Learn from PRs Skill
A Claude Code command that analyzes patterns in pull request feedback -- from CodeRabbit, SonarQube, and human reviewers -- then recommends configuration updates to catch those same issues locally before the next PR. It's a feedback loop that turns review comments into preventive rules: scan 5 recent PRs, find recurring themes (style issues, logic errors, security concerns, testing gaps), discover existing config files, and propose specific updates prioritizing the earliest intervention point.
Key Quotes
"This command only suggests changes, never modifies files."
Key Themes
The "learn from your mistakes" pattern is powerful because it's self-improving. Every PR review becomes training data for your local config. The shift-left philosophy -- catch issues as early as possible in the pipeline -- connects to AI PR Reviewer (catching at PR time) and Fresh Eyes (catching at commit time). This skill pushes the boundary even further: catch it before you even start the next task.
The read-only design ("only suggests, never modifies") is a good trust-building choice. You want humans reviewing the proposed config changes, not auto-applying them.
Critical Analysis
The concept is excellent; the execution details matter enormously. The value depends entirely on pattern quality -- can it distinguish between "reviewer has a style preference" and "reviewer caught a genuine bug class"? Grouping by frequency and severity helps, but false positives in config recommendations could create friction that makes developers disable the whole thing. See Awesome Agentic Patterns for the broader pattern catalogue this fits into (Feedback Loops category), and 14 More lessons from 14 years at Google's point about small PRs and code review standards.
Sources: Learn from PRs Skill Last updated: 2026-05-14