This paper presents an LLM-based pipeline for analyzing CI/CD workflows as executable operational policy. It combines repository enrichment, anti-pattern detection, workflow-stage mining, and repository-level recommendation generation. From 59,550 GitHub repositories with at least 1,000 stars, the study identifies 34,225 projects using CI/CD, collects 127,559 configuration files, and analyzes 75,201 workflows. The reported results include 434,769 anti-pattern findings, statistically significant differences in stage usage across programming languages, and distinct domain profiles such as greater release and cache usage in mobile projects. Few-shot prompting produces the strongest recommendation results, averaging 8.25 recommendations per repository with 96.1% YAML-valid snippets.
No heat snapshots are available in the last 24 hours.