/tsh-review-codebase
Agent: Architect
File: .github/prompts/tsh-review-codebase.prompt.md
Performs a comprehensive code quality analysis covering dead code, duplications, and improvement opportunities.
Usage
/tsh-review-codebase
What It Does
Phase 1: Context Discovery
- Loads project conventions from
tsh-technical-context-discoveringskill. - Identifies repository type (monorepo vs single system).
- Identifies tech stack, frameworks, and dependencies per layer/app.
Phase 2: Codebase Analysis
Runs parallel subagents per layer/app to detect:
Dead Code:
- Unused imports, exports, components, routes, and endpoints.
- Unreachable code paths.
- Deprecated code no longer referenced.
- Files not imported anywhere.
- Code only imported in its own test file.
Duplications:
- Duplicated functions, UI components, API patterns.
- Duplicated validation logic, utilities, type definitions.
- Copy-pasted code blocks differing only in variable names.
Improvement Opportunities:
- High cyclomatic complexity, deeply nested logic.
- Single Responsibility Principle violations.
- Missing or inconsistent error handling.
- Excessive use of
anyin TypeScript. - Outdated patterns that could be modernized.
- Performance issues and security concerns.
Phase 3: Architecture Review
- Module/package boundary evaluation.
- Shared code extraction assessment.
- Dependency graph analysis (circular dependencies).
- Separation of concerns verification.
Phase 4: Report Generation
All findings are prioritized:
- 🔴 Critical — Must be fixed (bugs, security issues, maintenance burden).
- 🟡 Important — Should be fixed (quality, readability, maintainability).
- 🟢 Nice to Have — Optional (polish, optimization).
Skills Loaded
tsh-codebase-analysing— Structured analysis process.tsh-technical-context-discovering— Project conventions and architecture patterns.tsh-code-reviewing— Code quality standards and security considerations.
Output
A code-quality-report.md file saved to the repository root, containing:
- Executive summary of overall code health.
- Findings organized by layer/app.
- Dead code, duplications, and improvement tables with file paths and line numbers.
- Architecture observations.
- Summary counts by severity.
- Recommended action plan (immediate, short-term, long-term).