How the Agent works

QUBE Agent is an automated assistant designed to help experienced NIPO ODIN scripters diagnose questionnaire defects and prepare targeted corrections. The agent is entirely human-directed: it acts only when an authorized editor selects an existing ticket and explicitly requests an investigation.

Requesting an investigation

The agent works on ODIN tickets. It does not run on an automated schedule, scan scripts in the background, or provide general conversational chat.

When requesting agent work, you choose a model option and approve a maximum charge. The investigation uses:

• The exact script version linked to the ticket.
• The captured interview path and user notes describing the defect.
• Surrounding questionnaire logic and structure.

Proposals, not replacements

The agent never overwrites an existing script version or silently alters your source files. When it identifies a solution, it publishes a separate, immutable script version as a proposed fix (or candidate version).

An AI proposal is not proof of correctness. Automated checks cannot guarantee that an interview will behave properly under all field conditions. Before accepting a change:

1. Compare source: Review the line-by-line differences between the reported script version and the proposed version.
2. Replay the journey: Open the proposed version in the debugger and replay the captured interview path to confirm the fix, checking any surrounding routing logic that might be affected.

Ticket lifecycle and verification

Every investigation leaves an attributable comment on the ticket explaining what was analyzed and what changed. If an investigation fails or produces no fix, the run logs the findings and the ticket remains open for human attention.

flowchart TD
  ticket[Human selects ticket & requests agent] --> run[Agent investigates using script context]
  run --> outcome{Outcome}
  outcome -->|Fix proposed| proposed[Publishes immutable version; marks Fixed]
  outcome -->|No fix or failure| record[Records findings; ticket stays open]
  proposed --> review[Human compares diff & replays interview]
  review -->|Verified| resolve[Human promotes source & resolves ticket]
  review -->|Issue remains| reopen[Human reopens ticket with feedback]

• Fixed: The agent marks a ticket Fixed when a proposed version is ready for review. In QubeSurvey, Fixed means ready for human verification, not resolved. The agent never resolves a ticket or approves its own work.
• Resolved: Only a human reviewer can verify the result, promote the proposed version into use, and mark the ticket Resolved. If the defect persists, the reviewer reopens the ticket with feedback.

For review and ticket practices, see Tickets and proposed fixes.

Funding and credit

Agent work uses prepaid account credit, separate from your subscription. Each run costs the provider’s measured charge plus 50%, within the maximum charge you approved. Unused reserved credit is released when the run settles. See People, access and billing for current payment availability.