Skip to content

[Feature]: Run 流水视图:一屏看清节点状态、耗时、Token、产物与失败原因 #1197

Open

Description

@linus-liu-web

Problem

真实项目(secops-delivery SOP,assetgraph 智能资产图谱)跑了两轮:14 个节点全部 succeeded,但想知道「每个节点做了什么、花了多少 token、产物在哪、为什么重试」需要分别翻节点卡片、产物列表、workspace 文件和平台日志——页面没有一个能一屏看懂整条流水的地方。

叠加已有碎片问题,体验更割裂:

数据其实是有的:run_nodes(状态/attempt/activity/error)、run_events、run_telemetry(usage)、run_artifacts(大小/类型/内容)都在 MySQL 里,主要缺一层前端聚合。正在进行的 #1183 是 runs/gates/chat/console UX 修复批次,本提案是其中「Run 流水视图」的信息架构 umbrella;#1191/#1101/#1102/#660 是具体点,本 issue 不重复它们。#1186(运维诊断页)解决的是「部署是否健康」,本 issue 解决的是「这次 run 发生了什么」,两者互补。

Proposed behavior

在运行详情页 / 对话内 run 区域提供一个聚合的流水视图:

  1. 顶部 summary:状态、总耗时、总 token、触发人/需求、交付分支与产物数。
  2. 节点流水:每个节点的状态、耗时、重试次数、token、产物数,加一句实时 activity(流式更新,而不是只有「运行中」)。
  3. 展开节点:真实错误(stderr / 门禁 errors / 结构化 failureDetail)、产物(大小、类型、预览、下载)、工具调用摘要(不是 raw 日志)。
  4. 事件时间线:node / artifact / error / approval 事件合并,可按节点或级别过滤。
  5. 失败引导:明确告诉用户该「重试」「修改上游产物后重跑」还是「看日志」,并附原因。
  6. 数据来源用现有的 run_nodes / run_events / run_telemetry / run_artifacts,不新增表;主要是前端聚合,必要时补少量 API 字段。

验收(建议):一个刚跑完的 7 节点 SOP,不打开 workspace/日志/数据库,就能从页面回答「哪些节点失败/重试、每步耗时与 token、产物在哪、失败原因是什么」。

Scope

Platform — UI / integrations

Alternatives considered

Checks

  • This proposal does not require organization-specific infrastructure or credentials.
  • I searched existing issues and discussions for similar proposals.

Activity

  1. cocofhu00 commented on Nov 21, 2026

    @cocofhu00
    Contributor

    感谢反馈。

    我理解的需求是:希望在运行详情里有一个聚合的流水视图,能在一屏里看到每个节点的状态、耗时、Token、产物和失败原因,不用分别翻节点卡片、产物和日志。

    这条 Issue 已经看到,会按现有排期查看。

  2. linus-liu-web commented on Nov 21, 2026

    @linus-liu-web
    Author

    Implemented as #1198: a read-only 流水 (pipeline) tab on the run detail page.

    It aggregates the data the run endpoint already returns into one place:

    • summary: run status, completed/failed node counts, wall time, total tokens, tool calls, retries/rework, artifact count + size;
    • node rows in workflow order: status, duration, attempts/rework, tokens, tool calls, artifact count + size, plus current activity or the failed node's real error;
    • expandable per-node detail: full error, activity, artifacts (opens the existing preview) and truncated output;
    • run event timeline (last 40 events).

    Frontend-only (6 files, +385/-4): no API/schema/engine change. Failed-node detail improves automatically once #1191 lands (backend capturing the real script/gate error). Tests: go test ./web -run TestRunPipelinePane, full go test ./web, go vet ./web, check-public.py pass.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions