Self-Improve Skill

Automatically detect, analyze, and fix issues in the oh-my-gemini project itself using multi-agent orchestration.

Purpose

Enable oh-my-gemini to improve itself through:

• Automated bug detection and fixing
• Code quality improvements
• Test coverage enhancement
• Performance optimization
• Documentation updates

Workflow

Phase 1: Detection

1. Run Diagnostics

   • Execute npm test to find failing tests
   • Run npm run lint to find linting errors
   • Run npm run build to find compilation errors
   • Check TypeScript errors with lsp_diagnostics

2. Code Analysis

   • Use lsp_diagnostics to find type errors
   • Use ast_grep to find code patterns that need improvement
   • Use grep to find TODO/FIXME comments
   • Check for deprecated patterns

3. Dependency Check

   • Run npm outdated to find outdated dependencies
   • Check for security vulnerabilities with npm audit

Phase 2: Prioritization

Create a prioritized list of issues:

1. Critical: Build failures, test failures, type errors
2. High: Linting errors, security vulnerabilities
3. Medium: Code quality issues, deprecated patterns
4. Low: Documentation, minor improvements

Phase 3: Fixing

For each issue:

1. Analyze the Problem

   • Use Read to examine relevant files
   • Use lsp_definition to understand code structure
   • Use lsp_references to find all usages

2. Create Fix Plan

   • Break down into atomic tasks
   • Use TodoWrite to track progress

3. Implement Fix

   • Use Edit to make changes
   • Follow existing code patterns
   • Ensure type safety

4. Verify Fix

   • Run lsp_diagnostics on changed files
   • Run npm run build to verify compilation
   • Run npm test to verify tests pass
   • Check for regressions

Phase 4: Continuous Improvement

1. Code Quality

   • Refactor complex functions
   • Improve type safety
   • Add missing error handling
   • Improve documentation

2. Test Coverage

   • Identify untested code paths
   • Add missing test cases
   • Improve test quality

3. Performance

   • Identify performance bottlenecks
   • Optimize slow operations
   • Reduce bundle size

4. Documentation

   • Update outdated docs
   • Add missing examples
   • Improve clarity

Agent Delegation

Delegate tasks to specialized agents:

• architect: For architectural decisions and complex refactoring
• executor: For implementation tasks
• code-reviewer: For code quality review
• tester: For test-related tasks
• writer: For documentation updates

Example Usage

> self-improve: Find and fix all TypeScript errors
> 자가발전: 모든 버그를 찾아서 수정해줘
> auto-fix: 린트 에러와 타입 에러를 모두 수정해줘

Safety Measures

1. Backup Before Changes

   • Create git commit before major changes
   • Use git stash for experimental changes

2. Incremental Changes

   • Fix one issue at a time
   • Verify each fix before proceeding

3. No Breaking Changes

   • Maintain backward compatibility
   • Update tests for any API changes

4. Review Before Commit

   • Show changes before committing
   • Get user approval for major changes

Configuration

Can be configured via .gemini-cli/GEMINI.md:

## Self-Improve Settings

- Auto-fix: true
- Auto-commit: false (requires approval)
- Test-before-commit: true
- Max-changes-per-run: 10