Agent

Knowledge Base Management

AE Multi-Agent Collaboration

Manage AI Agent Teams (create/update/delete/AI-generate config) and run TeamRuns (start/chat/cancel/reply/poll results/fetch artifacts) through ae-cli, upgrading multi-agent collaboration from manual orchestration to command-line-driven.

IndustriesAll industries

Agent · Skills Live

skill · loadedindustry+100

Run this Skill on my data

Agent orchestration

multi-agent

Analysis readyevidence attached

The AE Multi-Agent Collaboration Skill is a multi-agent team orchestration and execution capability in ThinkingAI's industry Skill library. It helps teams manage AI Agent Teams (list/create/update/delete/AI-generate config/browse templates) and run TeamRuns (start/chat/cancel/reply/poll results/fetch artifacts) through the ae-cli command-line tool, upgrading from manually orchestrating multi-agent collaboration flows on the web to an efficient one-command launch of multi-agent tasks with results fetched automatically. This is ThinkingAI's decade of industry experience, packaged into ready-to-use Skills.

Live walkthrough

AgentSave as AgentExport session

ae-agent-team-management

I want to build an Agent team that can automatically run 'monthly retention analysis + optimization recommendations,' but I genuinely can't manage writing that Team config JSON. Generate the config, build the team, then run this month's right away.

ae-cli team +ai-generate with the goal description, and AI generates a draft team config

Generated a five-role config: 1 Leader + 4 members

Human review of the draft: correct the knowledge-base bindings and the data-tool authorization scope

+create the team 'Retention Analysis Group'

Created successfully, with zero hand-written JSON

+run-start to launch this month's retention-analysis TeamRun, +run-watch to stream via SSE

Status waiting_user: read the pendingQuestion and use +run-reply to answer with the '7-day retention' definition

The run resumed automatically

+run-result to get the conclusion, +run-artifacts to download the artifacts

Status completed, 2 artifacts

The team is built and the first run is complete. From generating the config to getting the analysis report took 15 minutes, and that Team config JSON that was giving you a headache, you never had to touch it.

1. Team roles (1 Leader + 4 members)

Role	Agent	Responsibility	Status
Leader	Retention Analysis Leader	Break down the task, dispatch members, compile the final conclusion	First run complete
Member	Analytics Agent	Pull retention data, drill down by segment, locate the source of anomalies	First run complete
Member	Forecasting Agent	Fit the retention curve and extrapolate 30 days out	First run complete
Member	Strategy Agent	Draft optimization recommendations from the analysis and forecast	First run complete
Member	Review Agent	Validate the number definitions and feasibility of recommendations, and block unreliable conclusions	First run complete

← Scroll to see more

2. TeamRun execution summary

Monitored via SSE throughout, moving through five states: pending → running → waiting_user → running → completed

waiting_user paused at one point, 'confirm the retention definition': answered with the 7-day retention definition to fit your hardcore genre, and the run resumed automatically

The Analytics Agent located this month's 7-day retention at 31% → 27%, with the decline concentrated in channel B's new users; the Forecasting Agent extrapolates a slide toward 25% next month without intervention

The Strategy Agent produced 3 intervention recommendations, and the Review Agent validated the definitions before producing the final version

2 artifacts downloaded: 'This Month's Retention Analysis Report' and 'Optimization Recommendations List'

Recommendation

The 'Retention Analysis Group' is now a reusable asset: next month a single +run-start reruns it, or you can attach it to an automation to run on the 1st of each month. The four steps (analysis, forecasting, strategy, review) are handled by the team's division of labor, and the only moment you need to show up is to make the call on the definition at waiting_user, just like this time.

The 'Retention Analysis Group' team is saved, and the 2 artifacts are archived to the project attachment library.

On your data

That was a simulated run

Leave your work email and we will run a live walkthrough on your real business data.

The problem

Multi-agent collaboration is the standard mode for complex analysis tasks, but over 65% of teams are inefficient at orchestration and execution. Creating a Team requires manually configuring the roles, tools, MCP servers, Skill, and knowledge base bindings of multiple Agents (with a complex config JSON structure and scattered docs), taking 1-2 hours on average with an error rate over 30%. Execution is even harder: after starting a TeamRun, you must manually poll status (running/waiting for user input/waiting for approval/completed/failed), read the pendingQuestion and reply when it waits for input, all while switching back and forth between web and command line. Sharing configs across teams is a total gap: no one knows which Teams already exist or which templates to reference, so everyone starts from scratch every time.

What it does

Full Team lifecycle management: list to create to update to delete to AI-generate config to browse templates, covering the whole process from creation to maintenance

TeamRun execution loop: start to SSE stream watching to auto-reply while waiting to fetch results and artifacts, completed in one chain

AI-generated config: input a goal description and AI automatically generates a Team config draft, no more hand-writing complex JSON

Multi-turn chat mode: run-chat supports multi-turn interaction, with session_id keeping context continuous

Complete status reference: 7 states (pending/running/waiting_user/waiting_approval/paused/completed/failed/cancelled) clearly defined, with the follow-up action for each state made explicit

When to use it

Creating a multi-agent collaboration team to run a complex analysis task

Starting a TeamRun and automatically fetching the analysis results and artifacts

Using AI to generate a Team config draft to avoid hand-writing complex JSON

Multi-turn interactive execution (auto-replying to continue when an Agent waits for user input)

Browsing built-in Team templates to quickly create a standardized collaboration team

Viewing the list of existing Teams and available projects

In the field

Case

A game operations team · retention analysis and optimization suggestions

The team needed to "analyze this month's retention and automatically generate optimization suggestions" but was unsure how to configure a multi-agent Team. With the AE Multi-Agent Collaboration Skill, the system input the goal description via +ai-generate, and AI automatically generated a Team config draft containing a data-analysis Agent and an optimization-suggestion Agent; after creating the Team with +create, it started the task with +run-start and watched the execution status via +run-watch SSE streaming; when the TeamRun paused in the waiting_user state, it auto-replied with +run-reply to continue, and finally fetched the analysis report and optimization-suggestion artifacts with +run-artifacts. The whole process from configuration to results was command-line-driven, cutting time from 2 hours of manual work to 15 minutes.

FAQ

How is waiting for user input handled during a TeamRun?

A +run-watch exit code of 2 means the waiting_user state, and stdout contains the pendingQuestion. After reading the question, reply with the user's answer via +run-reply, then re-run +run-watch to keep watching.

How good is the AI-generated config?

The AI-generated config draft needs human review and adjustment, but it greatly reduces the work of hand-writing JSON. We recommend generating a draft with +ai-generate, then reviewing and adjusting before +create.

What is the difference from ae-analysis?

ae-analysis handles single-Agent analysis operations (one command, one operation), while ae-team handles multi-agent collaboration orchestration (multiple Agents dividing the work to complete a complex task). Use ae-analysis for simple queries and ae-team for complex multi-step tasks.