From Case to Agent This prototype changes the case view into an Agent workspace. The main conversation stays clear, while delegated sub tasks do the work. A familiar place to ask First, ask which servers the agent can reach. The main thread acknowledges the request and delegates verification to a worker. Working means the whole task Agent is working remains visible while the worker checks access. Commands and tool output never appear as conversation messages. The answer comes back to you When verification finishes, the main agent summarizes the result. These example results are simulated, and are not a fresh check of the servers. Details stay out of the way Hover over the agent response to reveal Logs. Keyboard focus and a touch menu provide the same access. Hover does not fetch execution details. Open the sub-task list Click Logs to open the sub task drawer. It shows task names, status, and timing. Select a worker to inspect its execution. Fetch only when requested Opening a worker starts an authorized request through Agents Hub to Robodev Agents. This loading state represents the proposed fetch, not a live backend connection. One source of execution history The drawer shows commands, tool results, and timestamps fetched from Robodev's canonical history. Hub keeps references and status, rather than another stored transcript. Close details; keep the conversation Closing the drawer stops its detail stream. The conversation and final answer remain available, without keeping raw worker logs in the chat. A question can involve several workers Next, ask why the latest deployment returns errors. The main agent starts separate workers for application logs and deployment health. Parallel work, one conversation Both workers appear in one task tree. The activity indicator stays on until the work and final summary finish. Their tool calls remain in details. Inspect one worker's evidence Select the log worker to inspect its evidence from Robodev. This demonstration uses fixture events, including a simulated database connection timeout. A concise, evidence-based update The main agent combines the results into a concise update. It reports the suspected cause and suggests a next step, without making an unrequested production change. Secrets use a dedicated form For a task that needs a missing credential, show a secure secret request in the main thread. The person enters the secret into a dedicated form. Waiting is different from working The task shows Waiting for a secret. The secret value goes to the vault, never to chat or logs. The main agent only receives the request status. Resume without exposing the secret Once the vault confirms receipt, the worker resumes with scoped access. Agent is working returns, and the main thread reports the result when it is ready. Two views, one coherent experience Agents Hub presents messages, questions, forms, and task status. Robodev runs the orchestrator and workers, and remains the source of truth for execution details. Ready for review This is an animated interface prototype with simulated conversations and fetches. It illustrates the architecture proposal; the production application has not been changed.