Tickets and proposed fixes

Tickets connect a reported problem to the script version and interview journey where it occurred. Comments keep the people on a project informed.

Describe and discuss

Use a short, specific title. In the description, separate what happens now from what should happen. Comments support Markdown: paragraphs, lists, emphasis, links and fenced code blocks. Raw HTML is not executed.

Work with QUBE Agent

QUBE Agent is optional. An authorized editor selects tickets and requests agent work with a model and spending ceiling. Agent access also requires the account’s permissions and available prepaid credit.

The agent investigates and may prepare a new script version for review. It does not silently promote that version or approve its own work. Read the handoff comment and inspect the linked version; a status message alone is not evidence that the problem is fixed.

If a run fails or produces no reviewable version, the ticket still needs human attention. Keep the report open and use the visible run information when investigating or requesting another attempt.

Understand the outcome

A demonstration ticket with a proposed fix in version 2, source comparison, testing and human review controls.

The report stays alongside its proposed fix. Start with Compare changes and Test proposed fix; use Resolve only after checking the result. The discussion records what was changed and what still needs your attention.

Fixed means an implementation is ready for a human to check. It does not mean that review is complete. Resolved means a human has verified the result and closed the ticket. Reopen it when the problem remains or returns.

Test the candidate and compare its source with the reported version before resolving. See Testing a script.

flowchart TD
  report[Open ticket] --> work[Human or agent investigates]
  work -->|No proposed fix| followup[Record findings; ticket stays open]
  work -->|New version prepared| fixed[Fixed: ready for review]
  fixed --> check[Human compares and tests]
  check -->|Verified| resolved[Resolved]
  check -->|Still wrong| reopen[Reopen with feedback]
  reopen --> work

Every agent attempt should leave a human-readable follow-up. Only a human decides that the tested result is satisfactory.