github-issue-triage_skill

Overview

This skill triages GitHub issues using a streaming, per-issue background-task architecture. It fetches every issue in a time range with exhaustive pagination, launches one background task per issue, and streams each issue's analysis as soon as it completes. At the end it aggregates results into a final comprehensive report.

How this skill works

The skill first registers todos and exhaustively collects issues (and PRs) for the requested time window. It then launches one independent background task for each issue (run_in_background=true) and repeatedly polls background_output() non-blocking to stream each task's result as it finishes. After all tasks complete it compiles a final summary report and actionable lists (critical, close, auto-respond, investigate, features, needs info).

When to use it

• You need fast, parallel triage across many issues with live progress updates.
• You want isolated analysis per issue so failures don’t block others.
• You need early detection of critical bugs while other analyses run.
• You must produce a final aggregated report after streaming completes.
• You want draft responses or auto-categorization for maintainers.

Best practices

• Always use exhaustive pagination to fetch every relevant issue before launching tasks.
• Enforce 1 issue = 1 background task; never batch multiple issues into one task.
• Stream results immediately using background_output(block=false) to avoid long silence.
• Register todos at initialization to track phases and state changes.
• Rate-limit polling loops (small delays) to avoid API or task system throttling.
• Prioritize critical issues for immediate attention and surface them first.

Example use cases

• Triage issues opened in the last 48 hours and surface critical bugs immediately.
• Run nightly triage across all open issues and generate draft responses for common questions.
• Correlate issues with recent PRs to identify probable fixes while streaming results.
• Provide maintainers live progress during large-scale issue reviews (hackathons, release windows).
• Automatically categorize issues into close, needs-info, feature-backlog, and needs-investigation lists.