sandbox_report

Reads a human report again by its stable reportId, with notes, annotated images, recordings and timed transcripts.

{ "id": "<boxId>", "reportId": "<reportId>" }

Use the report ID shown in the app or returned by sandbox_review. The tool returns the account-owned report for that box even after an earlier tool delivered it. It refreshes file URLs, preserves the box's conversation holder and leaves other pending messages available for their recipient. Reports and their media remain available for seven days.

Read the report's notes, images, recordings and timed transcripts before continuing the task. The app receipt's dispatchState: "read" means the registered native conversation called this tool for the complete report and its native turn finished successfully. tool_delivered records that the complete tool result was prepared for handoff. Interpret deliveredAt, the delivery time, together with that state; neither establishes that the AI understood the feedback or completed every requested change. The submitted report's receipt remains available for seven days even after its review link closes.

For a Codex, Claude Code or Gemini native conversation started and registered by parallelsandbox-agent, submitting in the app notifies the persistent supervisor even after the AI finishes. It resumes the saved original conversation ID and asks the AI to call this tool. Recover a stopped supervisor using its exact session directory; its saved model and permission profile are reused (Quick start, native session feedback setup).

For an ordinary, unregistered MCP connection, press Copy message for AI on the app's report receipt and paste it into the original conversation to continue. It asks the AI to run sandbox_report { "id": "<boxId>", "reportId": "<reportId>" } and read the complete saved feedback and attachments before proceeding. The copied message also works after the report was delivered.

Every tool and topic is listed in the tool reference.