Feedback: Issue Closure Recommendation agentic workflow

[denelon]

Feedback: Issue Closure Recommendation agentic workflow

This discussion collects feedback on an experimental workflow that identifies open winget-cli issues that may have been fully resolved by merged pull requests.

Tracking issue: #6422

The workflow is intended to help maintainers find stale open issues without making closure decisions for them. It would:

• consider only merged pull requests from microsoft/winget-cli;
• compare the issue's requested scope with the pull request's implemented scope;
• inspect comments chronologically for confirmation, remaining work, partial fixes, or continued reproduction;
• add Needs-Attention and post a short evidence-based recommendation only when confidence is high;
• avoid repeating a recommendation already posted by the workflow.

It would never close an issue, remove Needs-Triage, treat a merged cross-reference as sufficient evidence, or recommend closure when broader or unresolved work remains.

Initial pilot

The initial analysis covered:

• 1,231 open issues;
• 147 issues with merged pull-request cross-references;
• 49 issues with a merged pull request specifically in microsoft/winget-cli;
• two high-confidence closure-review candidates after semantic and live-thread verification.

The low yield is intentional. Precision is more important than volume for a workflow that can generate contributor notifications.

Feedback requested

• What evidence should be required before recommending closure?
• Should release verification be mandatory, or is a merged fix sufficient?
• Are there issue types that should always be excluded?
• Is Needs-Attention the right label for maintainer review?
• How long after a pull request merges should the workflow wait before evaluating an issue?

The workflow will begin in trial mode and remain recommend-only.