github-triage_skill

Overview

This skill provides unified GitHub triage for both issues and pull requests, launching one background subtask per item to analyze, act, and report. It processes all open items in parallel, tracking results via TaskCreate and streaming findings as each subagent finishes. The skill is geared for automated, read-only analysis and safe automated merges for trivial bugfixes when strict conditions are met.

How this skill works

The skill fetches all open issues and PRs using the gh CLI or a bundled script, classifies each item into precise categories, then creates one TaskCreate entry and spawns a background task (category="free") per item. Each subagent follows a strict, explicit prompt for its item type (question, bug, feature, other, PR bugfix, PR other), posts comments or merges only when rules allow, and reports results back through background_output().

When to use it

• Routine maintenance to triage a large backlog of open issues and PRs.
• Automated first-pass analysis to identify clear bug confirmations and trivial merges.
• When you want parallel background processing without blocking the main workflow.
• As a pre-screen before human review to surface high-priority or problematic items.
• To generate an auditable record of triage actions and suggested next steps.

Best practices

• Grant the skill read-only GH CLI/API access and ensure status checks are visible to the agent.
• Limit auto-merge to PRs that meet all merge conditions (passing CI, approved, non-draft, mergeable).
• Supply comments and TaskCreate templates that include repository context so subagents never guess.
• Monitor TaskCreate outputs and review any ACTIONS marked NEEDS_HUMAN_DECISION before taking further automated steps.
• Keep subagent prompts explicit and conservative—never allow guessing or implicit assumptions.

Example use cases

• Answering reproducible how-to and configuration questions by searching the codebase and closing resolved issues.
• Confirming bugs by locating the exact problematic function and recommending precise fixes for maintainers.
• Automatically merging a small bugfix PR when CI is green and reviews approve, otherwise flagging for human review.
• Assessing feature requests for feasibility and mapping where implementation would live in the codebase.
• Rapidly triaging a spike in incoming issues after a release to surface critical regressions.