)
  gzip assets/cytoscape.esm-BB4DxJjf.js  443036 → 141741 (−68.0%)
  br   assets/dagre-GXQ25YYZ-4R6bcum3.js  10676 → 3680 (−65.5%)
  gzip assets/dagre-GXQ25YYZ-4R6bcum3.js  10676 → 4046 (−62.1%)
  br   assets/defaultLocale-DX6XiGOO.js  4688 → 2012 (−57.1%)
  gzip assets/defaultLocale-DX6XiGOO.js  4688 → 2166 (−53.8%)
  br   assets/diagram-S7CK7UJ4-T-Bxl4t1.js  8217 → 3326 (−59.5%)
  gzip assets/diagram-S7CK7UJ4-T-Bxl4t1.js  8217 → 3677 (−55.3%)
  br   assets/diagram-UQ7AKVKN-BnBJ_2R3.js  6079 → 2367 (−61.1%)
  gzip assets/diagram-UQ7AKVKN-BnBJ_2R3.js  6079 → 2617 (−57.0%)
  br   assets/diagram-VSXAHHWV-DLnsWaIP.js  10716 → 3782 (−64.7%)
  gzip assets/diagram-VSXAHHWV-DLnsWaIP.js  10716 → 4154 (−61.2%)
  br   assets/diagram-VX7I27RA-Duotn1-e.js  15957 → 5171 (−67.6%)
  gzip assets/diagram-VX7I27RA-Duotn1-e.js  15957 → 5707 (−64.2%)
  br   assets/diagram-Z3DM3KII-iO8-m9Hl.js  4211 → 1642 (−61.0%)
  gzip assets/diagram-Z3DM3KII-iO8-m9Hl.js  4211 → 1818 (−56.8%)
  br   assets/docxRenderer-Ilu9VYYG.js  40904 → 10539 (−74.2%)
  gzip assets/docxRenderer-Ilu9VYYG.js  40904 → 11977 (−70.7%)
  br   assets/downloadDocx-Ep6Y-mWJ.js  1130 → 612 (−45.8%)
  gzip assets/downloadDocx-Ep6Y-mWJ.js  1130 → 767 (−32.1%)
  br   assets/ebnfDiagram-PWID7BFC-CAlTqs1O.js  2005 → 784 (−60.9%)
  gzip assets/ebnfDiagram-PWID7BFC-CAlTqs1O.js  2005 → 884 (−55.9%)
  br   assets/erDiagram-RLTQ6QDP-BDpOdOIc.js  30790 → 9175 (−70.2%)
  gzip assets/erDiagram-RLTQ6QDP-BDpOdOIc.js  30790 → 10542 (−65.8%)
  br   assets/flowDiagram-HODETNUW-BGwlzTt0.js  61847 → 16384 (−73.5%)
  gzip assets/flowDiagram-HODETNUW-BGwlzTt0.js  61847 → 19717 (−68.1%)
  br   assets/ganttDiagram-EL5Y4UJY-ijkjYohu.js  69695 → 20666 (−70.3%)
  gzip assets/ganttDiagram-EL5Y4UJY-ijkjYohu.js  69695 → 23425 (−66.4%)
  br   assets/gitGraphDiagram-WWUBYQGX-3ajohVe1.js  29417 → 7789 (−73.5%)
  gzip assets/gitGraphDiagram-WWUBYQGX-3ajohVe1.js  29417 → 8695 (−70.4%)
  br   assets/index-BXtlAAcH.js  365745 → 89917 (−75.4%)
  gzip assets/index-BXtlAAcH.js  365745 → 104064 (−71.5%)
  br   assets/index-Bq7tVxen.css  6576 → 1656 (−74.8%)
  gzip assets/index-Bq7tVxen.css  6576 → 1974 (−70.0%)
  br   assets/index-cbAtdfhj.js  814041 → 187943 (−76.9%)
  gzip assets/index-cbAtdfhj.js  814041 → 236064 (−71.0%)
  br   assets/ishikawaDiagram-5VMMS53U-Cg4LfJyo.js  17469 → 6007 (−65.6%)
  gzip assets/ishikawaDiagram-5VMMS53U-Cg4LfJyo.js  17469 → 6634 (−62.0%)
  br   assets/journeyDiagram-3NMN7TZE-DXSneWK0.js  23571 → 7531 (−68.0%)
  gzip assets/journeyDiagram-3NMN7TZE-DXSneWK0.js  23571 → 8331 (−64.7%)
  br   assets/kanban-definition-UXKFOSKX-BALPAjTC.js  20291 → 6409 (−68.4%)
  gzip assets/kanban-definition-UXKFOSKX-BALPAjTC.js  20291 → 7172 (−64.7%)
  br   assets/katex-BJxdM42i.js  261435 → 63900 (−75.6%)
  gzip assets/katex-BJxdM42i.js  261435 → 77355 (−70.4%)
  br   assets/katex-BOvhZl6_.css  30249 → 6978 (−76.9%)
  gzip assets/katex-BOvhZl6_.css  30249 → 8010 (−73.5%)
  br   assets/katex-HP8lGamR.js  261313 → 63694 (−75.6%)
  gzip assets/katex-HP8lGamR.js  261313 → 77109 (−70.5%)
  br   assets/layout-BrexQxSk.js  35436 → 11571 (−67.3%)
  gzip assets/layout-BrexQxSk.js  35436 → 12751 (−64.0%)
  br   assets/linear-BDZQyaJG.js  5661 → 2117 (−62.6%)
  gzip assets/linear-BDZQyaJG.js  5661 → 2318 (−59.1%)
  br   assets/mermaid.core-C3o1vFDn.js  682569 → 140363 (−79.4%)
  gzip assets/mermaid.core-C3o1vFDn.js  682569 → 168123 (−75.4%)
  br   assets/mindmap-definition-YA3MSWOX-pgZMQ8uL.js  23175 → 6998 (−69.8%)
  gzip assets/mindmap-definition-YA3MSWOX-pgZMQ8uL.js  23175 → 7799 (−66.3%)
  br   assets/ordinal-Cboi1Yqb.js  1189 → 497 (−58.2%)
  gzip assets/ordinal-Cboi1Yqb.js  1189 → 574 (−51.7%)
  br   assets/pegDiagram-XKGWAZYB-BRKb5jpA.js  1956 → 821 (−58.0%)
  gzip assets/pegDiagram-XKGWAZYB-BRKb5jpA.js  1956 → 927 (−52.6%)
  br   assets/pieDiagram-E7YTZNPT-nrtuGrNr.js  6294 → 2407 (−61.8%)
  gzip assets/pieDiagram-E7YTZNPT-nrtuGrNr.js  6294 → 2629 (−58.2%)
  br   assets/quadrantDiagram-AXDQQJYC-DHNugfXZ.js  34010 → 8523 (−74.9%)
  gzip assets/quadrantDiagram-AXDQQJYC-DHNugfXZ.js  34010 → 9876 (−71.0%)
  br   assets/railroadDiagram-O6MQD6OU-3o3NI4V-.js  1606 → 683 (−57.5%)
  gzip assets/railroadDiagram-O6MQD6OU-3o3NI4V-.js  1606 → 758 (−52.8%)
  br   assets/renderClipboardHtml-BXoFMW-3.js  5180 → 1452 (−72.0%)
  gzip assets/renderClipboardHtml-BXoFMW-3.js  5180 → 1778 (−65.7%)
  br   assets/requirementDiagram-BXWQKSXE-2YVxwHhy.js  31000 → 8241 (−73.4%)
  gzip assets/requirementDiagram-BXWQKSXE-2YVxwHhy.js  31000 → 9661 (−68.8%)
  br   assets/sankeyDiagram-P5KCCOFB-cHAs8eQM.js  23198 → 7819 (−66.3%)
  gzip assets/sankeyDiagram-P5KCCOFB-cHAs8eQM.js  23198 → 8565 (−63.1%)
  br   assets/sequenceDiagram-WJ2MYXX4-BqEzakUS.js  116962 → 25735 (−78.0%)
  gzip assets/sequenceDiagram-WJ2MYXX4-BqEzakUS.js  116962 → 30955 (−73.5%)
  br   assets/stateDiagram-D77RDMKH-DVKtcg-v.js  10361 → 3334 (−67.8%)
  gzip assets/stateDiagram-D77RDMKH-DVKtcg-v.js  10361 → 3623 (−65.0%)
  br   assets/swimlanes-42K2YHIH-BaJ44xBg.js  115791 → 36679 (−68.3%)
  gzip assets/swimlanes-42K2YHIH-BaJ44xBg.js  115791 → 42170 (−63.6%)
  br   assets/timeline-definition-24CTP7MA-d7EudEZG.js  30829 → 9133 (−70.4%)
  gzip assets/timeline-definition-24CTP7MA-d7EudEZG.js  30829 → 10143 (−67.1%)
  br   assets/vennDiagram-4TSXK5OY-Dswu0LDK.js  41961 → 13993 (−66.7%)
  gzip assets/vennDiagram-4TSXK5OY-Dswu0LDK.js  41961 → 15643 (−62.7%)
  br   assets/wardleyDiagram-VM6X3IG4-83YMC1gW.js  25502 → 6077 (−76.2%)
  gzip assets/wardleyDiagram-VM6X3IG4-83YMC1gW.js  25502 → 6751 (−73.5%)
  br   assets/xlsx-Cig0qvpi.js  424392 → 117659 (−72.3%)
  gzip assets/xlsx-Cig0qvpi.js  424392 → 140814 (−66.8%)
  br   assets/xlsxRenderer-BQPATLFB.js  2785 → 1354 (−51.4%)
  gzip assets/xlsxRenderer-BQPATLFB.js  2785 → 1677 (−39.8%)
  br   assets/xychartDiagram-S5SC5T6Z-Do0p5dm5.js  44509 → 11058 (−75.2%)
  gzip assets/xychartDiagram-S5SC5T6Z-Do0p5dm5.js  44509 → 12534 (−71.8%)
  br   fonts/OFL.txt  4474 → 1437 (−67.9%)
  gzip fonts/OFL.txt  4474 → 1980 (−55.7%)
 precompress: wrote 66 .br and 66 .gz
 NPM_BUILD:0
@@ -0,0 +1,651 @@
 > shuimu-workbench@1.0.0 test
 > tsx --test "src/**/!(*InteractiveCard*|*chatWindowArtifact*|*deliverableSync*).test.ts" && vitest run src/components/chat/InteractiveCard.test.ts src/components/chat/chatWindowArtifact.test.ts src/components/chat/deliverableSync.test.ts
 ✔ AuthClient: 基础登录与 .env 超级管理员鉴权 (2.388929ms)
 ✔ AuthClient: 自主注册受控流程（仅填用户名密码姓名，职位强制绑定访客待指派） (0.448527ms)
 ✔ AuthClient: 开启注册需审批时，注册账号处于待审状态并禁止登录 (0.495266ms)
 ✔ AuthClient: 管理员创建账号必须在首次登录修改密码 (0.258262ms)
 ✔ authErrorMessage: 优先信封 error 领域原因，避免裸 JSON / 无权访问误导 (0.227843ms)
 ✔ mapUserAccount: snake_case Go contract → camelCase (1.231233ms)
 ✔ mapUserAccount: prefers snake_case when both present (0.164515ms)
 ✔ mapLoginResult/mapMeResult use top-level must_change_password as sole authority (0.211534ms)
 ✔ mapUserAccountList maps admin users list (0.116027ms)
 ✔ AuthClient.login + fetchMe map real snake_case wire payloads (36.529405ms)
 ✔ shouldAutoRetryChat: disabled until idempotency (0.913842ms)
 ✔ shouldAutoRetryTask: transient only, max 2 (0.229563ms)
 ✔ shouldAutoRetryTask: document busy never retries (0.136656ms)
 ✔ shouldAutoRetry alias delegates to task path (0.137816ms)
 ✔ autoRetryDelayMs: 1s then 2s (0.137336ms)
 ✔ fallbackErrorClass: network → transient, workspace → resumable (1.663131ms)
 ✔ upsertChatSession surfaces safe INTERNAL envelope instead of raw JSON text (79.905537ms)
 [sse] 忽略无法解析的帧: {not valid json SyntaxError: Expected property name or '}' in JSON at position 1 (line 1 column 2)
    at JSON.parse (<anonymous>)
    at parseFrame (/home/weichen/projects/Agent-System/.worktrees/refresh-task-state-recovery-2026-09-28/web/src/api/chatStream.ts:126:26)
    at parseSSEChunk (/home/weichen/projects/Agent-System/.worktrees/refresh-task-state-recovery-2026-09-28/web/src/api/chatStream.ts:199:20)
    at TestContext.<anonymous> (/home/weichen/projects/Agent-System/.worktrees/refresh-task-state-recovery-2026-09-28/web/src/api/chatStream.test.ts:41:31)
    at Test.runInAsyncScope (node:async_hooks:226:14)
    at Test.run (node:internal/test_runner/test:1402:25)
    at Test.processPendingSubtests (node:internal/test_runner/test:974:18)
    at Test.postRun (node:internal/test_runner/test:1542:19)
    at Test.run (node:internal/test_runner/test:1467:12)
    at async Test.processPendingSubtests (node:internal/test_runner/test:974:7)
 ✔ parseSSEChunk: parses event and data lines (1.00906ms)
 ✔ parseSSEChunk: keeps incomplete frame in rest (0.231474ms)
 ✔ parseSSEChunk: flush parses final frame without trailing delimiter (0.224793ms)
 ✔ parseSSEChunk: ignores heartbeat comments and keeps last id (0.139066ms)
 ✔ parseSSEChunk: skips a single malformed frame and keeps parsing the next valid frame (5.017202ms)
 ✔ parseSSEChunk: parses CRLF-delimited frames (0.229503ms)
 ✔ parseSSEChunk: mixed LF and CRLF delimiters (0.137336ms)
 ✔ API_BASE is empty relative base (1.037719ms)
 ✔ apiFetch attaches Bearer from session storage (64.929112ms)
 ✔ apiFetch retries GET on 503 then succeeds (201.72881ms)
 ✔ apiFetch does not retry POST by default (0.516094ms)
 ✔ readApiError preserves typed envelope for error cards (1.498965ms)
 ✔ apiFetch auth:false skips Authorization (0.530665ms)
 ▶ docgov lifecycle API client
  ✔ claimDocument posts to /api/v1/docgov/claim with author role (23.291168ms)
  ✔ submitForReview posts to /api/v1/docgov/submit-review (0.706349ms)
  ✔ reviewDecision posts action and optional reason (0.817416ms)
  ✔ approverDecision posts action and rollback_to (0.954792ms)
  ✔ proposeModification posts to /api/v1/docgov/proposal (1.198325ms)
  ✔ decideModificationProposal posts to /api/v1/docgov/proposal/decide (0.720289ms)
 ✔ docgov lifecycle API client (28.742506ms)
 ✔ updateFreeformDoc exposes CONCURRENCY_CONFLICT on 409 (22.916699ms)
 ▶ notifications inbox helpers
  ✔ ranks action_required / proposal ahead of info (0.65737ms)
 ✔ notifications inbox helpers (1.486036ms)
 ✔ addProjectMember exposes structured ProjectSettingsApiError on 403 (25.801144ms)
 ✔ updateProjectMember surfaces safe INTERNAL message on 500 duty-sync failure (0.822326ms)
 ▶ projectSettingsError
  ✔ keeps status, code, message and issues on PLAN_FROZEN (1.624582ms)
  ✔ maps revision and concurrent conflicts to distinct recovery (0.172775ms)
  ✔ joins validation issues but still keeps typed codes (0.189844ms)
  ✔ prefers envelope code over legacy error field (0.138986ms)
  ✔ unwraps nested business JSON inside VERSION_CONFLICT envelope (0.187354ms)
  ✔ maps bare VERSION_CONFLICT envelope to retry_after_switch (0.106467ms)
  ✔ retains the local draft when a late tab hits PLAN_FROZEN (0.114897ms)
  ✔ prefers error_code over legacy bare error string (DOC_BUSY) (0.128807ms)
 ✔ projectSettingsError (4.087899ms)
 ✔ fetchProjectTrace turns null arrays into empty arrays (25.555001ms)
 ✔ fetchProjectTrace keeps real arrays and rejects failed responses (1.302701ms)
 ✔ Project workflow types validation (2.125657ms)
 ✔ createTaskRun exposes DOC_BUSY detail without a retryable run (27.347498ms)
 ✔ parseSSEChunk tracks id seq for after_seq reconnect (0.474346ms)
 ✔ subscribeTaskRunEvents reconnects on EOF with advancing after_seq (6.424009ms)
 ✔ readTaskRunSSEOnce marks non-terminal EOF for reconnect (0.68087ms)
 ✔ subscribeTaskRunEvents fails fast on 401 without retrying (0.491175ms)
 ✔ subscribeTaskRunEvents fails fast on 404 without retrying (1.400458ms)
 ✔ buildTaskRequest: maps five project fields + task_key + user identity (0.757228ms)
 ✔ buildTaskRequest: omits optional fields when absent (0.164035ms)
 ✔ buildTaskRequest: includes materials when provided and non-empty (0.755688ms)
 ✔ buildResumeRequest: adds work_dir on top of task request (0.175045ms)
 ✔ buildTaskRequest: canonicalizes legacy doc-p doc_id (0.260083ms)
 ✔ buildTaskRequest: UI-05 pinned intake keeps task_key + doc_proj6109_intake (0.193434ms)
 ✔ buildResumeRequest: forwards add_steps when provided (0.153185ms)
 ✔ CategoryDock: 渲染顶部 🔔 未读提醒项及角标数字 (20.272788ms)
 ✔ CategoryDock: isAllReadCompleted 为 true 时渲染完成态提示 (9.797689ms)
 ✔ CategoryDock: 0 未读时不渲染红色未读角标 (6.548825ms)
 ✔ CategoryDock: 分类项包含依次跳转与遍历提示信息 (5.285423ms)
 ✔ chat column lives outside the workbench shell (0.954951ms)
 ✔ InputArea keeps draft when send preflight fails (1.420548ms)
 ✔ InputArea clears draft only after accepted send (2.463867ms)
 ✔ InputArea shows thinking block message (0.197614ms)
 ✔ InputArea keeps draft when attachment registration fails (0.247822ms)
 ✔ TaskRunCard renders timeline with title badge and current step (13.771851ms)
 ✔ TaskRunCard completed state keeps card with duration and artifact (2.308901ms)
 ✔ TaskRunCard sanitizes frozen progress labels on completed steps (1.979611ms)
 ✔ TaskRunCard keeps shell when only taskRunId is present (0.864224ms)
 ✔ active document pane lives outside the workbench shell (1.437047ms)
 ✔ isTemplateOnly 交付物应正确识别为初始模板展示态 (7.921545ms)
 ✔ workbench shell lives outside App (4.726439ms)
 ✔ 左侧栏鼠标严密绝对跟随与夹紧测试 (Left Rail Exact Pointer Following & Clamping) (0.713178ms)
 ✔ 右侧栏鼠标严密绝对跟随与反向位移测试 (Right Rail Exact Pointer Following & Inverted Delta) (0.126506ms)
 ✔ trace panel is not mounted above the chat stream (5.641603ms)
 ✔ detects legacy mock and neuropulse/ten-projects ids (0.864744ms)
 ✔ detects colliding demo project ids (0.132876ms)
 ✔ applyRestoredFileRaw: 以 prev 为权威，只补丁缺失 fileRaw，不丢新增条目 (1.154566ms)
 ✔ applyRestoredFileRaw: 无可用 fileRaw 时返回同一引用 (0.187955ms)
 ✔ login member sync stays one request for 1, 30 and 100 projects (6.07364ms)
 ✔ viewer seed keeps session visibility without a member-list call (0.397488ms)
 ✔ later session response cannot overwrite a newer user (0.190665ms)
 ✔ in-flight member sync collapses to one call (0.263752ms)
 ✔ 每人独立私聊：同一项目不同用户拥有不同 sessionId (0.881204ms)
 ✔ 左侧会话列表按当前用户过滤，不可见他人私聊 (0.256262ms)
 ✔ 消息严格归属各自私聊 session，无跨用户混聊 (0.870234ms)
 ✔ 切换用户时 resolve 到该用户同项目私聊会话 (0.244812ms)
 ✔ resolve 优先有消息的私聊，跳过空壳 sess-p{N}-u{id} (1.166125ms)
 ✔ ensurePrivateSessionsForUser 为缺失用户补齐空私聊 (0.593802ms)
 ✔ 多项目会话消息隔离测试仍成立（endo vs ultrasound） (0.33904ms)
 ✔ window cache LRU: 驱逐带 userId 前缀的完整 cache key，不留孤儿条目 (1.36061ms)
 ✔ window cache: 无 schema 的旧载荷不当作有效缓存 (7.710341ms)
 ✔ window cache: 读写同一 userId 前缀，跨用户不串读 (0.273892ms)
 ✔ 残留的 legacy demo sessionId 被读取侧清除并返回 fallback (0.905753ms)
 ✔ 真实 sessionId 不被误清 (0.152565ms)
 ✔ 无 window 环境（SSR）安全返回 fallback (0.068078ms)
 ✔ same account and first login do not reset workbench storage (1.144056ms)
 ✔ switching accounts resets workbench storage (0.182865ms)
 ✔ resetWorkbenchStorage preserves org user directory for cross-account project create (1.219384ms)
 ✔ parseAnchorHash: 正确解析标准深链 hash (1.438807ms)
 ✔ formatAnchorHash: 正确构造深链格式 (0.106107ms)
 ✔ navigateToAnchor: 快路径直接命中 DOM 节点 (0.388478ms)
 ✔ navigateToAnchor: 慢路径拉取分片后命中 (79.542228ms)
 ✔ resolveTurnAnchorId: 方案 A 轮次归一化测试 (0.507675ms)
 ✔ 全链路场景 1: 用户自主受控注册（仅填用户名/密码/姓名，职位严格绑定访客待指派） (1.552684ms)
 ✔ 全链路场景 2: 系统准入策略（关闭自主注册与开启强制审批流转） (0.647091ms)
 ✔ 全链路场景 3: 管理员开通账号首次强制改密闭环 (0.242653ms)
 ✔ 全链路场景 4: 环境标识随登录会话返回（dev/test/main） (0.261022ms)
 ✔ cleanHeadingContent: 手工多级序号前缀清洗与专业词汇保护 (2.277532ms)
 ✔ 标题多层级自动编号：单 H1 文档总标题自适应模式 (1 -> 1.1 -> 1.1.1) (0.34501ms)
 ✔ 标题多层级自动编号：多 H1 标准大章模式 (0.137276ms)
 ✔ 跨章节图、表、公式多级按章重编号测试 (表 1-1, 表 2-1, 图 3-1, 式 (4-1)) (0.154156ms)
 ✔ 有序列表与无序列表编号与渲染功能测试 (2.399788ms)
 ✔ 动态块操作闭环重编号验证 (Insert, Move, Delete) (3.589093ms)
 ✔ findMatchingBlockForReview: 能够精准按审查关键词（电气安全/EMC/竞品）定位段落块 (0.848975ms)
 ✔ findMatchingBlockForReview: 能够按外部指定 hint 线索精确命中目标块 (0.170635ms)
 ✔ createRejectConfirmationCard: 生成合规危险级驳回门禁卡片 (0.605742ms)
 ✔ isAffirmativeResponse: 准确识别双轨交互中的肯定意图输入 (0.205574ms)
 ✔ findLastActiveCard: 正确在会话历史中提取最新的待决策卡片 (0.201834ms)
 ✔ decrementActiveCardsTtl: 多轮对话超时递减与自动失效归档 (0.163195ms)
 ✔ applyConfirmRejectToDoc & applyUndoRejectToDoc: 驳回挂载批注与60s窗口期撤销完整闭环 (0.251372ms)
 ✔ resolveCardInMessages & cancelCardInMessages: 消息流中卡片终态沉淀 (0.151626ms)
 ✔ classifyRejectUtterance: 询问原因 → QUERY_REJECT_REASON (1.295522ms)
 ✔ classifyRejectUtterance: 明确执行 → EXECUTE_REJECT (0.177635ms)
 ✔ classifyRejectUtterance: 模糊 → OTHER (0.197495ms)
 ✔ resolveDocGovCard: 编制人查询 → 信息卡且无 confirm_reject (0.788967ms)
 ✔ resolveDocGovCard: 审核人执行 + reviewing → 确认卡 (0.289862ms)
 ✔ resolveDocGovCard: 编制人误说执行驳回 → 无权纯文本 (0.141016ms)
 ✔ resolveDocGovCard: 批准人 + approving → 终审确认卡 (0.274622ms)
 ✔ resolveDocGovCard: 编制人 OTHER → 引导查询无危险按钮 (0.128557ms)
 ✔ canConfirmReject: 仅审核/批准 (0.537784ms)
 ✔ deriveFallbackCategory identifies categories correctly (0.870374ms)
 ✔ enrichMessagesWithCategories backfills categories without overwriting existing ones (0.313601ms)
 ✔ Markdown ➔ Block AST 解析测试 (标题自适应编号与图表公式全链路) (4.275193ms)
 ✔ 动态重编号引擎测试 (多 H1 规范章节重排验证) (0.980431ms)
 ✔ Block AST ➔ Markdown 双向无损往返还原测试 (0.843655ms)
 ✔ 三层交付物多用户隐私绝对隔离测试 (Multi-user Strict Privacy Isolation) (1.088098ms)
 ✔ 一键共享至项目参考库流转测试 (Promote to Shared Reference Test) (0.221383ms)
 ✔ 撤回共享恢复私有草稿测试 (Revoke Shared Reference Test) (0.408948ms)
 ✔ 受控工作流纯净性红线检验 (Workflow Purity Security Red Line) (0.129456ms)
 ✔ isSameMessage: clientTurnId match (1.861744ms)
 ✔ isSameMessage: serverId match without turn (0.175055ms)
 ✔ filterPendingLiveMessages: real temp vs server id (0.137566ms)
 ✔ mergeLiveOverHistory preserves server id from live (0.173595ms)
 ✔ mergeLiveOverHistory does not let inflight live mask terminal history (0.134716ms)
 ✔ mergeLiveOverHistory keeps live terminal over history inflight (0.112497ms)
 ✔ applyPhaseEvent: appends first phase as running node (1.403038ms)
 ✔ applyPhaseEvent: completes predecessors when a new phase arrives (0.249813ms)
 ✔ applyPhaseEvent: same phase refreshes label in place (no duplicate node) (0.243692ms)
 ✔ applyPhaseEvent: unknown phase falls back to hourglass meta (0.130056ms)
 ✔ applyPhaseEvent: prefers backend message over short label (0.114206ms)
 ✔ applyPhaseEvent: step_completed closes running entry without appending (0.239473ms)
 ✔ applyPhaseEvent: stage_completed closes entries (0.163325ms)
 ✔ applyPhaseEvent: task_failed appends warning node (0.133246ms)
 ✔ applyPhaseEvent: decorates label with step index / total steps (0.137436ms)
 ✔ applyPhaseEvent: repeated step_executing with step_index gets separate nodes (0.200365ms)
 ✔ applyPhaseEvent: ignores event without phase payload (0.097807ms)
 ✔ applyPhaseEvent: step_failed marks running entry as failed (keeps step label) (0.114317ms)
 ✔ applyPhaseEvent: step_failed appends failed node when no running entry exists (1.234163ms)
 ✔ applyPhaseEvent: retry of a failed step resets it to running (0.141366ms)
 ✔ applyPhaseEvent: stores detail from phase.message (0.075458ms)
 ✔ PHASE_META: document_parsing and document_summarizing are registered (0.114747ms)
 ✔ PLAN_CONFIRM 卡片来自 PlanView 且不触达 AST (1.988191ms)
 ✔ 409 冲突把卡片标为 outdated 并换上最新 revision (0.252702ms)
 ✔ GET reconcile 用权威 PlanView 覆盖本地 executing，不假装成功 (0.226923ms)
 ✔ 过期计划映射 expired (0.125736ms)
 ✔ 非计划卡不被误判 (0.112107ms)
 ✔ 集成验证：项目切换时交付物严格隔离，新建项目即刻包含 7 份模板副本 (6.355481ms)
 ✔ 集成验证：三层隐私与项目物理隔离双重约束 (8.259005ms)
 ✔ 用例 1: DEFAULT_ROLES 包含 5 种默认角色且具备预期权限 (4.314872ms)
 ✔ 用例 2: 系统管理员 (ADMIN) 与具备 perm:manage 用户的通配豁免特权 (0.226003ms)
 ✔ 用例 3: 常规工作流基线可见性 (workflow:view_all) 判定 (0.249052ms)
 ✔ 用例 4: 单文档覆盖 - designated_only (仅编制/审核/批准人可见，其他用户被精确拦截) (0.145636ms)
 ✔ 用例 5: 单文档覆盖 - role_whitelist (仅白名单内角色可见，其他角色被精确拦截) (0.127306ms)
 ✔ 用例 6: 单文档覆盖 - public (放宽可见性，无 workflow:view_all 权限的访客也能查阅) (0.100067ms)
 ✔ 用例 7: 私密文档隔离强红线 (category === private 无论任何管理员或单文档覆盖，非创建者绝对不可见) (0.111796ms)
 ✔ 用例 8: 核心业务动作权限拦截判定 (canPerformDocAction & hasPermission) (0.149616ms)
 ✔ 用例 9: filterVisibleDeliverablesWithRBAC 能够批处理过滤大盘交付物 (5.913474ms)
 ✔ 集成场景 1: 动态新增合规审计员角色，并配置其仅能查阅特定单文档覆盖 (1.047099ms)
 ✔ 集成场景 2: 服务端 capabilities 变更后权限即刻生效 (0.168345ms)
 ✔ 集成场景 3: 单文档覆盖 designated_only 严格收紧与撤回恢复全流程 (0.148375ms)
 ✔ renderInlineMarkdown: renders bold, italic, strikethrough (1.362779ms)
 ✔ renderInlineMarkdown: escapes raw HTML tags to prevent XSS (0.653981ms)
 ✔ renderInlineMarkdown: blocks javascript: link hrefs (0.305061ms)
 ✔ renderInlineMarkdown: renders inline code and preserved line breaks (0.290592ms)
 ✔ renderInlineMarkdown: renders headings lists and tables for chat replies (0.487936ms)
 ✔ renderInlineMarkdown loads katex only after a formula is seen (19.51147ms)
 ✔ sanitizeMermaidCode: auto wraps unquoted labels containing brackets or comparison operators (0.422077ms)
 (node:1877659) ExperimentalWarning: localStorage is not available because --localstorage-file was not provided.
 (Use `node --trace-warnings ...` to show where the warning was created)
 ✔ DOCX 渲染器行内文本与公式解析测试 (parseInlineContent) (4.879835ms)
 ✔ DOCX 渲染器 LaTeX 原生 OMML 转换引擎测试 (latexToOmml) (2.009401ms)
 ✔ DOCX 渲染器 Mermaid 语法净化与容错测试 (sanitizeMermaidCode) (0.3103ms)
 ✔ DOCX 渲染器 SVG 净化防 Canvas 污染测试 (cleanSvgForCanvas) (0.367549ms)
 ✔ DOCX 渲染器全量编译测试 (buildDocxModel & exportDocumentToDocx) (61.249852ms)
 ✔ XLSX 渲染器多表提取与工作簿编译测试 (buildXlsxWorkbook & exportDocumentToXlsx) (6.587554ms)
 ▶ Reply Document IR & Export Engine
  ✔ deriveReplyTitle and sanitizeFilename derive title and safe file name (1.768507ms)
  ✔ deriveReplyTitle handles text without heading (0.285461ms)
  ✔ parseReplyMarkdown parses headers, lists, code, and tables (1.728138ms)
  ✔ renderClipboardHtml renders self-contained styling for office tools (0.462936ms)
  ✔ renderClipboardHtml converts LaTeX formulas to native MathML for Word/Office (4.016571ms)
  ✔ renderPlainText strips formatting cleanly (0.481286ms)
  ✔ serializeMarkdown round-trips to valid markdown (0.33855ms)
  ✔ stripUnsafeHtml strips script tags and dangerous event handlers (0.549014ms)
  ✔ escapeHtml escapes math error payloads used by BlockEditorModal catch path (0.196264ms)
 ✔ Reply Document IR & Export Engine (11.579927ms)
 ✔ createProjectTemplateDeliverables: 必须精确生成 7 份全生命周期受控文档副本 (3.852105ms)
 ✔ resolveRequirementsTaskKey prefers intake for 初始需求表 (1.199405ms)
 ✔ resolveRequirementsTaskKey selects user_requirements for explicit URS (0.156685ms)
 ✔ UI-02 first-round prompt with 首轮+用户需求+摸底 must be intake not user_requirements (0.478466ms)
 ✔ 首轮 / 需求摸底 / 摸底 beat bare 用户需求 (0.106416ms)
 ✔ bare 用户需求 without V0.1 markers stays user_requirements (0.107757ms)
 ✔ matchTemplateDeliverable uses intake key when pool only has URS template (0.258242ms)
 ✔ resolveTaskTemplateForMessage keeps intake key with URS-only pool (0.149816ms)
 ✔ matchTemplateDeliverable prefers intake when both templates exist (0.126717ms)
 ✔ read-only request does not start a document task: 总结一下用户初始需求表 (0.294921ms)
 ✔ read-only request does not start a document task: 查看用户初始需求表 (0.145125ms)
 ✔ read-only request does not start a document task: 用户初始需求表现在谁在编辑 (0.090967ms)
 ✔ read-only request does not start a document task: 分析一下用户需求表里的风险 (0.070788ms)
 ✔ read-only request does not start a document task: 用户初始需求表 (0.054098ms)
 ✔ explicit write action still starts intake (0.136246ms)
 ✔ explicit update action still starts URS (0.190135ms)
 ✔ UI-05 prompt without active doc still keyword-routes to feasibility_reg (0.294701ms)
 ✔ UI-05 prompt + registered intake pins intake and doc_proj6109_intake (0.235983ms)
 ✔ read-only current-doc request does not start a task (0.071978ms)
 ✔ summary/analysis with write verbs stays chat (read-only priority) (0.089587ms)
 ✔ pure write intent still routes to task (0.053668ms)
 ✔ template-only active slot is not pinned; keywords still apply (0.096717ms)
 ✔ explicit named deliverable can switch away from active intake (0.165225ms)
 ✔ domain 注册法规 without current-doc deixis still routes feasibility_reg (0.088787ms)
 ✔ mismatched title vs docId suffix prefers docId taskKey (0.093428ms)
 ✔ deixis 当前文档 pins registered intake (0.095807ms)
 ✔ deixis 本文件 pins registered intake (0.047478ms)
 ✔ deixis 这份文档 pins registered intake (0.050108ms)
 ✔ deixis 本文档 pins registered intake (0.044218ms)
 ✔ deixis 本受控文档 pins registered intake (0.041788ms)
 ✔ no active doc does not invent a pin (0.096247ms)
 ✔ UI-11 named 系统设计输入 beats negated 需求 background (0.076237ms)
 ✔ mixed generate-design + do-not-cover-requirements stays prelim_design (0.055108ms)
 ✔ current-doc pin still wins when no other deliverable is named (0.083417ms)
 ✔ v3 bd_ task key wins over unprefixed intake alias (0.124346ms)
 ✔ word-order variants of the intake sheet stay on intake, not the shorter URS title (0.195294ms)
 ✔ resolveTaskTemplateForMessage blocks cross-workflow document mentions (2.185825ms)
 ✔ resolveTaskTemplateForMessage keeps in-scope main docs when active is main (0.129286ms)
 ✔ weighted mode does not let a cross-workflow title hijack an in-scope write (0.537474ms)
 ✔ progress question is read-only; explicit start still resolves research (0.093728ms)
 ▶ document revision overlay modes
  ✔ pure hides all decorations (0.811806ms)
  ✔ manual only marks human hunks (0.160956ms)
  ✔ all marks human and agent differently (0.137306ms)
  ✔ same human hunk stays amber in manual and all (no pairwise style guarantee) (0.135026ms)
  ✔ exposes needs-content error code for empty orphan docs (0.094397ms)
  ✔ matches overlay hunks by section or text when ids differ (0.272582ms)
  ✔ aligns client block ids from head_block_map by order (0.217134ms)
 ✔ document revision overlay modes (2.925713ms)
 ▶ toolbar refresh + report entry reachability
  ✔ pickLatestChangeReportId prefers highest version with report (0.33787ms)
  ✔ first squad run alone still exposes a reachable report id (0.165395ms)
  ✔ second squad run keeps both history report entries and latest shortcut (0.778567ms)
  ✔ revisionRefreshKey is an independent refresh signal from docId (0.127386ms)
 ✔ toolbar refresh + report entry reachability (1.69243ms)
 ▶ change report view model
  ✔ surfaces risk items, project impact, doc impacts with advice, and open questions (0.67743ms)
  ✔ marks complete analysis as PI narrative vs factual fields (0.187924ms)
  ✔ preferReportId prioritizes pending over committed latest (0.096757ms)
 ✔ change report view model (1.147986ms)
 ▶ revisionToolbarMount
  ✔ exposes clear non-error template preview hint copy (2.417999ms)
  ✔ template doc-p + isTemplateOnly=true does not mount (0.162285ms)
  ✔ legacy doc-001 does not mount (0.156295ms)
  ✔ unregistered doc-upload orphan does not mount (must register first) (0.145726ms)
  ✔ server-registered upload (doc-free) and freeform mount (0.156935ms)
  ✔ canonical doc_proj and generated non-template doc-p mount (0.193634ms)
 ✔ revisionToolbarMount (4.318292ms)
 ✔ login workspace sync lives in one hook (1.204984ms)
 ✔ block editing lives in one hook (0.957632ms)
 ✔ interactive card actions live in one hook (1.048618ms)
 ✔ useChatWindow: 所有 saveSessionWindowCache 调用必须传入 userId (1.066258ms)
 ✔ loadOlder race: stale resolve ignored by generation (1.557084ms)
 ✔ window fetch race: stale response ignored by generation (4.517886ms)
 ✔ mapRecordToMessage: 正确映射服务端消息记录为前端 ChatMessage (1.654151ms)
 ✔ refresh window: 空 messages 日桶不算有效窗口 (0.192455ms)
 ✔ mapRawBucket: 正确映射服务端按日分片桶 (0.312121ms)
 ✔ Category navigation: chip 选中后准确匹配目标分类最新轮次并触发归一化定位 (0.286702ms)
 ✔ mapRecordToMessage: restores task phase snapshot into card fields (0.236783ms)
 ✔ mapRecordToMessage: run anchor without task_status stays streaming (0.180904ms)
 ✔ controlled document hydration lives in one hook (1.096208ms)
 ✔ project creation lives in one hook called before the login return (0.755868ms)
 ✔ document lifecycle actions live in one hook (1.109877ms)
 ✔ freeform document actions live in one hook (1.755868ms)
 ✔ in-flight states cover pending and running only (6.667272ms)
 ✔ hasInFlightParse detects parse polling needs (0.246373ms)
 ✔ hook + reparse API surface wired (0.298551ms)
 ✔ project member writes live in one hook before the login return (1.318591ms)
 ✔ replay retries 502 and stops on auth or not-found (0.66772ms)
 ✔ sending a message lives in one hook (1.01656ms)
 ✔ server document list sync lives in one hook (1.393729ms)
 ✔ chat and task attempts live in one hook (1.34513ms)
 ✔ stage unresolved and not-in-active-stage degrade to chat once (0.155916ms)
 ✔ failed run with error_message surfaces TASK_FAILED not NETWORK_ERROR (0.770138ms)
 ✔ snake_case blocked_pending_governance is governance, not a network error (0.34381ms)
 ✔ unknown 4xx business code is not reported as a network error (0.377559ms)
 ✔ turn runtime: session A stream survives session B switch (0.649891ms)
 ✔ abortAllSessions aborts controllers across all sessions (0.192954ms)
 ✔ App persists workbench state through one hook and migrates legacy messages once (0.968861ms)
 ✔ active task cache keeps an in-progress run and drops it when finished (1.106048ms)
 ✔ syncAdminUserDirectories calls local reload then parent onRefreshUsers (1.624912ms)
 ✔ syncAdminUserDirectories fails closed when onRefreshUsers is a dead/missing prop (0.614612ms)
 ✔ syncAdminUserDirectories surfaces parent refresh errors (no silent swallow) (2.138076ms)
 ✔ afterProjectMemberPersisted syncs members then refreshes workflow docs (1.172795ms)
 ✔ afterProjectMemberPersisted does not refresh when member sync fails (0.463506ms)
 ▶ blockEditorDialogPolicy
  ✔ pins an accessible dialog label and exact save name (0.492506ms)
  ✔ rejects page-wide fuzzy /保存/ that would hit private-library buttons (0.258003ms)
 ✔ blockEditorDialogPolicy (1.557624ms)
 ▶ resolveSaveTargetBlock
  ✔ keeps the editor and does not fake success when the block id is gone (0.635451ms)
  ✔ never retargets another block that happens to share the same content (0.705099ms)
 ✔ resolveSaveTargetBlock (2.244724ms)
 ▶ editor session document binding
  ✔ fail-closes when activeDocId drifts away from the opened editor doc (0.173244ms)
  ✔ accepts a save only when the editor is still bound to the active doc (0.099087ms)
  ✔ blocks switching documents while the editor is open (0.139706ms)
 ✔ editor session document binding (0.614882ms)
 ▶ interpretBlockSaveError
  ✔ 409 keeps the draft in the editor (0.170955ms)
  ✔ other PATCH failures also keep the editor (0.192604ms)
 ✔ interpretBlockSaveError (0.782167ms)
 ▶ blockedGovernanceCopy
  ✔ recognizes blocked run status and fixed hint, not generation success (0.736178ms)
 ✔ blockedGovernanceCopy (1.654011ms)
 ▶ resolveChatComposerStreaming
  ✔ is false when no inflight assistant bubbles (0.472606ms)
  ✔ is true when inflight bubble and live turn (0.106117ms)
  ✔ is true when inflight bubble and live task run (0.078888ms)
  ✔ is false for orphan streaming_status after runtime cleared (0.059708ms)
 ✔ resolveChatComposerStreaming (1.616372ms)
 ▶ commentPolicy
  ✔ applies stage matrix for add (0.6784ms)
  ✔ keeps viewer read-only (0.206644ms)
  ✔ resolves author role literally (0.119417ms)
  ✔ gates resolve by stage (0.206814ms)
 ✔ commentPolicy (4.325182ms)
 ▶ controlledContentInvalidation
  ✔ hydrate success then late list_sync keeps cache/ready and does not bump resume (1.722119ms)
  ✔ post_task invalidation drops active cache and requires a new hydrate epoch (0.221513ms)
  ✔ editor open / save epoch keeps draft and rejects a stale GET after PATCH (0.351609ms)
  ✔ never drops the active doc without signaling rehydrate (0.138126ms)
  ✔ candidate helper keeps template / doc-001 / freeform / upload out of wipe lists (0.450837ms)
  ✔ dropRecordKeys is a no-op when ids are absent (0.219383ms)
 ✔ controlledContentInvalidation (4.35165ms)
 ▶ controlledDocHydrate
  ✔ shouldLoadControlledContent when non-template workflow doc (3.012441ms)
  ✔ shouldLoadControlledContent rejects local legacy/mock ids like doc-001 (0.181415ms)
  ✔ shouldLoadControlledContent keeps canonical, freeform-off, and template preview rules (0.164175ms)
  ✔ hydrateControlledDocumentContent never requests content/revision for doc-001 (0.383468ms)
  ✔ hydrateControlledDocumentContent never requests for template preview or registered freeform (0.181555ms)
  ✔ shouldLoadControlledContent stays inside toolbar server-backed eligibility (0.313571ms)
  ✔ resolveCanonicalDocId maps doc-p56-intake to doc_proj56_intake (0.490665ms)
  ✔ canonicalDocIdForTask maps doc-p56-research to doc_proj56_research (0.230063ms)
  ✔ hydrate treats projection missing as ledger content, not 404 (1.823986ms)
  ✔ attaches hydrate ETag as PATCH baseline and ignores a later GET (1.028349ms)
  ✔ rejects hydrate when revision-state network or 5xx fails (0.841565ms)
  ✔ aligns markdownToBlockAST random ids onto revision-state head_block_map (1.830716ms)
  ✔ leaves Math.random ids when head_block_map is omitted (why gate + fail-closed remain) (1.067668ms)
  ✔ rejects controlled hydrate on NOT_FOUND or NEEDS_CONTENT (0.377559ms)
  ✔ fingerprint stays stable when only object identity changes (0.269112ms)
  ✔ dedupes in-flight hydrate for the same doc fingerprint (0.449247ms)
  ✔ aligned late hydrate still cannot overwrite a PATCH-success baseline (0.455086ms)
 ✔ controlledDocHydrate (15.550208ms)
 ✔ createClientTurnId uses randomUUID when available (0.815456ms)
 ✔ createClientTurnId falls back to getRandomValues UUID v4 (0.421077ms)
 ✔ createClientTurnId throws when no secure random API exists (0.35659ms)
 ✔ selectNonCreatorInitialMembers skips creator only (1.587573ms)
 ✔ buildManagedMemberUpsertPayload sends document_duty on add/update/restore (0.270002ms)
 ✔ buildInitialMemberUpsertPayload maps reviewer UI role to member API role (0.191914ms)
 ✔ persistInitialProjectMembers awaits each add and stops on failure (0.65355ms)
 ✔ ensureCreatedProjectWithMembers skips recreate on bootstrap id and retries members (0.549973ms)
 ✔ beginCreateProjectSubmit rejects concurrent second entry until end (31.813816ms)
 ✔ resolveCreateProjectSubmitPhase uses bootstrap to choose syncing label (0.723658ms)
 ✔ createProjectSubmitButtonLabel and close block while submitting (0.200754ms)
 ✔ shouldCloneMedicalTemplates skips the 7-pack for software domain (0.131006ms)
 ▶ resolveCurrentDocRole
  ✔ resolves A as author by authoritative user id (0.752448ms)
  ✔ does not promote reviewer B or stranger to author (0.123077ms)
  ✔ resolves real USER_B id as reviewer and never name-promotes placeholder 2 (0.095468ms)
  ✔ fail-closes unknown identity to viewer even if projectPermission says author (0.076527ms)
  ✔ matches name only for legacy local docs without duty ids (0.105767ms)
  ✔ does not name-promote when server controlled duties are zero ids (0.160535ms)
 ✔ resolveCurrentDocRole (3.184245ms)
 ▶ currentProjectRestore
  ✔ rejects empty project names (0.827155ms)
  ✔ rejects page-wide project name when banner and sidebar both match (0.185724ms)
  ✔ scopes the visible entry to complementary exact title (2.776877ms)
  ✔ scopes the current project assertion to banner exact text (0.35035ms)
 ✔ currentProjectRestore (5.071659ms)
 ▶ deliverableLeaseSync
  ✔ returns the same array reference when leases are unchanged (0.753007ms)
  ✔ returns a new array only when a lease actually changes (0.177275ms)
 ✔ deliverableLeaseSync (1.904263ms)
 ▶ docGovActionVisibility
  ✔ hides all approval actions on unmanaged docs (1.605673ms)
  ✔ gates separated duties by user id, not folded role label (0.172974ms)
  ✔ author_only: same user holds all three duties and sees all actions (0.106427ms)
  ✔ author_only: non-duty user sees no actions (0.104037ms)
 ✔ docGovActionVisibility (3.05592ms)
 ▶ documentBlockDomAttrs
  ✔ marks a real DocumentBlock in pure mode without overlay kind (1.241673ms)
  ✔ keeps the same stable attrs when overlay is later attached separately (0.197974ms)
 ✔ documentBlockDomAttrs (2.573574ms)
 ✔ author can edit drafting when unlocked (0.720368ms)
 ✔ active project member can edit drafting when unlocked (deny_only_submit) (0.129316ms)
 ✔ non-member viewer cannot edit drafting (0.089708ms)
 ✔ lease blocks edit even for author and member (0.093477ms)
 ✔ reviewing is not editable via drafting affordance (0.080617ms)
 ▶ durableBodySnippet
  ✔ rejects empty persisted snippets (0.850954ms)
  ✔ lets a later successful revision replace the old snippet (0.988341ms)
  ✔ does not promote a conflict-failed draft to the expected body (0.174115ms)
 ✔ durableBodySnippet (2.940223ms)
 ▶ dutyCellAssertion
  ✔ parses author, reviewer, and viewer from select value or badge role (0.890893ms)
  ✔ must not infer author from a row whose option labels include 编制负责人 (0.499835ms)
  ✔ accepts only a disabled non-assignable reviewer select (0.122916ms)
  ✔ rejects an enabled reviewer select when the reviewer cannot assign leads (0.098187ms)
  ✔ rejects canAssignLead=true even if the reviewer select is disabled (0.249163ms)
  ✔ rejects viewer, author, approver, and unknown values for USER_B (0.265472ms)
 ✔ dutyCellAssertion (6.592245ms)
 ▶ decideHydrationApply
  ✔ applies only when gen, doc, save epoch match and editor is closed (1.170076ms)
  ✔ rejects a response for another document (0.146755ms)
  ✔ rejects a superseded generation (0.104287ms)
 ✔ decideHydrationApply (2.515795ms)
 ▶ createEditorHydrationGate
  ✔ does not start a new hydrate while the editor is open (UI-06 fill-window GET) (0.449467ms)
  ✔ deferred hydrate that resolves after the editor opens does not replace the session (0.162255ms)
  ✔ covers hydrate resolving between fill and save (0.149286ms)
  ✔ asks for a hydrate resume after editor_open blocked the first apply (0.370859ms)
  ✔ asks for a hydrate resume when beginHydration is skipped while editing (0.298281ms)
  ✔ requestHydrationResume is consumed once after a protected edit session (0.252152ms)
  ✔ late hydrate after a successful PATCH does not roll back the new baseline (0.412608ms)
 ✔ createEditorHydrationGate (2.69034ms)
 ▶ formatFileSize
  ✔ renders bytes below 1KB without rounding to 0 (0.834686ms)
  ✔ renders KB/MB with sensible precision (0.260662ms)
  ✔ degrades gracefully for invalid input (0.270781ms)
 ✔ formatFileSize (2.69142ms)
 ▶ deliverableFromServerSummary
  ✔ restores uploaded file routing metadata from server facts (0.903473ms)
  ✔ keeps plain freeform docs without upload metadata on the editor path (0.214224ms)
  ✔ handles legacy rows with filename but zero size gracefully (0.130076ms)
  ✔ maps shared space to shared category (0.139396ms)
 ✔ deliverableFromServerSummary (3.529395ms)
 ✔ redactResourceUrl drops query and origin (1.247093ms)
 ✔ summarizeResources flags deferred heavy chunks (1.146446ms)
 ✔ firstPaintHeavyHits lists only deferred module paths (0.186145ms)
 ✔ countInitRequests separates member fanout from session lists (0.283052ms)
 ✔ percentile uses nearest-rank and stays null on empty input (0.167675ms)
 ✔ summarizeLatencies reports p50/p75/p95 (0.216413ms)
 ✔ mergeSessionUser prefers session identity over stale catalog entry (1.251713ms)
 ▶ playwrightPhaseArtifacts
  ✔ defaults unknown or empty UI_PHASE to main (0.723149ms)
  ✔ keeps post-restart distinct from main so the second run cannot overwrite traces (0.178675ms)
  ✔ keeps main results at the historical filename and isolates post-restart recordPass (2.574403ms)
  ✔ shares phase-state and secrets across phases (0.211744ms)
  ✔ defaults assets to the 2026-09-22 rerun tree, not project-143 history (0.330121ms)
  ✔ refuses historical controlled-content and Round20 evidence dirs (2.076958ms)
  ✔ lists closeout groups covering UI-03/05/06/07-08-09/11/13/post-restart (0.811166ms)
 ✔ playwrightPhaseArtifacts (9.540066ms)
 ✔ resolvePostTaskActiveDocId prefers server doc_proj for taskKey (0.9999ms)
 ✔ resolvePostTaskActiveDocId honors preferred server doc id (0.195745ms)
 ✔ shouldSwitchTemplateSlotToServer when active is local template slot (0.169585ms)
 ✔ pollUntilPostTaskServerDoc retries until same taskKey server doc appears (0.495385ms)
 ✔ pollUntilPostTaskServerDoc returns null after maxAttempts without relying on setState (0.206394ms)
 ✔ mapServerProjectToProjectInfo maps creator and preserves local members (0.899954ms)
 ✔ mergeProjectLists drops local-only demo seed projects by default (0.32681ms)
 ✔ mergeProjectLists can retain local-only demo when explicitly opted in (0.158635ms)
 ✔ documentDutyFromMemberRole maps UI reviewer and creator (0.739068ms)
 ✔ exclusiveMemberRoleChange keeps a single reviewer (0.189214ms)
 ✔ memberRoleFromDocumentDuty does not invent reviewer from access role (0.116077ms)
 ✔ toApiProjectRole maps doc roles to member and creator to owner (0.735568ms)
 ✔ fromApiProjectRole maps owner to creator for project creator (0.143375ms)
 ✔ mapApiMemberToProjectMember builds UI record (0.188675ms)
 ▶ projectSettingsDuties
  ✔ toggles full to none by atomically clearing duties (1.726969ms)
  ✔ blocks partial saves and locates required/governance issue codes (0.607871ms)
  ✔ blocks activation preview when no document is under human approval (0.31043ms)
  ✔ allows an all-none configurable draft without three members (0.430327ms)
  ✔ requires three members once any document is managed (0.135766ms)
  ✔ resolves author_only to A/A/A without three-member requirement (0.243022ms)
  ✔ rejects author_only without author or with conflicting reviewers (0.159025ms)
  ✔ mixed full + author_only still requires three members for the full doc (0.324181ms)
  ✔ toggleAuthorOnly clears review/approve slots and restores on disable (0.178825ms)
  ✔ folds later rdev stages and batches human approval keys (0.263173ms)
 ✔ projectSettingsDuties (6.032391ms)
 ▶ projectSettingsMode
  ✔ enters frozen when can_edit_plan is false or the project is active (0.766867ms)
  ✔ keeps an active workflow editable except the definition switch (0.161595ms)
  ✔ stays planning only for an editable draft (0.076008ms)
  ✔ lists enabled duties for the frozen snapshot and skips disabled tasks (0.714029ms)
  ✔ keeps the local draft after PLAN_FROZEN refresh and switches to frozen (0.123676ms)
 ✔ projectSettingsMode (2.846246ms)
 ▶ recheckAck
  ✔ requires an explicit acknowledgement before advance (0.745668ms)
 ✔ recheckAck (1.588373ms)
 ▶ revisionEtag
  ✔ uses stale hydrate ETag for PATCH and never the latest GET (0.492745ms)
  ✔ does not invent 0:0 when controlled hydrate misses etag or errors (0.153046ms)
  ✔ refuses controlled save without a hydrate ETag (0.493985ms)
  ✔ allows freeform bootstrap 0:0 only (0.125246ms)
 ✔ revisionEtag (2.533935ms)
 ▶ saveControlledBlockEdit
  ✔ successful save calls PATCH once with bound docId, hydrate ETag, body and client_block_map (2.507325ms)
  ✔ does not PATCH when editorDocId and activeDocId diverge (0.239383ms)
  ✔ does not PATCH on stale block id and keeps the draft (0.185995ms)
  ✔ 409 does not close the editor or drop the draft (0.854364ms)
  ✔ network errors do not close the editor or drop the draft (0.441767ms)
 ✔ saveControlledBlockEdit (5.514056ms)
 ✔ parseProjectIdFromSessionId decodes sess-p{id}-u{user} (1.520394ms)
 ✔ resolveSessionProjectId prefers explicit id then session id encoding (0.212014ms)
 ✔ mapServerSessionToProjectSession fills project metadata (0.627971ms)
 ✔ mergeSessionLists keeps local stage while applying server counts (0.223494ms)
 ▶ squadCompletionObservation
  ✔ accepts a directly completed adjacent assistant without requiring inflight (1.388318ms)
  ✔ accepts a turn that passed through inflight then completed (0.127666ms)
  ✔ rejects error terminals, missing correspondence, and visible errors (0.093147ms)
  ✔ keeps waiting while still inflight and does not treat inflight as success (0.067458ms)
  ✔ cancelInflightAssistantMessages settles only this-session assistant inflight (0.128636ms)
 ✔ squadCompletionObservation (2.6832ms)
 ✔ running bubble without phases needs replay (0.889653ms)
 ✔ interrupted bubble without phases needs replay (1.515235ms)
 ✔ bubble that already has phases is not replayed (0.164685ms)
 ✔ completed bubble without snapshot needs replay for legacy rows (0.096507ms)
 ✔ completed bubble with snapshot is not replayed (0.113237ms)
 ✔ phase replay key ignores completed chatter (0.153825ms)
 ✔ resolveTaskRunCardTitle prefers artifact title (0.909703ms)
 ✔ resolveTaskRunCardTitle falls back to user message then taskKey (0.208703ms)
 ✔ truncateTitle ellipsizes long text (0.164275ms)
 ✔ formatTaskDuration covers seconds minutes hours (0.160255ms)
 ✔ resolveTaskRunCardBadge maps terminal and live states (0.775347ms)
 ✔ shouldRenderTaskRunCard keeps card shape when run id exists (0.158936ms)
 ✔ runningPhaseDetail prefers live label when it differs from step (0.142095ms)
 ✔ 403 revision denial is not reported as a network error (0.930763ms)
 ✔ 403 author-duty denial surfaces draft-vs-submit guidance (0.157385ms)
 ✔ DOC_DUTY_FORBIDDEN keeps draft-vs-submit guidance (0.754727ms)
 ✔ TASK_SCOPE_UNRESOLVED names the missing workflow stage (0.094747ms)
 ✔ resolveReconnectRunId prefers explicit opts and ignores attempt without opts (0.622101ms)
 ✔ shouldSubscribeExistingRun only for in-flight statuses (0.139905ms)
 ✔ isRemoteTaskTerminal covers confirmation and quality_failed (0.112477ms)
 ✔ nextAttemptOpts carries retained runId for reconnect without new POST (1.610892ms)
 ✔ template documents are labeled as preview, not 编制中 or 已生效受控 (0.958361ms)
 ✔ turnAttachmentRefs keeps doc and parse facts (1.178685ms)
 ✔ describeChatHttpFailure prefers the public validation error (0.460687ms)
 ▶ uploadFreeformRegister
  ✔ buildUploadClientKey prefers explicit hint then doc-upload stamp (0.955591ms)
  ✔ registerUploadAttachments hard-fails without creating orphan on API error (16.603197ms)
  ✔ registerUploadAttachments rejects local doc-upload id from server (0.370409ms)
  ✔ registerUploadAttachments returns server doc-free ids and freeform facts (0.548384ms)
  ✔ deliverableFromFreeformUpload maps server id and upload facts (0.167475ms)
  ✔ buildUploadRegistrationMarkdown embeds real text content when provided (0.144066ms)
  ✔ buildUploadRegistrationMarkdown keeps placeholder for binary files without textContent (0.250482ms)
  ✔ registerUploadAttachments persists source file metadata as server facts (0.301371ms)
 ✔ uploadFreeformRegister (20.688145ms)
 ▶ workflowCategoryNavigation
  ✔ private menu cannot see doc_proj* so it must switch to workflow before exact select (1.049129ms)
  ✔ fail-closes when the workflow menu is showing but the exact doc is missing (0.190554ms)
  ✔ label and taskKey opens always require the workflow category (0.170195ms)
 ✔ workflowCategoryNavigation (3.520565ms)
 ✔ rdev selection explains the CDO prerequisite (0.787547ms)
 ✔ login shell does not statically import export, preview, admin, or chart engines (27.783405ms)
 ▶ no external Google Fonts on login shell
  ✔ index.html has no fonts.googleapis / fonts.gstatic (0.930232ms)
  ✔ styles reference self-hosted fonts.css (0.365889ms)
 ✔ no external Google Fonts on login shell (2.449468ms)
 ▶ 用户上传文档多格式在线预览体系与渲染分流测试
  ✔ 1. 上传附件时正确注入 sourceType: "uploaded_file" 与完整 fileMeta 元数据 (0.865104ms)
  ✔ 2. 扩展名分类规则准确分发至对应的高保真渲染器类型 (0.258543ms)
  ✔ 3. 受控工作流交付物严格隔离：绝不命中 FilePreviewPanel，保持 BlockRenderer 纯净 (0.176665ms)
  ✔ 4. 上传文档支持受控流转：从个人私密一键共享至项目参考库后，在线预览依然可用且权限标态正确 (0.179385ms)
 ✔ 用户上传文档多格式在线预览体系与渲染分流测试 (2.772037ms)
 ▶ 项目立项、成员职责分配与退出留痕治理测试套件
  ✔ 1. 默认项目库包含完整元数据与法定三权成员 (active 状态) (0.6818ms)
  ✔ 2. 跨用户会话可见性严格隔离：未加入成员完全不可见 (0.273272ms)
  ✔ 3. 创建者添加新成员并分配职责后，新成员即刻可见该项目会话 (0.165275ms)
  ✔ 4. 中途加入后期退出 (exited)：保留项目会话但标记已退出只读留痕 (0.162316ms)
  ✔ 5. 超级管理员 (ADMIN) 享有全量通配可见与管理兜底特权 (0.093258ms)
 ✔ 项目立项、成员职责分配与退出留痕治理测试套件 (2.459847ms)
 ▶ 新建项目工程文档模板质量与纯净性保障套件（对标官方参考系统）
  ✔ 1. 必须精确生成 7 份官方全流程受控工作流模板且标记纯净 (0.848355ms)
  ✔ 2. 杜绝旧项目 CUSA-NeuroPulse 串号泄露，动态注入新项目信息 (0.701719ms)
  ✔ 3. 严格对标参考系统模板规范要素，绝非 3 行残缺摘要 (0.411778ms)
  ✔ 4. 缓存丢失场景下即时自愈重建 AST 能力验证 (0.532624ms)
 ✔ 新建项目工程文档模板质量与纯净性保障套件（对标官方参考系统） (7.526547ms)
 ▶ 用户上传文档私密归档与共享流转测试
  ✔ 10 个评审项目均包含已共享的 URS 交付物与 AST，支持跨角色合规查阅且会话均以 assistant 闭环 (1.469307ms)
  ✔ 用户上传附件自动生成私密交付物并在 AST 中建立隔离，支持流转至项目参考库 (1.644061ms)
  ✔ 单轮对话支持一次上传多个文档(上限5个)，批量生成私密交付物并遵循数量约束 (0.552124ms)
 ✔ 用户上传文档私密归档与共享流转测试 (4.558774ms)
 ▶ 工作流切换与消息展示时序测试套件
  ✔ 1. 当存在历史 seed 消息且追加新实时消息时，新消息必须严格沉淀在末尾 (1.243133ms)
  ✔ 2. 切回项目时相同正文不应因 temp/server id 差异重复展示 (0.185805ms)
 ✔ 工作流切换与消息展示时序测试套件 (2.221984ms)
 ✔ DeliverableItem 必须包含 projectId 与 isTemplateOnly 属性 (0.775697ms)
 ℹ tests 532
 ℹ suites 41
 ℹ pass 532
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 1586.5667
 RUN  v5.0.0 /home/weichen/projects/Agent-System/.worktrees/refresh-task-state-recovery-2026-09-28/web
 Test Files  3 passed (3)
      Tests  34 passed (34)
   Start at  06:12:03
   Duration  333ms (transform 67%, import 25%, tests 6%, worker 2%)
 NPM_TEST:0
@@ -83,6 +83,31 @@ func TestMaterializeTargets_MissingAndEmptyFail(t *testing.T) {
 	}
 }
 func TestMaterializeTargets_EmptyPatchKeepsExistingFile(t *testing.T) {
 	workDir := t.TempDir()
 	body := []byte("# 调研报告\n已写在盘上的正文\n")
 	path := filepath.Join(workDir, "output.md")
 	if err := os.WriteFile(path, body, 0o644); err != nil {
 		t.Fatal(err)
 	}
 	got, err := MaterializeTargets(workDir, []string{"output.md"}, map[string][]byte{
 		"output.md": nil,
 	})
 	if err != nil {
 		t.Fatal(err)
 	}
 	if string(got["output.md"]) != string(body) {
 		t.Fatalf("collected %q", got["output.md"])
 	}
 	disk, err := os.ReadFile(path)
 	if err != nil {
 		t.Fatal(err)
 	}
 	if string(disk) != string(body) {
 		t.Fatalf("disk overwritten with %q", disk)
 	}
 }
 func TestMaterializeTargets_RejectsSymlinkEscape(t *testing.T) {
 	root := t.TempDir()
 	outside := t.TempDir()
@@ -244,6 +244,7 @@ func Assemble(cfg *config.Config, svc *bizdev.Service) (*Runtime, error) {
 	stages := stageflow.New(svc)
 	coordEngine.SetStageTaskRunner(stages)
 	coordEngine.SetContextServices(parts.Ctx, parts.KB)
 	coordEngine.SetSessionRunReader(sessionRunAdapter{mgr: taskMgr})
 	if instSvc != nil && instSvc.Available() {
 		coordEngine.SetInstanceService(instSvc)
 	}
@@ -7,6 +7,7 @@ import (
 	"chat-agent-system/internal/app/intentflow"
 	"chat-agent-system/internal/bizdev"
 	"chat-agent-system/internal/chat"
 	"chat-agent-system/internal/coordinator"
 	"chat-agent-system/internal/ctxmgr"
 	"chat-agent-system/internal/taskrun"
 )
@@ -117,6 +118,10 @@ func (a chatTaskStarter) CreateAndStart(ctx context.Context, runID string, recor
 	}, bizdev.StageRunRequest{}, resumeDir)
 }
 func (a chatTaskStarter) BindAssistantMsgID(ctx context.Context, runID string, assistantMsgID int64) error {
 	return a.inner.BindAssistantMsgID(ctx, runID, assistantMsgID)
 }
 // BindChatSession attaches composition-root adapters onto a chat session.
 // Callers that keep a mutable intentflow.Service (e.g. server test hooks) should
 // prefer AdaptChat* helpers with their own fields so swaps stay visible.
@@ -138,3 +143,25 @@ func (rt *Runtime) BindChatSession(sess *chat.Session, env string) {
 		Environment: env,
 	}
 }
 type sessionRunAdapter struct {
 	mgr *taskrun.Manager
 }
 func (a sessionRunAdapter) LatestSessionRun(ctx context.Context, sessionID string) (*coordinator.SessionRunSnapshot, error) {
 	if a.mgr == nil {
 		return nil, nil
 	}
 	rec, err := a.mgr.LatestSessionRun(ctx, sessionID)
 	if err != nil || rec == nil {
 		return nil, err
 	}
 	return &coordinator.SessionRunSnapshot{
 		RunID:        rec.RunID,
 		TaskKey:      rec.TaskKey,
 		Status:       rec.Status,
 		WorkspaceID:  rec.WorkspaceID,
 		ErrorMessage: rec.ErrorMessage,
 		UpdatedAt:    rec.UpdatedAt,
 	}, nil
 }
@@ -70,6 +70,7 @@ type IntentStepView struct {
 // TaskStarter starts a task run without exposing bizdev/stageflow request types to chat.
 type TaskStarter interface {
 	CreateAndStart(ctx context.Context, runID string, record *taskrun.RunRecord, start TaskStartRequest, resumeDir string) (*taskrun.RunRecord, error)
 	BindAssistantMsgID(ctx context.Context, runID string, assistantMsgID int64) error
 }
 // TaskStartRequest carries the chat-owned fields needed to start or resume a run.
@@ -465,6 +465,9 @@ func (s *Session) sendMessageInternal(ctx context.Context, userInput string, str
 		streamer.EmitPhase("dialogue_generating", "正在整理回复...", "assistant", 0, 0, 40)
 	}
 	s.EnsureDocResolver()
 	if s.UserCtx != nil {
 		s.UserCtx.ChatSessionID = s.ID
 	}
 	coordRes, err := s.Coordinator.ProcessIntent(ctx, recRes, s.UserCtx)
 	if err != nil {
 		if streamer != nil {
@@ -473,8 +476,12 @@ func (s *Session) sendMessageInternal(ctx context.Context, userInput string, str
 		return nil, fmt.Errorf("coordination execution failed: %w", err)
 	}
 	var startedRunID string
 	if coordRes.PendingAction != nil {
 		execResult, execErr := s.executeAction(ctx, coordRes.PendingAction, s.UserCtx)
 		if execResult != nil {
 			startedRunID = execResult.RunID
 		}
 		coordRes.NaturalReply = renderActionReply(coordRes.PendingAction, execResult, execErr)
 		coordRes.PendingAction = nil
 	}
@@ -592,8 +599,17 @@ func (s *Session) sendMessageInternal(ctx context.Context, userInput string, str
 			asstMeta["client_turn_id"] = clientTurnID
 			asstMeta["client_id"] = clientTurnID + ":assistant"
 		}
 		if startedRunID != "" {
 			asstMeta["run_id"] = startedRunID
 			if asstMeta["task_status"] == nil || asstMeta["task_status"] == "" {
 				asstMeta["task_status"] = "running"
 			}
 		}
 		asstMsgID, _ := s.CtxMgr.RecordUserMessageWithID(ctx, s.ID, 0, "assistant", coordRes.NaturalReply, string(coordRes.Intent), string(recRes.Branch), asstMeta)
 		if startedRunID != "" && asstMsgID > 0 && s.Actions != nil && s.Actions.TaskRuns != nil {
 			_ = s.Actions.TaskRuns.BindAssistantMsgID(ctx, startedRunID, asstMsgID)
 		}
 		// 异步记录问答切片到向量记忆库 (P2 检索增强，不阻塞流式或同步答复)
 		sessID := s.ID
@@ -222,6 +222,8 @@ type UserProjectContext struct {
 	RegulationHits []kb.HitItem `json:"-"`
 	// UpgradeOffered 路由升级建议冷却标记（每会话至多主动建议一次，防打扰——对齐"参谋不强推"人设）
 	UpgradeOffered bool `json:"-"`
 	// ChatSessionID 当前对话会话，供进度问询读取真实 task run。不序列化。
 	ChatSessionID string `json:"-"`
 	// DocumentSummaryReader 仅用于服务端鉴权后的只读正文访问，不序列化。
 	DocumentSummaryReader DocumentSummaryReader `json:"-"`
 	// RealtimeSnippet 本轮实时检索（天气/新闻等）成功结果，供 Prompt 注入；不序列化。
@@ -83,6 +83,7 @@ type CoordinatorEngine struct {
 	docResolver      DocResolver
 	pendingSupps     PendingSupplementReader
 	reopenLister     ReopenLister
 	sessionRuns      SessionRunReader
 	// OnSupplement is optional; composition typically wires supplementimpact.TriggerAsync.
 	// Nil-safe: unset hooks are skipped.
 	OnSupplement          func(ctx context.Context, projectID, userID int64, sessionID, summary, markdown string)
@@ -108,6 +109,27 @@ func (e *CoordinatorEngine) tasks() StageTaskRunner {
 	return e.stageTasks
 }
 // SessionRunSnapshot 是进度问询可读的任务运行投影，不含工作目录绝对路径。
 type SessionRunSnapshot struct {
 	RunID        string
 	TaskKey      string
 	Status       string
 	WorkspaceID  string
 	ErrorMessage string
 	UpdatedAt    time.Time
 }
 // SessionRunReader 按会话读取最近一次任务运行。未注入时进度问询保持模板答复。
 type SessionRunReader interface {
 	LatestSessionRun(ctx context.Context, sessionID string) (*SessionRunSnapshot, error)
 }
 func (e *CoordinatorEngine) SetSessionRunReader(reader SessionRunReader) {
 	if e != nil {
 		e.sessionRuns = reader
 	}
 }
 func (e *CoordinatorEngine) SetContextServices(ctxMgr *ctxmgr.ContextManager, kbClient *kb.Client) {
 	if e != nil {
 		e.ctxMgr = ctxMgr
@@ -151,6 +151,13 @@ func (e *CoordinatorEngine) handleQueryProgress(ctx context.Context, cand *inten
 		sb.WriteString(brief.CommunicationKit)
 	case "progress":
 		if snap := e.latestSessionRun(ctx, userCtx); snap != nil {
 			sb.WriteString(formatSessionRunReply(snap))
 			if brief != nil {
 				sb.WriteString(fmt.Sprintf("\n项目成熟度仍为 %s（%d/100）。", brief.MaturityLevel, brief.MaturityScore))
 			}
 			break
 		}
 		sb.WriteString(fmt.Sprintf("【进展汇报】 项目当前成熟度等级为 %s（评分 %d/100）。\n", brief.MaturityLevel, brief.MaturityScore))
 		sb.WriteString(fmt.Sprintf("诊断结论：%s\n\n", brief.MaturityHeadline))
 		if len(brief.NextBestActions) > 0 {
@@ -196,6 +203,49 @@ func (e *CoordinatorEngine) handleQueryProgress(ctx context.Context, cand *inten
 	}), nil
 }
 func (e *CoordinatorEngine) latestSessionRun(ctx context.Context, userCtx *UserProjectContext) *SessionRunSnapshot {
 	if e == nil || e.sessionRuns == nil || userCtx == nil || strings.TrimSpace(userCtx.ChatSessionID) == "" {
 		return nil
 	}
 	snap, err := e.sessionRuns.LatestSessionRun(ctx, userCtx.ChatSessionID)
 	if err != nil || snap == nil || snap.RunID == "" {
 		return nil
 	}
 	return snap
 }
 func formatSessionRunReply(snap *SessionRunSnapshot) string {
 	label := map[string]string{
 		"pending":                     "排队中",
 		"running":                     "正在编写",
 		"completed":                   "已完成",
 		"failed":                      "执行失败",
 		"finalizing_failed":           "收尾失败",
 		"interrupted":                 "已中断，可以继续",
 		"cancelled":                   "已取消",
 		"pending_author_confirmation": "待编制人确认",
 		"quality_failed":              "质量未通过",
 		"blocked_pending_governance":  "待激活工作流",
 	}
 	status := label[snap.Status]
 	if status == "" {
 		status = snap.Status
 	}
 	taskName := snap.TaskKey
 	if taskName == "" {
 		taskName = "当前任务"
 	}
 	var b strings.Builder
 	b.WriteString(fmt.Sprintf("「%s」当前状态：%s。", taskName, status))
 	if snap.WorkspaceID != "" {
 		b.WriteString(fmt.Sprintf("工作区标识：%s。", snap.WorkspaceID))
 	}
 	if snap.ErrorMessage != "" && (snap.Status == "failed" || snap.Status == "interrupted" || snap.Status == "finalizing_failed") {
 		b.WriteString("原因：" + snap.ErrorMessage + "。")
 	}
 	return b.String()
 }
 // synthesizeBaselineBrief 在用户尚未执行过任何阶段/任务（无持久化参谋简报）时，
 // 基于黑板事实与项目信息合成 intake 级基准参谋简报，确保查询类意图始终有实质研判可用。
 func synthesizeBaselineBrief(userCtx *UserProjectContext) *squad.AdvisorBriefing {
@@ -0,0 +1,45 @@
 package coordinator
 import (
 	"context"
 	"strings"
 	"testing"
 )
 type fakeSessionRuns struct {
 	snap *SessionRunSnapshot
 	err  error
 }
 func (f fakeSessionRuns) LatestSessionRun(context.Context, string) (*SessionRunSnapshot, error) {
 	return f.snap, f.err
 }
 func TestFormatSessionRunReply_RunningIncludesWorkspace(t *testing.T) {
 	got := formatSessionRunReply(&SessionRunSnapshot{
 		TaskKey:     "research",
 		Status:      "running",
 		WorkspaceID: "ws_research_1",
 	})
 	for _, want := range []string{"research", "正在编写", "ws_research_1"} {
 		if !strings.Contains(got, want) {
 			t.Fatalf("reply %q missing %q", got, want)
 		}
 	}
 }
 func TestLatestSessionRun_ReadsInjectedReader(t *testing.T) {
 	engine := NewCoordinatorEngine(nil)
 	engine.SetSessionRunReader(fakeSessionRuns{snap: &SessionRunSnapshot{
 		RunID:   "run-1",
 		TaskKey: "research",
 		Status:  "running",
 	}})
 	got := engine.latestSessionRun(context.Background(), &UserProjectContext{ChatSessionID: "sess-1"})
 	if got == nil || got.RunID != "run-1" {
 		t.Fatalf("latest = %#v", got)
 	}
 	if engine.latestSessionRun(context.Background(), &UserProjectContext{}) != nil {
 		t.Fatal("empty chat session must not invent a run")
 	}
 }
@@ -11,6 +11,30 @@ import (
 	"chat-agent-system/internal/squad"
 )
 // isProgressInquiry 为真时，带任务名词的疑问句应让行 QUERY_PROGRESS，而不是 TRIGGER_TASK。
 func isProgressInquiry(input string) bool {
 	if !isInterrogative(input) {
 		return false
 	}
 	markers := []string{"进度", "进展", "怎么样了", "状态如何", "写完", "完成了吗", "完成没有", "好了吗", "好了没有"}
 	hit := false
 	for _, marker := range markers {
 		if strings.Contains(input, marker) {
 			hit = true
 			break
 		}
 	}
 	if !hit {
 		return false
 	}
 	for _, verb := range []string{"开始", "重新", "再来", "重跑", "再生成", "重新生成"} {
 		if strings.Contains(input, verb) {
 			return false
 		}
 	}
 	return true
 }
 // isInterrogative 判断输入是否疑问句：句尾问号/「吗」「呢」，或句中出现
 // 「怎么/如何/什么/多少/哪/是否/大概/多久/为什么」等疑问词。
 // 用于把「查询/咨询」类输入从祈使句「指令」类意图（阶段推进等）中排除。
@@ -248,6 +272,11 @@ func requirementsTriggerPhrase(taskKey string) string {
 }
 func matchTriggerTask(input string, userCtx *ContextSnapshot) *CandidateIntent {
 	// 疑问句里的「编写进度怎么样了」是在问状态，不是再启动一份交付物。
 	// 显式启动词（开始/重新/重跑）仍走任务触发。
 	if isProgressInquiry(input) {
 		return nil
 	}
 	type phrase struct {
 		text string
 		key  string
@@ -72,6 +72,30 @@ func TestIntentRecognition_AdvanceStage(t *testing.T) {
 	}
 }
 func TestIntentRecognition_ProgressQuestionDoesNotStartTask(t *testing.T) {
 	rec := intent.NewRecognizer()
 	ctx := &intent.ContextSnapshot{SnapshotID: "ctx_progress"}
 	progress, err := rec.Recognize(context.Background(), "调研报告编写进度怎么样了", ctx)
 	if err != nil {
 		t.Fatalf("recognize progress: %v", err)
 	}
 	if progress.TopIntent.Intent != intent.IntentQueryProgress {
 		t.Fatalf("progress question intent = %s, want QUERY_PROGRESS", progress.TopIntent.Intent)
 	}
 	start, err := rec.Recognize(context.Background(), "开始做调研报告", ctx)
 	if err != nil {
 		t.Fatalf("recognize start: %v", err)
 	}
 	if start.TopIntent.Intent != intent.IntentTriggerTask {
 		t.Fatalf("start intent = %s, want TRIGGER_TASK", start.TopIntent.Intent)
 	}
 	if start.TopIntent.FilledSlots[intent.SlotTargetTask] != "research" {
 		t.Fatalf("start task = %s, want research", start.TopIntent.FilledSlots[intent.SlotTargetTask])
 	}
 }
 func TestIntentRecognition_TriggerTask(t *testing.T) {
 	rec := intent.NewRecognizer()
 	ctx := &intent.ContextSnapshot{
@@ -230,6 +230,7 @@ func (s *Server) Routes() http.Handler {
 	mux.HandleFunc("POST /api/v1/tasks/run/stream", s.handleRunTaskStream)
 	mux.HandleFunc("GET /api/v1/tasks/run/stream", s.handleRunTaskStream)
 	mux.HandleFunc("POST /api/v1/tasks/runs", s.handleCreateTaskRun)
 	mux.HandleFunc("GET /api/v1/tasks/runs", s.handleListTaskRuns)
 	mux.HandleFunc("GET /api/v1/tasks/runs/{run_id}", s.handleGetTaskRun)
 	mux.HandleFunc("GET /api/v1/tasks/runs/{run_id}/events", s.handleTaskRunEvents)
 	mux.HandleFunc("POST /api/v1/tasks/runs/{run_id}/cancel", s.handleCancelTaskRun)
@@ -49,6 +49,12 @@ func (failInsertStore) HeartbeatLease(context.Context, string, string, time.Dura
 func (failInsertStore) ReleaseLease(context.Context, string, string) error { return nil }
 func (failInsertStore) ResetForResume(context.Context, string, int) error  { return nil }
 func (failInsertStore) MarkOrphanedRunning(context.Context) (int, error)   { return 0, nil }
 func (failInsertStore) ListActiveBySession(context.Context, string) ([]*taskrun.RunRecord, error) {
 	return nil, nil
 }
 func (failInsertStore) LatestBySession(context.Context, string) (*taskrun.RunRecord, error) {
 	return nil, nil
 }
 func TestRunTaskStream_InvalidJSON_Returns400(t *testing.T) {
 	srv := newTestServer(t)
@@ -209,6 +209,9 @@ func (s *Server) handleCreateTaskRun(w http.ResponseWriter, r *http.Request) {
 				"doc_id":         req.DocID,
 			},
 		}
 		if req.ClientTurnID != "" {
 			asstMeta["client_id"] = req.ClientTurnID + ":assistant"
 		}
 		projectinstance.ApplyScopeToMetadata(asstMeta, taskScope)
 		assistantMsgID, _ = ctxMgr.RecordUserMessageWithID(r.Context(), req.SessionID, 0, "assistant", "任务执行中…", "TRIGGER_TASK", "", asstMeta)
 		claimed.AssistantMsgID = assistantMsgID
@@ -613,6 +616,45 @@ func (s *Server) cancelPiAgentRunner(workspaceID string) {
 	_ = resp.Body.Close()
 }
 func (s *Server) handleListTaskRuns(w http.ResponseWriter, r *http.Request) {
 	user, ok := s.requirePrincipal(w, r)
 	if !ok {
 		return
 	}
 	sessionID := strings.TrimSpace(r.URL.Query().Get("session_id"))
 	if sessionID == "" {
 		writeSafeError(w, r, http.StatusBadRequest, "session_id required")
 		return
 	}
 	if r.URL.Query().Get("active") != "1" {
 		writeSafeError(w, r, http.StatusBadRequest, "active=1 required")
 		return
 	}
 	if !s.assertSessionOwnedByUser(w, r, sessionID, user.ID) {
 		return
 	}
 	if s.taskRuns() == nil {
 		writeJSONResponse(w, http.StatusOK, map[string]any{"runs": []taskRunResponse{}})
 		return
 	}
 	listed, err := s.taskRuns().ListActiveBySession(r.Context(), sessionID)
 	if err != nil {
 		writeDomainError(w, r, http.StatusInternalServerError, err)
 		return
 	}
 	out := make([]taskRunResponse, 0, len(listed))
 	for _, rec := range listed {
 		if rec == nil {
 			continue
 		}
 		if !s.isSystemAdminUser(r.Context(), user) && rec.UserID > 0 && rec.UserID != user.ID {
 			continue
 		}
 		out = append(out, s.buildTaskRunResponse(rec))
 	}
 	writeJSONResponse(w, http.StatusOK, map[string]any{"runs": out})
 }
 func (s *Server) handleGetTaskRun(w http.ResponseWriter, r *http.Request) {
 	user, ok := s.requirePrincipal(w, r)
 	if !ok {
@@ -26,7 +26,7 @@ import (
 // 步骤状态常量（与既有 workflow.json 的 legacy 状态字符串完全兼容）
 const (
 	StepStatePending   = "pending"
-	StepStateExecuting = "running"    // legacy 值保持不变
+	StepStateExecuting = "running"   // legacy 值保持不变
 	StepStateVerifying = "verifying" // 新增：校验态
 	StepStateRepairing = "repairing" // 新增：修复态
 	StepStateCompleted = "completed"
@@ -213,15 +213,23 @@ func (l *BudgetLedger) UsageSummary() string {
 // TraceEntry 单次底层 agent 调用的轨迹记录
 type TraceEntry struct {
-	Seq              int       `json:"seq"`
+	Seq                 int       `json:"seq"`
-	StepID           string    `json:"step_id"`
+	StepID              string    `json:"step_id"`
-	Role             string    `json:"role"`
+	Role                string    `json:"role"`
-	Timestamp        time.Time `json:"ts"`
+	Timestamp           time.Time `json:"ts"`
-	PromptDigest     string    `json:"prompt_digest"`
+	PromptDigest        string    `json:"prompt_digest"`
-	OutputDigest     string    `json:"output_digest"`
+	OutputDigest        string    `json:"output_digest"`
-	SimilarityToPrev float64   `json:"similarity_to_prev"`
+	SimilarityToPrev    float64   `json:"similarity_to_prev"`
-	DurationMs       int64     `json:"duration_ms"`
+	DurationMs          int64     `json:"duration_ms"`
-	Success          bool      `json:"success"`
+	Success             bool      `json:"success"`
 	ExecErr             string    `json:"exec_err,omitempty"`
 	ResultSuccess       bool      `json:"res_success,omitempty"`
 	ResultError         string    `json:"res_error_reason,omitempty"`
 	HasTrustedPatch     bool      `json:"has_trusted_patch,omitempty"`
 	WroteThisInvocation bool      `json:"wrote_this_invocation,omitempty"`
 	MatErr              string    `json:"mat_err,omitempty"`
 	RecoveredFromDisk   bool      `json:"recovered_from_disk,omitempty"`
 	CancelReason        string    `json:"cancel_reason,omitempty"`
 }
 // AppendTrace 追加一条执行轨迹（jsonl，天然即历史轨迹审计源）
@@ -462,6 +470,3 @@ func SortedStepIDs(wf *WorkflowPlan) []string {
 	sort.Strings(ids)
 	return ids
 }
@@ -5,6 +5,7 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
 	"log/slog"
 	"os"
 	"path/filepath"
 	"strings"
@@ -264,28 +265,31 @@ func (d *WorkflowDriver) Execute(ctx context.Context, wsDir string, backend agen
 				for _, s := range parallelGroup {
 					sCopy := s
 					wg.Add(1)
-		go func() {
+					go func() {
-			defer wg.Done()
+						defer wg.Done()
-			defer func() {
+						defer func() {
-				if rec := recover(); rec != nil {
+							if rec := recover(); rec != nil {
-					mu.Lock()
+								mu.Lock()
-					if groupErr == nil {
+								if groupErr == nil {
-						groupErr = fmt.Errorf("parallel step panic: %v", rec)
+									groupErr = fmt.Errorf("parallel step panic: %v", rec)
-					}
+								}
-					mu.Unlock()
+								mu.Unlock()
-				}
+							}
-			}()
+						}()
-			if err := d.runOneStep(runCtx, wsDir, sCopy, backend, timeout, opts.StepNotify, &mu, &producedFiles, ledger); err != nil {
+						if err := d.runOneStep(runCtx, wsDir, sCopy, backend, timeout, opts.StepNotify, &mu, &producedFiles, ledger); err != nil {
-				mu.Lock()
+							mu.Lock()
-				if groupErr == nil {
+							if groupErr == nil {
-					groupErr = err
+								groupErr = err
-				}
+							}
-				mu.Unlock()
+							mu.Unlock()
-			}
+						}
-		}()
+					}()
 				}
 				wg.Wait()
 				if groupErr != nil {
 					if cerr := contextErrString(runCtx); cerr != "" {
 						slog.Warn("squad_parallel_cancelled", "error", groupErr.Error(), "cancel_reason", cerr, "parent_cancel", contextErrString(ctx))
 					}
 					wf.Status = "failed"
 					_ = d.SaveWorkflow(wsDir, wf)
 					_ = d.updateWorkspaceStatus(wsDir, "failed")
@@ -335,25 +339,25 @@ func (d *WorkflowDriver) Execute(ctx context.Context, wsDir string, backend agen
 					Duration:    time.Since(start),
 				}, rErr
 			}
-		// 治理：受控迁移至执行态（legacy 重跑场景直接放行）
+			// 治理：受控迁移至执行态（legacy 重跑场景直接放行）
-		markStepExecuting(wf, step.ID)
+			markStepExecuting(wf, step.ID)
-		if err := d.runOneStep(runCtx, wsDir, step, backend, timeout, opts.StepNotify, &mu, &producedFiles, ledger); err != nil {
+			if err := d.runOneStep(runCtx, wsDir, step, backend, timeout, opts.StepNotify, &mu, &producedFiles, ledger); err != nil {
-			// A5 非关键步骤降级：记录 degraded 继续推进，收口统一标 degraded
+				// A5 非关键步骤降级：记录 degraded 继续推进，收口统一标 degraded
-			if !step.IsCritical() {
+				if !step.IsCritical() {
-				markStepDegraded(wsDir, wf, step, err)
+					markStepDegraded(wsDir, wf, step, err)
-				stepsRun++
+					stepsRun++
 					_ = d.SaveWorkflow(wsDir, wf)
 					continue
 				}
 				wf.Status = "failed"
 				_ = d.SaveWorkflow(wsDir, wf)
-				continue
+				_ = d.updateWorkspaceStatus(wsDir, "failed")
 				return &WorkflowExecutionResult{
 					Success:     false,
 					ErrorReason: err.Error(),
 					Duration:    time.Since(start),
 				}, err
 			}
 			wf.Status = "failed"
 			_ = d.SaveWorkflow(wsDir, wf)
 			_ = d.updateWorkspaceStatus(wsDir, "failed")
 			return &WorkflowExecutionResult{
 				Success:     false,
 				ErrorReason: err.Error(),
 				Duration:    time.Since(start),
 			}, err
 		}
 			stepsRun++
 			// 治理：FSM 受控迁移 + 步骤归档锁定 + 软线提醒
 			if tErr := TransitionStep(wf, step.ID, StepStateCompleted); tErr == nil {
@@ -549,13 +553,13 @@ func (d *WorkflowDriver) runOneStep(
 	}
 	var (
-		res         *agent.Result
+		res        *agent.Result
-		execErr     error
+		execErr    error
-		matErr      error
+		matErr     error
-		duration    int64
+		duration   int64
-		ok          bool
+		ok         bool
-		lastAssess  agent.FailureAssessment
+		lastAssess agent.FailureAssessment
-		basePrompt  = string(taskData)
+		basePrompt = string(taskData)
 	)
 	for attempt := 1; attempt <= agent.MaxTransientStepInvocations; attempt++ {
 		onDisk, _ := os.ReadFile(prevOutPath)
@@ -611,22 +615,51 @@ func (d *WorkflowDriver) runOneStep(
 		if newContent == "" && wroteThisInvocation {
 			newContent = string(onDisk)
 		}
-		ok = execErr == nil && res != nil && res.Success && matErr == nil
+		recoveredFromDisk := false
 		if strings.TrimSpace(newContent) == "" {
 			if again, rerr := os.ReadFile(prevOutPath); rerr == nil && deliverableWroteThisInvocation(prevContent, again) {
 				newContent = string(again)
 				wroteThisInvocation = true
 				recoveredFromDisk = true
 			}
 		}
 		ok = stepOutputAccepted(execErr, matErr, wroteThisInvocation, newContent, res)
 		if !ok || recoveredFromDisk {
 			slog.Info("squad_step_output_judge",
 				"step_id", step.ID,
 				"exec_err", errString(execErr),
 				"res_success", res != nil && res.Success,
 				"res_error_reason", resultErrorReason(res),
 				"has_trusted_patch", hasTrustedPatch,
 				"wrote_this_invocation", wroteThisInvocation,
 				"mat_err", errString(matErr),
 				"recovered_from_disk", recoveredFromDisk,
 				"accepted", ok,
 			)
 		}
 		sim := OutputSimilarity(prevContent, newContent)
 		if mu != nil {
 			mu.Lock()
 		}
 		traceSeq := len(LoadTrace(wsDir)) + 1
 		_ = AppendTrace(wsDir, TraceEntry{
-			Seq:              traceSeq,
+			Seq:                 traceSeq,
-			StepID:           step.ID,
+			StepID:              step.ID,
-			Role:             step.Role,
+			Role:                step.Role,
-			Timestamp:        time.Now(),
+			Timestamp:           time.Now(),
-			PromptDigest:     digestString(string(taskData)),
+			PromptDigest:        digestString(string(taskData)),
-			OutputDigest:     digestString(newContent),
+			OutputDigest:        digestString(newContent),
-			SimilarityToPrev: sim,
+			SimilarityToPrev:    sim,
-			DurationMs:       duration,
+			DurationMs:          duration,
-			Success:          ok,
+			Success:             ok,
 			ExecErr:             errString(execErr),
 			ResultSuccess:       res != nil && res.Success,
 			ResultError:         resultErrorReason(res),
 			HasTrustedPatch:     hasTrustedPatch,
 			WroteThisInvocation: wroteThisInvocation,
 			MatErr:              errString(matErr),
 			RecoveredFromDisk:   strings.TrimSpace(newContent) != "" && !(res != nil && res.Success),
 			CancelReason:        contextErrString(ctx),
 		})
 		if mu != nil {
 			mu.Unlock()
@@ -662,7 +695,7 @@ func (d *WorkflowDriver) runOneStep(
 		}
 	}
-	if execErr != nil {
+	if !ok && execErr != nil {
 		if streamer != nil {
 			streamer.EmitPhase("step_failed", fmt.Sprintf("%s执行异常：%s", roleTitle, execErr.Error()), step.Role, 0, 0, 0)
 		}
@@ -680,7 +713,7 @@ func (d *WorkflowDriver) runOneStep(
 		return fmt.Errorf("step %s (%s) backend execute: %w", step.ID, step.Role, execErr)
 	}
-	if res != nil && !res.Success {
+	if !ok && res != nil && !res.Success {
 		errMsg := res.ErrorReason
 		if streamer != nil {
 			streamer.EmitPhase("step_failed", fmt.Sprintf("%s执行未成功：%s", roleTitle, errMsg), step.Role, 0, 0, 0)
@@ -761,6 +794,55 @@ func (d *WorkflowDriver) runOneStep(
 	return nil
 }
 // stepOutputAccepted 以本轮写盘产物为事实源。runner 把 Success 标成 false
 // （例如 typed result 没带回 patch）但文件已经写好时，不让这一处否决步骤。
 func stepOutputAccepted(execErr, matErr error, wrote bool, newContent string, res *agent.Result) bool {
 	if execErr != nil || matErr != nil {
 		return false
 	}
 	if res != nil && res.Success {
 		return true
 	}
 	if !wrote || strings.TrimSpace(newContent) == "" {
 		return false
 	}
 	// 连接中断或可重试失败留下的文件可能是截断的，不能当成交付成功。
 	// 交付物缺失/空结果但盘上已有正文时，以文件为准。
 	if res != nil && (res.Retryable || hardRunnerFailure(res.ErrorReason)) {
 		return false
 	}
 	return true
 }
 func hardRunnerFailure(reason string) bool {
 	reason = strings.ToLower(strings.TrimSpace(reason))
 	if reason == "" || strings.Contains(reason, "missing target deliverable") {
 		return false
 	}
 	return true
 }
 func errString(err error) string {
 	if err == nil {
 		return ""
 	}
 	return err.Error()
 }
 func resultErrorReason(res *agent.Result) string {
 	if res == nil {
 		return ""
 	}
 	return res.ErrorReason
 }
 func contextErrString(ctx context.Context) string {
 	if ctx == nil || ctx.Err() == nil {
 		return ""
 	}
 	return ctx.Err().Error()
 }
 func deliverableWroteThisInvocation(baseline string, onDisk []byte) bool {
 	if len(bytes.TrimSpace(onDisk)) == 0 {
 		return false
@@ -0,0 +1,31 @@
 package squad
 import (
 	"errors"
 	"testing"
 	"chat-agent-system/internal/agent"
 )
 func TestStepOutputAccepted_DiskContentBeatsMissingDeliverable(t *testing.T) {
 	missing := &agent.Result{Success: false, ErrorReason: "missing target deliverable output.md"}
 	if !stepOutputAccepted(nil, nil, true, "# 调研报告\n正文", missing) {
 		t.Fatal("non-empty file must be accepted when the runner only reported a missing deliverable")
 	}
 	truncated := &agent.Result{Success: false, ErrorReason: "Connection error.", Retryable: true}
 	if stepOutputAccepted(nil, nil, true, "# partial", truncated) {
 		t.Fatal("retryable connection failure must stay fail-closed even if a file exists")
 	}
 	if stepOutputAccepted(nil, nil, true, " \n", missing) {
 		t.Fatal("blank content must not override a failed runner result")
 	}
 	if stepOutputAccepted(errors.New("backend"), nil, true, "# 调研报告", missing) {
 		t.Fatal("execute error must reject the step")
 	}
 	if stepOutputAccepted(nil, errors.New("materialize"), true, "# 调研报告", &agent.Result{Success: true}) {
 		t.Fatal("materialize error must reject the step")
 	}
 	if !stepOutputAccepted(nil, nil, false, "", &agent.Result{Success: true}) {
 		t.Fatal("runner success without a new file still counts")
 	}
 }
@@ -0,0 +1,36 @@
 package taskrun
 import (
 	"context"
 	"testing"
 )
 func TestListActiveBySession_MemoryOnly(t *testing.T) {
 	m := NewManager(nil, nil)
 	m.mu.Lock()
 	m.runs["run-live"] = &activeRun{record: &RunRecord{
 		RunID: "run-live", SessionID: "sess-1", Status: "running", TaskKey: "research",
 	}}
 	m.runs["run-done"] = &activeRun{record: &RunRecord{
 		RunID: "run-done", SessionID: "sess-1", Status: "completed",
 	}}
 	m.runs["other"] = &activeRun{record: &RunRecord{
 		RunID: "other", SessionID: "sess-2", Status: "running",
 	}}
 	m.mu.Unlock()
 	listed, err := m.ListActiveBySession(context.Background(), "sess-1")
 	if err != nil {
 		t.Fatal(err)
 	}
 	if len(listed) != 1 || listed[0].RunID != "run-live" {
 		t.Fatalf("listed = %#v", listed)
 	}
 }
 func TestWriteRunningAnchor_NoContextManagerIsNoop(t *testing.T) {
 	m := NewManager(nil, nil)
 	m.writeRunningAnchor(context.Background(), &RunRecord{
 		RunID: "run-1", AssistantMsgID: 9, TaskKey: "research", Status: "running",
 	})
 }
@@ -91,6 +91,37 @@ func (m *memIdemStore) HeartbeatLease(context.Context, string, string, time.Dura
 func (m *memIdemStore) ReleaseLease(context.Context, string, string) error { return nil }
 func (m *memIdemStore) ResetForResume(context.Context, string, int) error  { return nil }
 func (m *memIdemStore) MarkOrphanedRunning(context.Context) (int, error)   { return 0, nil }
 func (m *memIdemStore) ListActiveBySession(_ context.Context, sessionID string) ([]*RunRecord, error) {
 	m.mu.Lock()
 	defer m.mu.Unlock()
 	out := make([]*RunRecord, 0)
 	for _, rec := range m.byID {
 		if rec == nil || rec.SessionID != sessionID {
 			continue
 		}
 		if rec.Status != "pending" && rec.Status != "running" {
 			continue
 		}
 		cpy := *rec
 		out = append(out, &cpy)
 	}
 	return out, nil
 }
 func (m *memIdemStore) LatestBySession(_ context.Context, sessionID string) (*RunRecord, error) {
 	m.mu.Lock()
 	defer m.mu.Unlock()
 	var best *RunRecord
 	for _, rec := range m.byID {
 		if rec == nil || rec.SessionID != sessionID {
 			continue
 		}
 		if best == nil || rec.UpdatedAt.After(best.UpdatedAt) {
 			cpy := *rec
 			best = &cpy
 		}
 	}
 	return best, nil
 }
 func TestClaimRun_ConcurrentSameKey_OneOwner(t *testing.T) {
 	store := newMemIdemStore()
@@ -255,8 +255,11 @@ func (m *Manager) BindAssistantMsgID(ctx context.Context, runID string, assistan
 		rec := *ar.record
 		ar.mu.Unlock()
 		if m.store != nil && m.store.Available() {
-			return m.store.UpdateRun(ctx, &rec)
+			if err := m.store.UpdateRun(ctx, &rec); err != nil {
 				return err
 			}
 		}
 		m.writeRunningAnchor(ctx, &rec)
 		return nil
 	}
 	if m.store != nil && m.store.Available() {
@@ -266,11 +269,121 @@ func (m *Manager) BindAssistantMsgID(ctx context.Context, runID string, assistan
 		}
 		rec.AssistantMsgID = assistantMsgID
 		rec.UpdatedAt = time.Now()
-		return m.store.UpdateRun(ctx, rec)
+		if err := m.store.UpdateRun(ctx, rec); err != nil {
 			return err
 		}
 		m.writeRunningAnchor(ctx, rec)
 		return nil
 	}
 	return nil
 }
 // writeRunningAnchor 在 run 仍进行中时就把 run_id / task_status 写进助手消息。
 // 刷新后窗口能按 taskRunId 重放进度，不必等终态 finalize。
 func (m *Manager) writeRunningAnchor(ctx context.Context, rec *RunRecord) {
 	if m == nil || m.ctxMgr == nil || rec == nil || rec.AssistantMsgID <= 0 || rec.RunID == "" {
 		return
 	}
 	meta := map[string]any{
 		"run_id":      rec.RunID,
 		"task_key":    rec.TaskKey,
 		"task_status": "running",
 	}
 	if rec.WorkspaceID != "" {
 		meta["workspace_id"] = rec.WorkspaceID
 	}
 	if err := m.ctxMgr.UpdateMessageMetadata(ctx, rec.AssistantMsgID, meta, "running"); err != nil {
 		slog.Warn("taskrun_running_anchor_failed", "run_id", rec.RunID, "message_id", rec.AssistantMsgID, "error", err)
 	}
 }
 // ListActiveBySession 返回该会话仍在 pending/running 且租约未过期的 run。
 // 内存中的活跃 run 与存储合并，覆盖尚未落库的窗口。
 func (m *Manager) ListActiveBySession(ctx context.Context, sessionID string) ([]*RunRecord, error) {
 	if m == nil || sessionID == "" {
 		return nil, nil
 	}
 	var stored []*RunRecord
 	if m.store != nil && m.store.Available() {
 		listed, err := m.store.ListActiveBySession(ctx, sessionID)
 		if err != nil {
 			return nil, err
 		}
 		stored = listed
 	}
 	return mergeRuns(stored, m.memorySessionRuns(sessionID, true)), nil
 }
 // LatestSessionRun 返回该会话最近一次 run（含已结束），供进度问询读取真实状态。
 func (m *Manager) LatestSessionRun(ctx context.Context, sessionID string) (*RunRecord, error) {
 	if m == nil || sessionID == "" {
 		return nil, nil
 	}
 	var best *RunRecord
 	if m.store != nil && m.store.Available() {
 		rec, err := m.store.LatestBySession(ctx, sessionID)
 		if err != nil {
 			return nil, err
 		}
 		best = rec
 	}
 	for _, mem := range m.memorySessionRuns(sessionID, false) {
 		if best == nil || mem.UpdatedAt.After(best.UpdatedAt) {
 			cpy := *mem
 			best = &cpy
 		}
 	}
 	return best, nil
 }
 func (m *Manager) memorySessionRuns(sessionID string, activeOnly bool) []*RunRecord {
 	m.mu.RLock()
 	defer m.mu.RUnlock()
 	out := make([]*RunRecord, 0)
 	for _, ar := range m.runs {
 		if ar == nil || ar.record == nil || ar.record.SessionID != sessionID {
 			continue
 		}
 		ar.mu.Lock()
 		rec := *ar.record
 		ar.mu.Unlock()
 		if activeOnly && rec.Status != "pending" && rec.Status != "running" {
 			continue
 		}
 		out = append(out, &rec)
 	}
 	return out
 }
 func mergeRuns(stored, memory []*RunRecord) []*RunRecord {
 	byID := make(map[string]*RunRecord, len(stored)+len(memory))
 	order := make([]string, 0, len(stored)+len(memory))
 	add := func(rec *RunRecord) {
 		if rec == nil || rec.RunID == "" {
 			return
 		}
 		if prev, ok := byID[rec.RunID]; ok {
 			if rec.UpdatedAt.After(prev.UpdatedAt) {
 				byID[rec.RunID] = rec
 			}
 			return
 		}
 		byID[rec.RunID] = rec
 		order = append(order, rec.RunID)
 	}
 	for _, rec := range stored {
 		add(rec)
 	}
 	for _, rec := range memory {
 		add(rec)
 	}
 	out := make([]*RunRecord, 0, len(order))
 	for _, id := range order {
 		out = append(out, byID[id])
 	}
 	return out
 }
 // StartClaimedRun 仅对 ClaimRun 成功的 owner 启 worker；禁止在 Insert 冲突后调用。
 func (m *Manager) StartClaimedRun(parent context.Context, record *RunRecord, taskReq bizdev.TaskRunRequest, stageReq bizdev.StageRunRequest, resumeWorkDir string) (*RunRecord, error) {
 	if record == nil || record.RunID == "" {
@@ -303,6 +416,8 @@ func (m *Manager) StartClaimedRun(parent context.Context, record *RunRecord, tas
 	m.runs[record.RunID] = ar
 	m.mu.Unlock()
 	m.writeRunningAnchor(parent, record)
 	if m.pwSvc != nil && record.ProjectID > 0 {
 		startedBy := record.UserID
 		_, _ = m.pwSvc.CreateWorkflowRun(workerCtx, &projectworkflow.WorkflowRun{
@@ -26,6 +26,8 @@ type Store interface {
 	GetByIdempotencyKey(ctx context.Context, key string) (*RunRecord, error)
 	AppendEvent(ctx context.Context, runID string, seq int64, ev stream.StreamEvent) error
 	GetRun(ctx context.Context, runID string) (*RunRecord, error)
 	ListActiveBySession(ctx context.Context, sessionID string) ([]*RunRecord, error)
 	LatestBySession(ctx context.Context, sessionID string) (*RunRecord, error)
 	ListEvents(ctx context.Context, runID string, afterSeq int64) ([]StoredEvent, error)
 	AcquireLease(ctx context.Context, runID, workerID string, ttl time.Duration) error
 	HeartbeatLease(ctx context.Context, runID, workerID string, ttl time.Duration) error
@@ -248,6 +250,108 @@ func (s *PGStore) GetRun(ctx context.Context, runID string) (*RunRecord, error)
 	return &rec, nil
 }
 func (s *PGStore) ListActiveBySession(ctx context.Context, sessionID string) ([]*RunRecord, error) {
 	if !s.Available() {
 		return nil, ErrStoreNotAvailable
 	}
 	if sessionID == "" {
 		return nil, nil
 	}
 	where := `WHERE session_id = $1 AND status IN ('pending', 'running')`
 	if s.hasLeaseColumns(ctx) {
 		where += ` AND (lease_expires_at IS NULL OR lease_expires_at > now())`
 	}
 	where += ` ORDER BY updated_at DESC`
 	return s.queryRuns(ctx, where, sessionID)
 }
 func (s *PGStore) LatestBySession(ctx context.Context, sessionID string) (*RunRecord, error) {
 	if !s.Available() {
 		return nil, ErrStoreNotAvailable
 	}
 	if sessionID == "" {
 		return nil, nil
 	}
 	rows, err := s.queryRuns(ctx, `WHERE session_id = $1 ORDER BY updated_at DESC LIMIT 1`, sessionID)
 	if err != nil {
 		return nil, err
 	}
 	if len(rows) == 0 {
 		return nil, nil
 	}
 	return rows[0], nil
 }
 func (s *PGStore) queryRuns(ctx context.Context, where string, args ...any) ([]*RunRecord, error) {
 	withLease := s.hasLeaseColumns(ctx)
 	q := runSelectSQL(withLease) + " " + where
 	rows, err := s.pool.Query(ctx, q, args...)
 	if err != nil {
 		return nil, err
 	}
 	defer rows.Close()
 	out := make([]*RunRecord, 0)
 	for rows.Next() {
 		rec, err := scanRunRecord(rows, withLease)
 		if err != nil {
 			return nil, err
 		}
 		out = append(out, rec)
 	}
 	return out, rows.Err()
 }
 func runSelectSQL(withLease bool) string {
 	if withLease {
 		return `SELECT run_id, session_id, project_id, user_id, client_turn_id, idempotency_key,
 			task_key, stage, status, workspace_id, assistant_message_id,
 			error_message, worker_id, lease_expires_at, attempt, created_at, updated_at
 			FROM task_runs`
 	}
 	return `SELECT run_id, session_id, project_id, user_id, client_turn_id, idempotency_key,
 		task_key, stage, status, workspace_id, assistant_message_id,
 		error_message, created_at, updated_at
 		FROM task_runs`
 }
 type runScanner interface {
 	Scan(dest ...any) error
 }
 func scanRunRecord(row runScanner, withLease bool) (*RunRecord, error) {
 	var rec RunRecord
 	var projectID, assistantMsgID *int64
 	var idempotencyKey *string
 	var err error
 	if withLease {
 		err = row.Scan(
 			&rec.RunID, &rec.SessionID, &projectID, &rec.UserID, &rec.ClientTurnID, &idempotencyKey,
 			&rec.TaskKey, &rec.Stage, &rec.Status, &rec.WorkspaceID, &assistantMsgID,
 			&rec.ErrorMessage, &rec.WorkerID, &rec.LeaseExpiresAt, &rec.Attempt,
 			&rec.CreatedAt, &rec.UpdatedAt,
 		)
 	} else {
 		err = row.Scan(
 			&rec.RunID, &rec.SessionID, &projectID, &rec.UserID, &rec.ClientTurnID, &idempotencyKey,
 			&rec.TaskKey, &rec.Stage, &rec.Status, &rec.WorkspaceID, &assistantMsgID,
 			&rec.ErrorMessage, &rec.CreatedAt, &rec.UpdatedAt,
 		)
 	}
 	if err != nil {
 		return nil, err
 	}
 	if projectID != nil {
 		rec.ProjectID = *projectID
 	}
 	if assistantMsgID != nil {
 		rec.AssistantMsgID = *assistantMsgID
 	}
 	if idempotencyKey != nil {
 		rec.IdempotencyKey = *idempotencyKey
 	}
 	return &rec, nil
 }
 // GetByIdempotencyKey 按稳定幂等键查找既有 run；无键或未命中返回 (nil, nil)。
 func (s *PGStore) GetByIdempotencyKey(ctx context.Context, key string) (*RunRecord, error) {
 	if !s.Available() {
@@ -302,7 +406,6 @@ func (s *PGStore) ListEvents(ctx context.Context, runID string, afterSeq int64)
 	return out, rows.Err()
 }
 // CancelOpenByProject cancels pending/running runs for a project inside an existing transaction.
 // Table ownership stays with taskrun; workflow terminate must call this instead of writing task_runs directly.
 func CancelOpenByProject(ctx context.Context, tx pgx.Tx, projectID int64, at time.Time, errMsg string) error {
@@ -500,6 +603,12 @@ func (noopStore) AppendEvent(context.Context, string, int64, stream.StreamEvent)
 func (noopStore) GetRun(context.Context, string) (*RunRecord, error) {
 	return nil, ErrStoreNotAvailable
 }
 func (noopStore) ListActiveBySession(context.Context, string) ([]*RunRecord, error) {
 	return nil, ErrStoreNotAvailable
 }
 func (noopStore) LatestBySession(context.Context, string) (*RunRecord, error) {
 	return nil, ErrStoreNotAvailable
 }
 func (noopStore) GetByIdempotencyKey(context.Context, string) (*RunRecord, error) {
 	return nil, ErrStoreNotAvailable
 }
@@ -120,6 +120,7 @@ import { saveFile } from './data/fileStore';
 import { cancelTaskRun, getTaskRun } from './api/taskRuns';
 import { cancelInflightAssistantMessages } from './lib/squadCompletionObservation';
 import { useRecoveredTaskPhases } from './hooks/useRecoveredTaskPhases';
 import { useActiveTaskRecovery } from './hooks/useActiveTaskRecovery';
 export {
  ALL_INITIAL_DELIVERABLES,
@@ -1012,6 +1013,14 @@ export default function App() {
    chatWindow.updateLiveMessage,
  );
  useActiveTaskRecovery(
    authenticatedSessionId,
    chatWindow.phase === 'ready',
    chatWindow.allMessages,
    chatWindow.updateLiveMessage,
    chatWindow.appendLiveMessage,
  );
  // P1: 多维度历史导航与分类统计、日历、引用状态
  const [categoryStats, setCategoryStats] = useState<ChatCategoryStat[]>([]);
  const [dateStats, setDateStats] = useState<ChatDateStat[]>([]);
@@ -230,6 +230,19 @@ export function mapRecordToMessage(r: RawChatMessageRecord): ChatMessage {
      : undefined;
  const msgId = meta.client_id || (r.id ? `msg-${r.id}` : `msg-tmp-${Date.now()}-${Math.random().toString(36).substring(2, 6)}`);
  const phaseSnapshot = parseTaskPhaseSnapshot(meta.task_phase_snapshot);
  const hasRunID = typeof meta.run_id === 'string' && meta.run_id !== '';
  const rawTaskStatus = typeof meta.task_status === 'string' ? meta.task_status : '';
  let messageStatus = mapTaskStatusToMessageStatus(meta.task_status, r.role);
  // 有 run 锚点但还没有终态快照、也没有 task_status 时，不能当成已完成。
  if (
    r.role === 'assistant' &&
    hasRunID &&
    !(phaseSnapshot?.steps && phaseSnapshot.steps.length > 0) &&
    rawTaskStatus === '' &&
    messageStatus === 'completed'
  ) {
    messageStatus = 'streaming_status';
  }
  return {
    id: msgId,
    clientTurnId: meta.client_turn_id || undefined,
@@ -249,7 +262,7 @@ export function mapRecordToMessage(r: RawChatMessageRecord): ChatMessage {
    interactiveCard: meta.interactive_card,
    recommendedCases: meta.recommended_cases,
    quotes: meta.quotes || (meta.quote ? [meta.quote] : undefined),
-    status: mapTaskStatusToMessageStatus(meta.task_status, r.role),
+    status: messageStatus,
    errorInfo: persistedError || legacyBusyError || legacyTaskError,
    taskRequest,
    taskRunId: typeof meta.run_id === 'string' ? meta.run_id : undefined,
@@ -180,6 +180,26 @@ export async function cancelTaskRun(runId: string): Promise<void> {
  }
 }
 export type ActiveTaskRun = {
  run_id: string;
  status: string;
  task_key?: string;
  workspace_id?: string;
  assistant_message_id?: number;
  error_message?: string;
  session_id?: string;
 };
 export async function listActiveTaskRuns(sessionId: string): Promise<ActiveTaskRun[]> {
  const url = `/api/v1/tasks/runs?session_id=${encodeURIComponent(sessionId)}&active=1`;
  const res = await apiFetch(url, { headers: getBearerAuthHeaders() });
  if (!res.ok) {
    throw new Error(`list active task runs failed: HTTP ${res.status}`);
  }
  const body = (await res.json()) as { runs?: ActiveTaskRun[] };
  return Array.isArray(body.runs) ? body.runs : [];
 }
 export async function getTaskRun(runId: string) {
  const res = await apiFetch(`/api/v1/tasks/runs/${encodeURIComponent(runId)}`, {
    headers: getBearerAuthHeaders(),
@@ -359,3 +359,9 @@ test('weighted mode does not let a cross-workflow title hijack an in-scope write
  const fuzzy = resolveTaskTemplateForMessage('编写建议', pool, null, 'main', 'weighted');
  assert.equal(fuzzy?.wrongWorkflow, undefined);
 });
 test('progress question is read-only; explicit start still resolves research', () => {
  assert.equal(hasDocumentWriteIntent('调研报告编写进度怎么样了'), false);
  assert.equal(resolveTaskKeyFromMessage('调研报告编写进度怎么样了'), undefined);
  assert.equal(resolveTaskKeyFromMessage('开始做调研报告'), 'research');
 });
@@ -0,0 +1,114 @@
 import { useEffect } from 'react';
 import { listActiveTaskRuns, type ActiveTaskRun } from '../api/taskRuns';
 import {
  placeholderFromActiveTask,
  readActiveTasks,
  removeActiveTask,
  upsertActiveTask,
  type ActiveTaskCacheEntry,
 } from '../lib/activeTaskCache';
 import type { ChatMessage } from '../types/workbench';
 const FRESH_MS = 20_000;
 /**
 * 刷新后先用本地进行中缓存画出进度卡片，再用会话活跃 run 列表收敛。
 * 服务端没有的旧缓存会被清掉；刚写入、列表还没跟上的条目保留一小段时间。
 */
 export function useActiveTaskRecovery(
  sessionId: string,
  windowReady: boolean,
  messages: ChatMessage[],
  updateMessage: (id: string, patch: Partial<ChatMessage>) => void,
  appendMessage: (message: ChatMessage) => void,
 ) {
  useEffect(() => {
    if (!sessionId || !windowReady) return;
    let cancelled = false;
    const cached = readActiveTasks(sessionId);
    for (const entry of cached) {
      if (messages.some((message) => message.taskRunId === entry.runId)) continue;
      const byClient = messages.find(
        (message) => message.id === entry.assistantClientId || message.clientMessageId === entry.assistantClientId,
      );
      if (byClient) {
        updateMessage(byClient.id, {
          taskRunId: entry.runId,
          status: 'streaming_status',
          livePhaseLabel: entry.lastPhaseLabel || byClient.livePhaseLabel || '任务执行中…',
          serverId: byClient.serverId || entry.serverMessageId,
        });
        continue;
      }
      appendMessage(placeholderFromActiveTask(entry));
    }
    void listActiveTaskRuns(sessionId)
      .then((runs) => {
        if (cancelled) return;
        const activeIDs = new Set(runs.map((run) => run.run_id).filter(Boolean));
        for (const run of runs) {
          applyServerRun(sessionId, run, messages, updateMessage, appendMessage);
        }
        for (const entry of readActiveTasks(sessionId)) {
          if (activeIDs.has(entry.runId)) continue;
          const age = Date.now() - Date.parse(entry.startedAt);
          if (Number.isFinite(age) && age >= 0 && age < FRESH_MS) continue;
          removeActiveTask(entry.runId);
        }
      })
      .catch((err) => {
        console.warn('active task sync failed', err);
      });
    return () => {
      cancelled = true;
    };
    // 窗口 ready 后再画本地缓存并拉活跃 run，避免被随后的窗口替换清掉。
    // eslint-disable-next-line react-hooks/exhaustive-deps
  }, [sessionId, windowReady]);
 }
 function applyServerRun(
  sessionId: string,
  run: ActiveTaskRun,
  messages: ChatMessage[],
  updateMessage: (id: string, patch: Partial<ChatMessage>) => void,
  appendMessage: (message: ChatMessage) => void,
 ) {
  if (!run.run_id) return;
  const existing =
    messages.find((message) => message.taskRunId === run.run_id) ||
    messages.find((message) => run.assistant_message_id && message.serverId === run.assistant_message_id);
  const label = existing?.livePhaseLabel || '任务执行中…';
  if (existing) {
    updateMessage(existing.id, {
      taskRunId: run.run_id,
      status: existing.status === 'completed' ? existing.status : 'streaming_status',
      taskWorkspaceId: run.workspace_id || existing.taskWorkspaceId,
      serverId: existing.serverId || run.assistant_message_id,
      livePhaseLabel: label,
    });
  } else {
    const entry: ActiveTaskCacheEntry = {
      sessionId,
      runId: run.run_id,
      assistantClientId: run.assistant_message_id ? `msg-${run.assistant_message_id}` : `ast-active-${run.run_id}`,
      taskKey: run.task_key || '',
      startedAt: new Date().toISOString(),
      lastPhaseLabel: label,
      serverMessageId: run.assistant_message_id,
    };
    appendMessage(placeholderFromActiveTask(entry));
  }
  const previous = readActiveTasks(sessionId).find((row) => row.runId === run.run_id);
  upsertActiveTask({
    sessionId,
    runId: run.run_id,
    assistantClientId: previous?.assistantClientId || existing?.id || `ast-active-${run.run_id}`,
    taskKey: run.task_key || previous?.taskKey || '',
    startedAt: previous?.startedAt || new Date().toISOString(),
    lastPhaseLabel: previous?.lastPhaseLabel || label,
    serverMessageId: run.assistant_message_id || previous?.serverMessageId,
  });
 }
@@ -165,3 +165,18 @@ test('mapRecordToMessage: restores task phase snapshot into card fields', () =>
  assert.equal(msg.phaseSnapshot?.title, '用户初始需求收集表');
  assert.equal(msg.phaseEvents?.[1].detail, '门禁');
 });
 test('mapRecordToMessage: run anchor without task_status stays streaming', () => {
  const msg = mapRecordToMessage({
    id: 42,
    session_id: 'sess-1',
    user_id: 1,
    role: 'assistant',
    content: '任务执行中…',
    metadata: { run_id: 'run-live', task_key: 'research' },
    day_key: '2026-09-28',
    created_at: '2026-09-28T03:27:00Z',
  });
  assert.equal(msg.taskRunId, 'run-live');
  assert.equal(msg.status, 'streaming_status');
 });
@@ -0,0 +1,18 @@
 import assert from 'node:assert/strict';
 import test from 'node:test';
 import {
  replayAttemptLimit,
  replayBackoffMs,
  shouldAbandonTaskReplay,
 } from './useRecoveredTaskPhases';
 test('replay retries 502 and stops on auth or not-found', () => {
  assert.equal(replayAttemptLimit(false), 4);
  assert.equal(replayAttemptLimit(true), 2);
  assert.equal(replayBackoffMs(0), 500);
  assert.equal(replayBackoffMs(20), 5000);
  assert.equal(shouldAbandonTaskReplay(502, 0, 4), false);
  assert.equal(shouldAbandonTaskReplay(502, 3, 4), true);
  assert.equal(shouldAbandonTaskReplay(401, 0, 4), true);
  assert.equal(shouldAbandonTaskReplay(404, 0, 4), true);
 });
@@ -1,6 +1,6 @@
 import { useEffect, useRef } from 'react';
 import type { StreamEvent } from '../api/chatStream';
-import { readTaskRunSSEOnce, subscribeTaskRunEvents } from '../api/taskRuns';
+import { readTaskRunSSEOnce, subscribeTaskRunEvents, TaskRunHttpError } from '../api/taskRuns';
 import { applyPhaseEvent } from '../engine/phaseEventMapper';
 import {
  messageNeedsPhaseReplay,
@@ -80,15 +80,38 @@ async function replayRun(
    cache.set(runId, { events: strip, label });
    updateMessage(messageId, { phaseEvents: strip, livePhaseLabel: label });
  };
-  try {
+  const attempts = replayAttemptLimit(live);
-    if (live) {
+  for (let attempt = 0; attempt < attempts; attempt++) {
-      await subscribeTaskRunEvents(runId, { onEvent }, signal);
+    if (signal.aborted) return;
    try {
      if (live) {
        await subscribeTaskRunEvents(runId, { onEvent }, signal);
        return;
      }
      await readTaskRunSSEOnce(runId, 0, onEvent, signal, { replayOnly: true });
      return;
-    }
+    } catch (err) {
-    await readTaskRunSSEOnce(runId, 0, onEvent, signal, { replayOnly: true });
+      if (signal.aborted) return;
-  } catch (err) {
+      const status = err instanceof TaskRunHttpError ? err.status : 0;
-    if (!signal.aborted) {
+      if (shouldAbandonTaskReplay(status, attempt, attempts)) {
-      console.warn('task phase replay failed', runId, err);
+        console.warn('task phase replay failed', runId, err);
        return;
      }
      await new Promise((resolve) => setTimeout(resolve, replayBackoffMs(attempt)));
    }
  }
 }
 /** 终态重放多试几次；进行中订阅内部已有重试，外层只再补一次。 */
 export function replayAttemptLimit(live: boolean): number {
  return live ? 2 : 4;
 }
 export function replayBackoffMs(attempt: number): number {
  return Math.min(500 * (attempt + 1), 5000);
 }
 export function shouldAbandonTaskReplay(status: number, attempt: number, attempts: number): boolean {
  if (status === 401 || status === 404) return true;
  return attempt >= attempts - 1;
 }
@@ -9,6 +9,7 @@ import { fetchChatTurnStatus } from '../api/chatTurn';
 import { mapTaskStatusToMessageStatus } from '../api/chatWindow';
 import { listProjectWorkflowDocuments } from '../api/projectDocuments';
 import { createTaskRun, getTaskRun, resumeTaskRun, subscribeTaskRunEvents } from '../api/taskRuns';
 import { readActiveTasks, removeActiveTask, touchActiveTaskPhase, upsertActiveTask } from '../lib/activeTaskCache';
 import { buildTaskRequest, streamTask } from '../api/taskStream';
 import { applyPhaseEvent } from '../engine/phaseEventMapper';
 import { resolveTaskTemplateForMessage, wrongWorkflowGuidance } from '../engine/templateMatch';
@@ -371,6 +372,24 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
    let accumulated = '';
    let settled = false;
    const rememberRun = (runId: string, label?: string, serverMessageId?: number) => {
      if (!sessionId || !runId) return;
      const previous = readActiveTasks(sessionId).find((row) => row.runId === runId);
      upsertActiveTask({
        sessionId,
        runId,
        assistantClientId: asstId,
        taskKey: task.taskKey || previous?.taskKey || '',
        startedAt: previous?.startedAt || new Date().toISOString(),
        lastPhaseLabel: label || previous?.lastPhaseLabel,
        serverMessageId: serverMessageId || previous?.serverMessageId,
      });
    };
    const forgetRun = (runId?: string) => {
      const id = runId || retainedRunId;
      if (id) removeActiveTask(id);
    };
    const historyOnly = !messages.some((m) => m.id === asstId);
    let historySnapshot = historyOnly
      ? chatWindow.allMessages.find((m) => m.id === asstId)
@@ -412,6 +431,7 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
        scheduleRetry(delayMs);
        return;
      }
      forgetRun();
      patchAsst({ status: 'failed', errorInfo: info, livePhaseLabel: undefined });
      taskRunStream.setThinking(sessionId, false);
    };
@@ -472,9 +492,11 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
      if (controller.signal.aborted || settled) return;
      switch (ev.type) {
        case 'phase': {
          const phaseLabel = ev.phase?.message;
          if (retainedRunId) touchActiveTaskPhase(retainedRunId, phaseLabel);
          patchAsst((m) => ({
            status: 'streaming_status',
-            livePhaseLabel: ev.phase?.message || m.livePhaseLabel,
+            livePhaseLabel: phaseLabel || m.livePhaseLabel,
            phaseEvents: applyPhaseEvent(opts?.phaseEvents || m.phaseEvents || [], ev.phase),
          }));
          break;
@@ -542,6 +564,7 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
        case 'done': {
          settled = true;
          const finishedRunId = retainedRunId || taskRunStream.activeTaskRunId;
          forgetRun(finishedRunId || undefined);
          taskRunStream.setRunId(sessionId, null);
          taskRunStream.setThinking(sessionId, false);
          void (async () => {
@@ -583,6 +606,7 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
      const view = readRemoteTaskRun(remote);
      if (view.status === 'quality_failed') {
        settled = true;
        forgetRun(runId);
        patchAsst({
          status: 'quality_failed',
          livePhaseLabel: undefined,
@@ -594,6 +618,7 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
      }
      if (isBlockedGovernanceStatus(view.status)) {
        settled = true;
        forgetRun(runId);
        patchAsst({
          status: 'blocked_pending_governance',
          content: GOVERNANCE_BLOCKED_HINT,
@@ -606,6 +631,7 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
      }
      if (view.status === 'completed' || view.status === 'pending_author_confirmation') {
        settled = true;
        forgetRun(runId);
        const mapped = mapTaskStatusToMessageStatus(view.status, 'assistant') || 'completed';
        patchAsst({
          status: mapped,
@@ -624,6 +650,7 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
      }
      if (view.status === 'interrupted') {
        settled = true;
        forgetRun(runId);
        patchAsst({
          status: 'failed',
          livePhaseLabel: undefined,
@@ -693,6 +720,7 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
        }
        const resumed = await resumeTaskRun(runId);
        retainedRunId = resumed.run_id || runId;
        rememberRun(retainedRunId, '正在继续任务…');
        taskRunStream.setRunId(sessionId, retainedRunId);
        taskRunStream.setThinking(sessionId, true);
        patchAsst({
@@ -720,6 +748,7 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
          const view = readRemoteTaskRun(remote);
          if (remote && shouldSubscribeExistingRun(view.status)) {
            retainedRunId = existingRunId;
            rememberRun(existingRunId, '正在重连任务…');
            taskRunStream.setRunId(sessionId, existingRunId);
            patchAsst({ taskRunId: existingRunId, livePhaseLabel: '正在重连任务…' });
            await subscribeWithGapHeal(existingRunId);
@@ -738,6 +767,7 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
                  ? buildInterruptedTaskErrorInfo(view.errorMessage, view.workspaceId)
                  : undefined,
            });
            forgetRun(existingRunId);
            taskRunStream.setThinking(sessionId, false);
            if (mapped === 'completed' || mapped === 'pending_author_confirmation') {
              await syncControlledDocsAfterTask(task.taskKey || 'intake');
@@ -760,6 +790,7 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
          typeof created.assistant_msg_id === 'number' && created.assistant_msg_id > 0
            ? created.assistant_msg_id
            : 0;
        rememberRun(created.run_id, '任务执行中…', assistantMsgId > 0 ? assistantMsgId : undefined);
        if (assistantMsgId > 0) {
          chatWindow.replaceMessageServerId(asstId, assistantMsgId);
          patchAsst({ taskRunId: created.run_id, serverId: assistantMsgId });
@@ -774,6 +805,7 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
            livePhaseLabel: undefined,
            taskRunId: created.run_id,
          });
          forgetRun(created.run_id);
          taskRunStream.setRunId(sessionId, null);
          taskRunStream.setThinking(sessionId, false);
          await chatWindow.refresh().catch(() => undefined);
@@ -789,6 +821,7 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
    void startObserver().catch(async (_err) => {
      if (settled) return;
      if (controller.signal.aborted) {
        forgetRun();
        patchAsst({ status: 'cancelled', livePhaseLabel: undefined });
        taskRunStream.setThinking(sessionId, false);
        return;
@@ -805,6 +838,7 @@ export function useTaskRunAttempts(input: TaskRunAttemptsInput) {
            // 原 run 仍在跑：重订阅，不 POST 新 run
            settled = false;
            retainedRunId = runId;
            rememberRun(runId, '正在重连任务…');
            patchAsst({ livePhaseLabel: '正在重连任务…', taskRunId: runId });
            await subscribeWithGapHeal(runId);
            return;
@@ -0,0 +1,44 @@
 import assert from 'node:assert/strict';
 import test from 'node:test';
 import {
  readActiveTasks,
  removeActiveTask,
  touchActiveTaskPhase,
  upsertActiveTask,
 } from './activeTaskCache';
 const mem = new Map<string, string>();
 test('active task cache keeps an in-progress run and drops it when finished', () => {
  (globalThis as { window?: unknown }).window = {
    localStorage: {
      getItem: (key: string) => mem.get(key) ?? null,
      setItem: (key: string, value: string) => {
        mem.set(key, value);
      },
    },
  };
  mem.clear();
  upsertActiveTask({
    sessionId: 'sess-1',
    runId: 'run-1',
    assistantClientId: 'turn:assistant',
    taskKey: 'research',
    startedAt: '2026-09-28T03:27:00Z',
  });
  upsertActiveTask({
    sessionId: 'sess-2',
    runId: 'run-2',
    assistantClientId: 'other',
    taskKey: 'intake',
    startedAt: '2026-09-28T03:28:00Z',
  });
  touchActiveTaskPhase('run-1', '正在撰写调研报告');
  const mine = readActiveTasks('sess-1');
  assert.equal(mine.length, 1);
  assert.equal(mine[0].runId, 'run-1');
  assert.equal(mine[0].lastPhaseLabel, '正在撰写调研报告');
  removeActiveTask('run-1');
  assert.equal(readActiveTasks('sess-1').length, 0);
  assert.equal(readActiveTasks('sess-2').length, 1);
 });
@@ -0,0 +1,95 @@
 import type { ChatMessage } from '../types/workbench';
 /** 进行中任务锚点。窗口缓存只存服务端已落库消息，刷新会丢掉还没写回的进度。 */
 const STORAGE_KEY = 'smdf_agent_workbench_v6_active_tasks';
 export type ActiveTaskCacheEntry = {
  sessionId: string;
  runId: string;
  assistantClientId: string;
  taskKey: string;
  startedAt: string;
  lastPhaseLabel?: string;
  serverMessageId?: number;
 };
 function readAll(): ActiveTaskCacheEntry[] {
  if (typeof window === 'undefined' || !window.localStorage) return [];
  try {
    const raw = window.localStorage.getItem(STORAGE_KEY);
    if (!raw) return [];
    const parsed = JSON.parse(raw) as unknown;
    if (!Array.isArray(parsed)) return [];
    return parsed.filter(isEntry);
  } catch {
    return [];
  }
 }
 function writeAll(entries: ActiveTaskCacheEntry[]): void {
  if (typeof window === 'undefined' || !window.localStorage) return;
  try {
    window.localStorage.setItem(STORAGE_KEY, JSON.stringify(entries));
  } catch {
    // 配额不足时放弃缓存，服务端锚点仍可恢复。
  }
 }
 function isEntry(value: unknown): value is ActiveTaskCacheEntry {
  if (!value || typeof value !== 'object') return false;
  const row = value as Partial<ActiveTaskCacheEntry>;
  return typeof row.sessionId === 'string' && typeof row.runId === 'string' && row.runId !== '';
 }
 export function readActiveTasks(sessionId: string): ActiveTaskCacheEntry[] {
  if (!sessionId) return [];
  return readAll().filter((entry) => entry.sessionId === sessionId);
 }
 export function upsertActiveTask(entry: ActiveTaskCacheEntry): void {
  if (!entry.sessionId || !entry.runId) return;
  const rest = readAll().filter((row) => row.runId !== entry.runId);
  writeAll([...rest, entry]);
 }
 export function touchActiveTaskPhase(runId: string, label: string | undefined): void {
  if (!runId || !label) return;
  const all = readAll();
  const hit = all.find((row) => row.runId === runId);
  if (!hit || hit.lastPhaseLabel === label) return;
  hit.lastPhaseLabel = label;
  writeAll(all);
 }
 export function removeActiveTask(runId: string): void {
  if (!runId) return;
  writeAll(readAll().filter((row) => row.runId !== runId));
 }
 export function placeholderFromActiveTask(entry: ActiveTaskCacheEntry): ChatMessage {
  return {
    id: entry.assistantClientId || `ast-active-${entry.runId}`,
    clientMessageId: entry.assistantClientId || undefined,
    serverId: entry.serverMessageId,
    sessionId: entry.sessionId,
    role: 'assistant',
    authorName: '项目助手',
    content: '任务执行中…',
    timestamp: new Date().toTimeString().substring(0, 5),
    status: 'streaming_status',
    livePhaseLabel: entry.lastPhaseLabel || '任务执行中…',
    taskRunId: entry.runId,
    taskRequest: entry.taskKey
      ? {
          taskKey: entry.taskKey,
          projectName: '',
          customerName: '',
          deviceType: '',
          sessionId: entry.sessionId,
          userId: 0,
          projectId: 0,
          clientTurnId: '',
        }
      : undefined,
  };
 }