# Descoped LLM Skills Code Review > Code Review Command Generator - Skill: `tomevault-io/descoped-llm-skills-code-review` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/descoped-llm-skills-code-review` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/descoped-llm-skills-code-review/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/descoped-llm-skills-code-review --- # Code Review Command Generator Generate a project-specific `/{PREFIX}-code-review` Claude Code command that performs deep clean code analysis tailored to the repo's tech stacks, module structure, and architecture. ## What This Skill Produces A single Claude Code command: `.claude/commands/{PREFIX}-code-review.md` The generated command includes: - Scope mapping (keyword → directory paths) - All 7 clean code categories with tech-appropriate examples and thresholds - Stack-specific quality checks for the project's languages/frameworks only - Per-category structured report tables - Post-report triage with issue command integration (when available) - Quality check commands per workspace area ## Workflow ### Phase 1: Auto-Detect Project Structure Scan the repo root for tech markers: | Marker | Tech | |--------|------| | `Cargo.toml` | Rust | | `pyproject.toml`, `setup.py`, `requirements.txt` | Python | | `go.mod` | Go | | `build.gradle`, `build.gradle.kts`, `pom.xml` | Java/Kotlin | | `package.json` (check `dependencies` for framework) | TypeScript + React/Next.js/Svelte | | `*.xcodeproj`, `Package.swift` | Swift/iOS | | `settings.gradle` with android | Kotlin/Android | Map detected technologies to workspace areas by examining directory structure. Look for common patterns: - Monorepo: `crates/`, `packages/`, `apps/`, `services/` - Standard: `src/`, `lib/`, `cmd/`, `internal/` - Frontend: `frontend/`, `web/`, `app/` - Mobile: `ios-app/`, `android-app/`, `apps/ios/`, `apps/android/` ### Phase 2: Confirm with User Present detected workspace areas and ask the user to confirm or adjust. Gather: 1. **Workspace areas** — present detected modules with paths; let user add/remove/rename 2. **Project prefix** — short name for the command (e.g., `frk` → `/{frk}-code-review`). If a `{PREFIX}-issue` or `{PREFIX}-start` command already exists in `.claude/commands/`, suggest reusing that prefix. 3. **Architecture style** — hexagonal, layered, flat, microservices, or custom. This determines the dependency order for reviewing code (e.g., core → infrastructure → API → frontend for hexagonal). 4. **Project-specific rules** — any architectural constraints to respect during review. Examples: "CMS 3-layer architecture is intentional", "adapter pattern in infrastructure is production-ready", "greenfield — delete dead code directly". These become extra rules in the generated command. 5. **Quality check commands** — per workspace area. Auto-suggest from `references/tech-checks.md` based on detected tech, let user confirm or override. ### Phase 3: Generate the Command Read the following references to compose the command: - `assets/code-review-command.md` — structural template with placeholders - `references/categories.md` — 7 categories with per-tech variants - `references/tech-checks.md` — stack-specific quality checks **Placeholder resolution:** | Placeholder | Source | |-------------|--------| | `{ISSUE_COMMAND_NOTE}` | If `/{PREFIX}-issue` exists: ` — does NOT create GitHub issues (use /{PREFIX}-issue for that)`. Otherwise: empty. | | `{SCOPE_TABLE}` | Workspace areas as argument → path mapping table. Always include `all` (full codebase, one area at a time) and `path/to/file` (single file or directory). | | `{DEPENDENCY_ORDER}` | `Work in dependency order: X → Y → Z.` based on architecture style. | | `{CATEGORIES}` | All 7 categories from `references/categories.md`. Include only the tech-specific items matching this project's stacks. Use thresholds for the project's languages. | | `{STACK_CHECKS}` | Relevant sections from `references/tech-checks.md` only. Do not include stacks the project doesn't use. | | `{ISSUE_COMMAND_REF}` | If `/{PREFIX}-issue` exists: ` (/{PREFIX}-issue)`. Otherwise: empty. | | `{PROJECT_RULES}` | User-provided architectural constraints as additional rule bullets. | | `{QUALITY_COMMANDS}` | `- **Quality checks**: ` followed by the per-area check commands. | Write the assembled command to `.claude/commands/{PREFIX}-code-review.md`. ### Phase 4: Commit Stage and commit: ```bash git add .claude/commands/{PREFIX}-code-review.md git commit -m "chore: add {PREFIX}-code-review command" ``` ## Design Principles - **Self-contained output** — the generated command works without this skill being installed - **Tech-appropriate** — only includes categories, examples, thresholds, and checks relevant to the project's actual tech stacks - **No false positives** — the generated command emphasizes verification before reporting - **Integrates with existing workflow** — detects and references existing issue/start commands - **Re-runnable** — can regenerate the command when repo structure changes --- > Source: [descoped/llm-skills](https://github.com/descoped/llm-skills) — distributed by [TomeVault](https://tomevault.io).