[Feature]:

# Proposal: AgentScope Studio support for visualization, interaction and debugging of implemented agents

## Description

Hi AgentScope team,

I would like to propose enhancing **AgentScope Studio** so that users can **visualize**, **interact with**, and **debug** the agents they implement with AgentScope—beyond the current project management and run tracing.

### Current state

From the [Studio tutorial](https://doc.agentscope.io/tutorial/task_studio.html), Studio already provides:

- **Project management** for agent application development
- **Visualization** of running applications and tracing (e.g. token usage, model invocations)
- The built-in **Friday** agent for secondary development

The [roadmap](https://github.com/agentscope-ai/agentscope/blob/main/docs/roadmap.md) also mentions: *"Provide enhanced AgentScope Studio with visual components for developing, tracing and debugging"*.

### Why this matters

As developers build more complex agent systems (multi-agent workflows, ReAct agents, tools, memory, planning), they need:

1. **Visualization**
- See the **topology** of their agents (which agents exist, how they are connected, which tools/skills they use).
- View **conversation and message flow** in a clear, navigable way (who said what, in which order, with which tool calls).

2. **Interaction**
- **Drive or replay** specific runs from the Studio UI (e.g. send a test message, trigger a workflow).
- **Inspect and edit** prompts, tool inputs/outputs, or intermediate state in a run, to understand behavior.

3. **Debugging**
- **Break down a failing run**: see exactly which step (agent reply, tool call, model call) led to an error or unexpected output.
- **Compare runs** (e.g. before/after a code or prompt change) to validate fixes.
- Optional: **step-through** or **pause/resume** for local/dev runs to debug interactively.

These capabilities would make Studio a central place for developing and validating AgentScope-based agents, especially as the framework grows (voice agents, multi-agent realtime, A2A, etc.).

### Proposed direction

A short **“Studio: visualization, interaction and debugging for implemented agents”** enhancement could:

1. **Visualization**
- In Studio, expose a **view of the agent graph** for a project/run (agents, connections, tools).
- Improve **run/conversation view** so that messages, tool calls, and traces are easy to browse and filter (e.g. by agent, by time, by type).

2. **Interaction**
- Allow **starting or triggering** a run (or a single agent) from the Studio UI with custom input.
- Allow **inspecting and, where safe, editing** key data (e.g. last user message, tool results) for a run to quickly test “what if” scenarios.

3. **Debugging**
- Make **errors and failures** in a run very visible (e.g. highlight the step that failed, show stack trace or error message).
- Add simple **run comparison** (e.g. side-by-side or diff of two runs) to support iterative debugging.

This would align with the existing roadmap item on *“visual components for developing, tracing and debugging”* and would make Studio more valuable for anyone building non-trivial agents with AgentScope.

If this direction sounds reasonable, I am happy to help refine it (e.g. as a GitHub Discussion) or contribute to design/docs. Thank you for your work on AgentScope and Studio.