# Code Review > Orchestrates a 3-agent parallel code review combining pattern compliance, Go conventions, and architectural analysis. Use when this capability is needed. - Skill: `tomevault-io/code-review-297` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/code-review-297` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/code-review-297/raw - Safety review: pending - 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/code-review-297 --- # Code Review Skill Orchestrate a comprehensive code review using three specialist agents in parallel, then synthesize their findings. ## Inputs This Skill reads `$ARGUMENTS`. Accept these patterns: - A file path, directory path, or glob pattern to review - A PR number (e.g., `#42` or `42`) - `--diff` to review staged/unstaged git changes - No arguments defaults to reviewing uncommitted changes (`git diff`) If the user didn't supply arguments, ask what they want reviewed. ## Step-by-step procedure ### Step 1: Determine review scope Identify what code to review: - **PR**: Run `git diff ...` to get changed files - **Files/directories**: Use the provided paths - **Staged changes**: Run `git diff --cached` - **Uncommitted changes**: Run `git diff` Collect the list of changed/target files and their contents. ### Step 2: Delegate to three agents in parallel Launch all three agents simultaneously using the Task tool: 1. **`code reviewer`** (tactical) - Pattern compliance, linting, best practices - Prompt: Provide the file paths and ask it to review against Cognee patterns, run linters, and return structured findings 2. **`solution architect`** (strategic) - Architectural concerns, design coherence - Prompt: Provide the file paths and ask it to evaluate architectural consistency, separation of concerns, and design coherence 3. **`go software architect`** (Go-specific) - Go coding conventions, naming conventions, implementation conventions - Prompt: Provide the Go file paths and ask it to check for idiomatic Go (stuttering, interface design, error wrapping, context propagation, package structure, naming) ### Step 3: Synthesize findings When all three agents return: 1. **Identify agreements** — Issues flagged by multiple agents 2. **Identify unique findings** — Issues only one agent caught 3. **Identify disagreements** — Conflicting recommendations ### Step 4: Reconcile disagreements If disagreements exist: 1. Present the conflicting views back to each disagreeing agent 2. Ask each to provide reasoning 3. Request they reach consensus 4. If no consensus, present both perspectives to the user ### Step 5: Compile unified findings **IMPORTANT**: Read the template at `skills/code-review/templates/code-review-template.md` BEFORE delegating. When delegating to the documentation agent: 1. Provide the **template path** so the agent reads it and follows its section order exactly 2. Provide a **prior review** from `docs/code-reviews/` as a style reference (e.g., the most recent one) 3. Provide the **review content** only (findings, metadata, compliance checks) — do NOT dictate document structure in the prompt 4. The agent must use the template's sections, not invent its own Create a new file under `docs/code-reviews/` named for the scope, for example `docs/code-reviews/phase-09-routing-engine.md`. Keep the template section order intact. If you must show a summary to the user, include only a short synopsis plus a linkable file reference to the completed review document. ### Step 6: Collaborate with user on resolution - Discuss trade-offs for architectural decisions - Get user approval before delegating fixes - Do NOT auto-fix without user consent ### Step 7: Delegate approved fixes After user approves specific fixes, delegate to appropriate specialists: - Go issues → `go-software-agent` - Shell issues → `shell-script-agent` - DevOps issues → `go-devops-agent` - Documentation → `documentation-agent` ## Key principles - Three perspectives: tactical (code-review) + Go-specific (go-architect) + strategic (solution-architect) - Agents reconcile disagreements BEFORE presenting to user - User is involved in resolution decisions, not just notified - Documentation updated to capture learnings from review - Pattern compliance checked against Cognee knowledge base ## Finding categories - **Pattern Violation**: Deviates from documented Cognee pattern - **Go Idiom**: Non-idiomatic Go (stuttering, interface placement, error handling) - **Architecture**: Structural or design concern - **Security**: Potential security issue - **Error Handling**: Missing or inadequate error handling - **Testing**: Missing tests or edge cases - **Performance**: Inefficient implementation - **Style**: Naming, formatting, organization issues --- > Converted and distributed by [TomeVault](https://tomevault.io/claim/twistingmercury) — claim your Tome and manage your conversions.