Review the output

Read what your App produced for a work item, check the deliverables, and decide whether to approve, run it again, or push back. You stay in control of every result.

Audience: analysts

Open a work item

Click a row in the rail. The right column fills with the work item's title, a status badge, and a row of actions.

The actions you will see, depending on the item's state:

• Start Conversation or Open Conversation. Open the review thread, where you can talk through this work item. Start Conversation appears once the run has reached a terminal state (Succeeded, Failed, or Cancelled).
• Open Task. Jump to the underlying task run to see every step that was taken.
• Open Audit. Open the audit trail, scoped to this work item.
• Rerun. Re-execute the work item from the beginning. Available once the current run has finished.

If the latest run failed, a red banner under the actions shows the error message. Attachments you sent in (or that arrived from a trigger) appear as downloadable file chips below the actions.

Read the Review section

Below the actions sits the Review section. This is where the actual outputs land.

The header shows two things:

• Coverage. For example, "3 / 4 outputs". This counts how many of the slots the App's published workflow defines actually came back populated.
• Status pill. One of Pending, Ready, Repairing, Failed, or No outputs. Pending means the run has not produced results yet. Ready means everything is in. Repairing means the system is patching missing pieces. Failed means the validator could not put the results in the expected shape.

Switch between output tabs

Each output the App expects shows up as a tab under the Review header. Click any tab to read that output. Tabs work like a notebook: you can move between them and your place is remembered until you switch work items.

If a tab is missing a required result, you will see a small dot next to its name once the run is no longer pending.

Read each output type

Outputs come in five shapes, and the page renders each one differently:

• Markdown. Formatted text with headings, lists, links, and inline file references.
• Table. A grid with column headers and rows. Read across, scan down.
• Verdict. A colored badge (green for pass, red for fail, yellow for review, grey for neutral) plus a written rationale. The raw payload sits behind a Show raw payload toggle if you want to see the underlying data.
• List. A bulleted set of values.
• JSON. Raw structured data shown in a code block. Useful when the next step is paste into a downstream tool.

Copy the output

Every populated output panel has a Copy button in the top right. Click it and the output is on your clipboard, ready to paste into an email, a ticket, or a spreadsheet. Markdown copies as plain text. Tables, verdicts, lists, and JSON copy as structured JSON so the recipient can paste them anywhere.

Download file deliverables

When an output references files in the workspace (paths under work-items/ or artifacts/), they appear as a Deliverables row of file chips above the output body. Click a chip to download the file directly. The filename shows in the chip and the full path is in the chip's tooltip.

Check the metadata

At the bottom of the page is a Metadata section you can expand. Inside is everything the system knows about this work item:

• When it was Created, Updated, Completed, or Failed.
• How many Steps the workflow ran out of how many total.
• The Current step (label or key) if the run is still going.
• The number of Attachments.
• Created via (the trigger name or how it was submitted).
• The Latest task identifier, if you need to share it with engineering.

Decide what to do next

After you read the output, you have three paths:

• The result is right. Move on. The output you saw is what the App publishes back to its source (the email reply, the Slack message, the row in the system of record). No extra step is needed.
• The result is wrong. Click Start Conversation and say what to fix. See Correct and send back.
• The run got stuck. Click Rerun to try again with the same inputs.