跳到主内容
智客 ZICQ

技能库 智客分类:Agent 工作流 adk-dev-console

Adk Dev Console

解释 ADK Dev 控制台 — 每个标签显示的内容, 如何读取代理步骤、 痕迹、 多代理仪表板、代理切换、 控制台模式, 以及 adk dev 期间在 localhost: 301 显示的其他 UI 特性

1544 安装量

官方网址:skills.sh

技能介绍

先看中文介绍;官方 description 原文单独保留,不改写 SKILL.md。

做什么

解释 ADK Dev 控制台 — 每个标签显示的内容, 如何读取代理步骤、 痕迹、 多代理仪表板、代理切换、 控制台模式, 以及 adk dev 期间在 localhost: 301 显示的其他 UI 特性

何时用

这个技能

代理如何加载

按 Agent Skills 渐进披露:启动时只加载 name 与 description(约 100 token);任务匹配后才读入整份 SKILL.md 正文;scripts/、references/、assets/ 仅在需要时再读。 本文件正文结构:ADK Dev Console、When to Use This Skill、Available Documentation、How to Answer、Quick Reference、Tab Groups。 其中含规范建议的小节:分步指令。

文件分析

文件分析:除 SKILL.md 外,正文引用了 references/agent-steps.md、references/pages.md、references/component-registry.md、references/multi-agent-dashboard.md,属于带资源的技能包,这些文件按需再读。

官方 description(原文)

Explains the ADK Dev Console — what each tab shows, how to read Agent Steps, traces, multi-agent dashboard, agent switching, console modes, and other UI features visible at localhost:3001 during adk dev

ADK Dev ConsoleWhen to Use This SkillAvailable DocumentationHow to AnswerQuick ReferenceTab GroupsAgent(0) Task DockMulti-Agent SidebarAgent Steps (Chat Page)Key URLsMulti-Agent CLI Commands

· 许可:MIT

来源分类:skills.sh agent-skill

SKILL.md 与 Agent 调用

官方规范 ↗
name
adk-dev-console
description
Explains the ADK Dev Console — what each tab shows, how to read Agent Steps, traces, multi-agent dashboard, agent switching, console modes, and other UI features visible at localhost:3001 during adk dev
许可
MIT
  1. 发现技能客户端向 Agent 提供名称与描述目录。
  2. 匹配与调用用户指定或任务匹配后,载入 SKILL.md 指令。
  3. 按需加载按步骤读取参考文档、使用脚本与素材。
指令中引用的文件 · 4
  • references/agent-steps.md
  • references/pages.md
  • references/component-registry.md
  • references/multi-agent-dashboard.md

以下路径提取自原文;文件是否齐全请以来源仓库中的完整目录为准。

具体调用语法与可用工具以目标 Agent 客户端为准。 查看调用机制说明 ↗

安装这个技能

Skills CLI ↗

先选择目标 Agent 和安装范围,保留技能包的附属文件,安装后检查客户端能否发现该技能。

该技能引用了附属文件,请从来源获取完整目录;仅复制 SKILL.md 可能缺少依赖。

交给 Agent 安装

复制安装指令给支持 Agent Skills 的代理,确认其中的目标目录与客户端匹配。

把 Agent Skill「adk-dev-console」安装到我的项目:SKILL.md 原文与官方 description 见 https://zicq.com/zh/skills/skl-9a2c07338c474922-%E9%98%BF%E5%BE%B7%E5%85%8B%C2%B7%E5%BE%B7%E5%A4%AB%E6%8E%A7%E5%88%B6%E5%8F%B0.html
请存为 .cursor/skills/adk-dev-console/SKILL.md 或 .claude/skills/adk-dev-console/SKILL.md,frontmatter 的 name 与 description 保持原样,不要改写。
该技能还带 scripts/、references/、assets/ 等文件,请从 https://github.com/botpress/skills 取完整目录,不要只建一个 SKILL.md。

GitHub 完整包 ↗

终端安装 · Skills CLI

需要 Node.js 与 npx。先查看仓库技能列表,确认实际名称。

npx skills add 'https://github.com/botpress/skills' --list

npx skills add 'https://github.com/botpress/skills' --skill 'adk-dev-console'

CLI 会交互选择目标 Agent,默认安装到项目;用户级安装使用 -g。先通过查看命令核对仓库内容,再用 npx skills list 检查已安装技能。

阅读排版

name: adk-dev-console description: Explains the ADK Dev Console — what each tab shows, how to read Agent Steps, traces, multi-agent dashboard, agent switching, console modes, and other UI features visible at localhost:3001 during adk dev license: MIT

ADK Dev Console

The Dev Console is a local app served at port 3001 (by default, but can be customized) during adk dev. It gives developers real-time visibility into their agent's behavior — conversations, execution traces, data, integrations, and configuration.

The Dev Console is a singleton shared dashboard: one UI server per user, with multiple adk dev agents registering and deregistering dynamically. Developers can switch between running agents in the sidebar, toggle between dev and prod targets, and connect to deployed Cloud bots — all from the same browser tab.

When to Use This Skill

Activate when users ask about:

  • UI concepts — "What are Agent Steps?", "What does the Observe tab show?", "What is a turn?", "What is the component registry?", "What is the Agent Map?"
  • Agent(0) UI — "What are the Tasks under Agent(0)?", "Where did the todo list come from?", "Why is Agent(0) showing task progress?"
  • Dev Console navigation — "What tabs are available?", "Where do I find traces?", "How do I test RAG?"
  • Reading execution data — "What do the steps mean?", "Why is my step red?", "What's the cost shown?"
  • Component registry — "Where do I find custom components?", "How do I install a component?", "What components are available?"
  • Multi-agent dashboard — "How do I switch between agents?", "How do I run multiple agents?", "What does the agent selector show?", "How do I see which agents are running?"
  • Console modes and targets — "What's the difference between dev and prod?", "How do I connect to a Cloud bot?", "What is Cloud Dev Console?"
  • Agent management CLI — "How do I list running agents?", "How do I stop an agent?", "What does adk ps show?", "How do I open the dashboard?"
  • Specific pages — "How do I use the evals page?", "Where do I configure integrations?", "How do I search knowledge?"
  • Comparing UI vs CLI — "Should I use the Dev Console or CLI for debugging?"
  • Mentions of localhost:3001, "dev console", "Dev Console", "component registry", "agent selector", "agent picker", "agent map", "Cloud Dev Console", or specific tab names (Chat, Build, Components, Data, Test, Observe, Config)

Available Documentation

| File | Contents | | ------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | | references/agent-steps.md | Agent Steps visualization — turns, iterations, tools, messages, state mutations, cost tracking, status indicators | | references/pages.md | Every page/tab in the Dev Console — what it shows, key features, layout | | references/component-registry.md | Component Registry — installed vs registry tabs, component lifecycle, runtime registry internals, UI features | | references/multi-agent-dashboard.md | Multi-agent architecture, agent selector, console modes (local/cloud), dev/prod targets, CLI commands (adk ps, adk kill, adk dashboard, adk status) |

How to Answer

Match depth to the question:

  • "What is X?" (e.g., "What are Agent Steps?") → One sentence definition + what the user sees. Don't dump the full data model.
  • "Where do I find X?" → Name the tab group and page, give the URL path.
  • "How do I read X?" → Explain the visual hierarchy and what each element means.
  • "What does this mean?" (pointing at something in the UI) → Identify the component, explain its meaning.

Quick Reference

Tab Groups

| Group | Pages | Purpose | | -------------- | ------------------------------------------------ | ------------------------------------------------------------------------------------------------ | | Chat | Chat | Test the agent via webchat + see Agent Steps | | Build | Agent Map | Interactive agent architecture graph (experimental, feature-flagged behind enable_agent_forge) | | Components | Webchat Components, Actions, Workflows, Triggers | Browse component registry, test bot primitives | | Test | RAG Search, Evals | Test knowledge search and run automated evals | | Data | Knowledge, Tables, Files | Manage knowledge bases, tables, and files | | Observe | Conversations, Traces, Logs | View conversation history, execution traces, runtime logs | | Config | Settings, Integrations | Agent config, secrets, LLM settings, integration management |

Agent(0) Task Dock

When Agent(0) uses its todo tool, the panel shows a compact Tasks dock with completed/total count, the active task, and an expandable task list. Status icons map to pending, in progress, completed, and cancelled; priorities are shown as low/medium/high chips.

Multi-Agent Sidebar

The sidebar header shows the currently selected agent with a status indicator and mode pill (dev/prod/cloud). Clicking it opens the Agent Selector dropdown:

  • Active agents — all running local agents with status dot, name, project path, and a close button
  • Cloud Dev Console — switch to a deployed Botpress Cloud bot (workspace + bot picker)
  • Recent projects — previously opened projects not currently running
  • Footer actions — Create new project, Open existing, Switch environment (dev↔prod), About

Agent Steps (Chat Page)

The right panel of the Chat page shows Agent Steps — a real-time visualization of what the agent did to process each message.

Hierarchy: Turn → Iterations → Tools / Messages / State Mutations

  • Turn = one conversation exchange (user message → agent processing → response)
  • Iteration = one loop of the autonomous agent (think → decide → act)
  • Tool = a tool call within an iteration (violet card)
  • Message = a bot message sent (blue card)
  • State Mutation = a state change (teal card, shows before/after)

Status indicators: ✓ green = ok, ✗ red = error, ⟳ blue spinning = running

AI metrics per iteration: model name, input/output tokens, cost (USD)

Key URLs

| Path | Page | | ---------------- | --------------------------------------------------------- | | /chat | Chat + Agent Steps | | /agent-map | Agent Map — interactive architecture graph (experimental) | | /components | Component registry + installed components | | /actions | Actions browser | | /workflows | Workflows + run history | | /search | RAG search testing | | /evals | Eval definitions + runs | | /knowledge | Knowledge base management | | /tables | Table data management | | /traces | Full trace viewer | | /conversations | Conversation history | | /logs | Runtime logs | | /settings | Agent configuration | | /integrations | Integration management |

Multi-Agent CLI Commands

| Command | Purpose | | --------------- | -------------------------------------------------------------------------------- | | adk ps | List running agents and processes with PIDs, ports, uptime (supports --watch) | | adk dashboard | Open the DevConsole in standalone mode (no agent required) | | adk kill | Stop agents or the entire DevConsole (--all, --current, --pid, or by name) | | adk status | Show project health and status info |

相关技能

Agent 工作流

Skill Creator

创造有效技能指南。 当用户想创造出新的技能(或更新现有的技能),以专业知识,工作流程,或工具集成来扩展克洛德的能力时,应该使用这种技能.

Agent 工作流

Clawdhub

使用ClawdHub CLI搜索,安装,更新并发布从taladhub.com的代理技能. 需要获取苍蝇上的新技能时使用,将安装的技能同步到最新版本或特定版本,或者发布 npm-instainddhub CLI 的新/更新的技能文件夹.

Agent 工作流

Agent Team Orchestration

管弦乐团多代理团队,任务设定周期,交接协议,审查工作流程. 使用时间: (1)建立2+特派员队伍,具有不同专业,(2)确定任务路线和生命周期(收录框_ spec_建设_审查_完成),(3)在特派员之间制定交接协议,(4)建立审查和质量关口,(5)管理特派员之间的交流和文物共享.

Agent 工作流

Superpowers

Spec-first,TDD,子代理驱动的软件开发工作流程. 当:(1)构建任何新功能或应用——触发脑暴_计划_子代理执行回路,(2)调试出一个bug或测试失败——触发系统性的根起过程,(3)用户说"让我们构建","帮助我计划","我想添加X",或"这个被打破",(4)完成一个功…