checklist-runner_skill

Overview

This skill parses and verifies GitHub PR and issue checklists, then auto-checks items that pass verification when safe. It classifies each unchecked item, runs the cheapest applicable verification (CI, shell checks, scans, or human queries), and updates checkboxes according to ownership and permission rules. The skill prioritizes safety: it avoids editing others' posts silently, confirms expensive scans, and prevents race conditions.

How this skill works

Given a PR/issue URL, number, or the current branch, the skill fetches the body and relevant comments, strips code blocks, and extracts checklist items with metadata (author, updated_at, nesting). Items are classified into Auto, CI, Shell, Scan, or Human categories by pattern matching and confidence scoring. Verifications execute in cost order (Auto → CI → Shell → Scan → Human), produce PASS/FAIL with evidence, and check off items only when ownership and updated_at checks allow it; otherwise it posts suggestions or reports.

When to use it

• When asked to “verify the checklist”, “check off items”, “run the checklist”, or “process checklist”
• When a PR or issue contains unchecked - [ ] items that need automated or semi-automated verification
• When you want an evidence-backed verification report before merging
• When you need to batch human confirmations for manual items

Best practices

• Classify items before executing to pick the cheapest verification path
• Check CI status once (no polling); offer wait/re-run/local options for pending checks
• Confirm before launching scan subagents and cap at 5 to control resource use
• Respect ownership: auto-edit only your own posts; use comments/suggestions for others
• Use updated_at to detect race conditions and abort edits if content changed
• Batch human questions into a single prompt and include evidence for each PASS/FAIL

Example use cases

• Run /checklist on a feature PR to auto-verify lint, tests, and file-based checks before merge
• Audit an issue checklist where reviewers added manual items requiring explicit approvals
• Detect and report secrets or changelog omissions using scan subagents (with user confirmation)
• Generate a summarized verification report showing passed, failed, pending, and already-checked items for a release PR