Agent-created ticket follow-through
Review a labeled Linear queue for stale, canceled and completed agent-created issues without editing them.
- 01Confirm the queue definition. Resolve the supplied teams, label and state mapping. Stop if a label is ambiguous or no reliable provenance convention exists. Use the tracker state mapping, not guesses from issue titles.
- 02Read the scoped issues. List matching issues with identifiers, links, state, creation and update times, owner and labels. Paginate within the configured cap, including completed and canceled issues when available. State any filtering or archive limitations that keep the queue incomplete.
- 03Review follow-through. Group the returned issues by current state and age. Identify open issues without an owner, issues past the stale threshold, and canceled issues that may deserve a review of the originating agent. Read a bounded set of comments only when needed to explain a finding.
- 04Separate evidence from inference. A completed issue is marked completed by the team, not independently verified as fixed. Cancellation is a signal to investigate, not proof the agent was wrong. An update timestamp is not a transition history; do not infer cycle throughput, reopened counts or resolution time from it.
- 05Return the report. Include the scoped counts, incomplete coverage, oldest unresolved issues, source links and a short list of review recommendations. Attribute issues to a particular agent only when a configured label or explicit issue record identifies it. If no issues match, report no matches within the checked scope.
Never create, edit, close, reopen, assign, comment on, or relabel an issue.
Configure read-only grants and restrict the review to the named teams.
Do not mix unrelated tickets into the queue to make the report look complete.
Do not report unavailable counts as zero or invent historical trends without supplied comparable snapshots.
Do not promise native outcome charts, automatic tuning, or verified business impact from issue status alone.