Review Design

Review the codebase for design and architecture problems, then file each finding as a Linear issue.

Overview

This skill dispatches a deep codebase analysis focused exclusively on design issues — poor separation of concerns, tight coupling, missing abstractions, god classes, inappropriate responsibility assignment, poor error propagation patterns, and concurrency model issues. Each finding is documented with code snippets and filed as a Linear issue.

Workflow

Phase 1: Gather Linear Context

Before reviewing code, collect the Linear team, project, and user information.

1. Fetch all data in parallel using mcp__claude_ai_Linear__list_teams, mcp__claude_ai_Linear__list_projects,

mcp__claude_ai_Linear__list_users, and mcp__claude_ai_Linear__list_issue_labels.

1. Ask the user all three questions in a single `AskUserQuestion` call with these questions:

• Team: Which team should the issues be created under? (list all teams from step 1). If only one team exists,

auto-select it and skip this question.

• Project: Which project should the issues be assigned to? (list all projects; if results can be filtered by

team, do so after team is known).

• Assignee: Who should the issues be assigned to? (list all users for the team, plus a "No assignee" option).

Important: AskUserQuestion requires at least 2 options per question. If a question would have only 1 option, auto-select that option and omit it from the AskUserQuestion call. If all questions can be auto-selected, skip the AskUserQuestion call entirely and inform the user of the auto-selected values.

1. Check existing labels from the fetched label data for the chosen team. If a label named design does not already

exist, create it using mcp__claude_ai_Linear__create_issue_label:

• design (color: #f39c12) — Architectural or design problems

Phase 2: Codebase Design Review

Dispatch a subagent to review the codebase. Use the feature-dev:code-explorer subagent type for deep analysis.

Design Reviewer Agent: Review all source files for design problems: poor separation of concerns, tight coupling, missing abstractions, god classes, inappropriate responsibility assignment, poor error propagation patterns, and concurrency model issues. For each finding, produce:

• Title (concise summary)
• Description (what the design problem is and how it manifests)
• Impact (what goes wrong if unfixed — maintainability, extensibility, testability)
• Importance (Critical / High / Medium / Low)
• Code snippet showing the problem (with file path and line numbers)
• Code snippet showing the suggested fix or restructuring

Phase 3: Present Findings to User

After the agent completes:

1. Compile all findings into a summary table.
2. Present the summary to the user showing: title, importance, and file location for each finding.
3. Ask for confirmation before creating Linear issues: "I found N design issues. Create all as Linear issues?"
4. Allow the user to exclude specific findings before filing.

Phase 4: File Linear Issues

For each approved finding, create a Linear issue using mcp__claude_ai_Linear__save_issue with:

title: <concise finding title>
team: <user-selected team>
project: <user-selected project>
assignee: <user-selected assignee, or omit if "No assignee" was chosen>
labels: ["design"]
priority: <mapped from importance: Critical=1, High=2, Medium=3, Low=4>
state: Todo
description: |
  ## Description
  <what the issue is and context>

  ## Impact
  <what goes wrong if left unfixed>

  ## Importance
  <Critical | High | Medium | Low> - <brief justification>

  ## Problem Code
  **File:** `<file_path>`

<code snippet showing the problem>

## Suggested Fix

<code snippet showing the corrected code or restructuring>

After all issues are created, present a final summary with the Linear issue identifiers and links.

Finding Quality Standards

Each finding must meet these criteria before filing:

• Specific: Point to exact code, not vague observations.
• Actionable: Include a concrete fix or restructuring, not just a complaint.
• Impactful: Explain real consequences for maintainability, extensibility, or testability.
• Non-trivial: Skip minor style preferences. Focus on findings that meaningfully affect the codebase's long-term health.

Priority Mapping