self-improvement_skill

Overview

This skill is a meta-skill that analyzes pull requests, issues, and user interactions to automatically improve Cursor rules and skills. It extracts concrete lessons from code reviews, bug reports, and user feedback, then creates or updates rules and tracks their effectiveness. The goal is continuous prevention of common mistakes and faster convergence on useful guidance.

How this skill works

The skill fetches PR and issue data, parses review comments and labels, and scans conversation transcripts for corrections and preference signals. It identifies recurring failure and success patterns, maps findings to existing rules, and either updates those rules or generates new ones with concrete examples and references. Finally, it records metrics to measure rule effectiveness over time.

When to use it

• After a closed PR (merged or rejected) to capture lessons learned
• When multiple issues reveal the same bug or misunderstanding
• Following repeated user corrections or clarification requests
• To create rules when a new recurring pattern emerges
• When validating that rule updates reduced future errors

Best practices

• Be specific: capture concrete examples and exact rejection reasons
• Always reference sources: include PR/issue numbers and snippets
• Prioritize high-impact patterns that cause most rejections
• Test rule changes in a safe environment before wide rollout
• Iterate: refine rules as more examples accumulate

Example use cases

• Analyze PR #3567 to extract deprecated-function errors and update common-mistakes
• Scan issues tagged with ‘bug’ to find missing conversation persistence patterns
• Monitor conversations to learn individual user preferences and update profiles
• Create a new rule when several PRs misuse audio storage flow
• Measure reduction in PR rejections after adding pre-implementation checks