Build an agent run workspace

Bring run progress, human approvals and source citations together so a person can inspect and direct an agent's work.

Agent run workspace

Review a research run, advance its local steps, and inspect the resulting brief.

Run plan

1
Scope the question
Compare warehouse onboarding friction
Complete
2
Read source notes
Three customer interviews attached
Complete
3
Draft the brief
Tie each recommendation to its evidence
Current step

Research brief

Draft

Make the first import easier

Participants could map customer names quickly, but hesitated when choosing date formats.

Show three sample rows before the first import. Keep the selected date format visible beside the preview.

Next: connect the recommendation to interview evidence.

Interactive example with illustrative data. Your application supplies the connected services.

Components in this recipe

2846Agent run workspaceBlocksReview a research run, advance its local steps, and inspect the resulting brief.

2866Human approval inboxBlocksReview proposed agent actions and record an approval or rejection with a reason.

2895Source citation panelBlocksConnect a written claim to its source excerpt and mark evidence as reviewed.

Build it yourself

Map your runner's events to stable run and step IDs. Render progress in the agent workspace and preserve completed steps after reconnecting.
Route consequential tool requests to the approval inbox. Send decisions to your backend and wait for acknowledgement before showing approval as applied.
Attach the citation panel to each generated artifact. Supply original source URLs and excerpts that the user is permitted to access.

Build with your agent

A ready-to-use task for your connected coding agent.

Build an agent run workspace

Use the Fieldbook MCP tools search_components and get_component. Select the framework already used by this project. Inspect access requirements and the returned props before editing.

Components: agent-run-workspace, human-approval-inbox, source-citation-panel.

Map your runner's events to stable run and step IDs. Render progress in the agent workspace and preserve completed steps after reconnecting.

Route consequential tool requests to the approval inbox. Send decisions to your backend and wait for acknowledgement before showing approval as applied.

Attach the citation panel to each generated artifact. Supply original source URLs and excerpts that the user is permitted to access.

These components do not run an agent or execute tools. Your backend owns streaming, retries, cancellation, approval enforcement and source access.

Test interrupted streams, rejected approval, retry, empty citations and duplicate events. Cancellation and approval must remain explicit user actions.

Preserve the project's styling conventions. Replace illustrative data with application inputs, add meaningful checks, and report integration work that remains.

Members only

Set up MCP Included with Pro

Connect your application

These components do not run an agent or execute tools. Your backend owns streaming, retries, cancellation, approval enforcement and source access.

Verify the result

Test interrupted streams, rejected approval, retry, empty citations and duplicate events. Cancellation and approval must remain explicit user actions.