rails-audit-thoughtbot（Rails 审核 thoughtbot）

Rails Audit Skill (thoughtbot Best Practices)

Perform comprehensive Ruby on Rails application audits based on thoughtbot's Ruby Science and Testing Rails best practices, with emphasis on Plain Old Ruby Objects (POROs) over Service Objects.

Audit Scope

The audit can be run in two modes:

1. Full Application Audit: Analyze entire Rails application
2. Targeted Audit: Analyze specific files or directories

Execution Flow

Step 1: Determine Audit Scope

Ask user or infer from request:

• Full audit: Analyze all of app/, spec/ or test/, config/, db/, lib/
• Targeted audit: Analyze specified paths only

Step 2: Collect Optional Metrics (SimpleCov + RubyCritic)

Ask the user both questions upfront in a single AskUserQuestion so they can decide once:

• Question: "Before starting the audit, would you like to collect automated metrics?\n\n1. SimpleCov — runs your test suite to capture actual code coverage percentages\n2. RubyCritic — analyzes code complexity, duplication, and smells (does not run tests)\n\nBoth are recommended for the most thorough audit."
• Options: "Yes to both (Recommended)" / "SimpleCov only" / "RubyCritic only" / "Skip both"

Based on the user's choice, spawn the accepted subagents in parallel using the Task tool. Both can run at the same time because SimpleCov modifies the test helper while RubyCritic only reads source files — they don't conflict.

SimpleCov subagent (if accepted):

Read the file agents/simplecov_agent.md and follow all steps described in it. The audit scope is: {{SCOPE from Step 1}}. Return the coverage data in the output format specified in that file.

RubyCritic subagent (if accepted):

Read the file agents/rubycritic_agent.md and follow all steps described in it. The audit scope is: {{SCOPE from Step 1}}. Return the code quality data in the output format specified in that file.

After both agents finish, clean up:

• If SimpleCov ran: rm -rf coverage/
• If RubyCritic ran: rm -rf tmp/rubycritic/

Interpreting responses:

• COVERAGE_FAILED / RUBYCRITIC_FAILED: no data for that tool — use estimation mode (SimpleCov) or omit the section (RubyCritic). Note the failure reason in the report.
• COVERAGE_DATA: parse and keep in context for Steps 4 and 5 (overall coverage, per-directory breakdowns, lowest-coverage files, zero-coverage files).
• RUBYCRITIC_DATA: parse and keep in context for Steps 4 and 5 (overall score, per-directory ratings, worst-rated files, top smells, most complex files).

Step 3: Load Reference Materials

Before analyzing, read the relevant reference files:

• references/code_smells.md - Code smell patterns to identify
• references/testing_guidelines.md - Testing best practices
• references/poro_patterns.md - PORO and ActiveModel patterns
• references/security_checklist.md - Security vulnerability patterns
• references/rails_antipatterns.md - Rails-specific antipatterns (external services, migrations, performance)

Step 4: Analyze Code by Category

Analyze in this order:

1. Testing Coverage & Quality

- If SimpleCov data was collected in Step 2, use actual coverage percentages instead of estimates - Cross-reference per-file SimpleCov data: files with 0% coverage = "missing tests" - Check for missing test files - Identify untested public methods - Review test structure (Four Phase Test) - Check for testing antipatterns

1. Security Vulnerabilities

- SQL injection risks - Mass assignment vulnerabilities - XSS vulnerabilities - Authentication/authorization issues - Sensitive data exposure

1. Models & Database

- Fat model detection - Missing validations - N+1 query risks - Callback complexity - Law of Demeter violations (voyeuristic models) - If RubyCritic data was collected, flag models with D/F ratings or high complexity

1. Controllers

- Fat controller detection - Business logic in controllers - Missing strong parameters - Response handling - Monolithic controllers (non-RESTful actions, > 7 actions) - Bloated sessions (storing objects instead of IDs) - If RubyCritic data was collected, flag controllers with D/F ratings or high complexity

1. Code Design & Architecture

- Service Objects → recommend PORO refactoring - Large classes - Long methods - Feature envy - Law of Demeter violations - Single Responsibility violations - If RubyCritic data was collected, cross-reference D/F rated files and high-complexity files with manual code review findings

1. Views & Presenters

- Logic in views (PHPitis) - Missing partials for DRY - Helper complexity - Query logic in views

1. External Services & Error Handling

- Fire and forget (missing exception handling for HTTP calls) - Sluggish services (missing timeouts, synchronous calls that should be backgrounded) - Bare rescue statements - Silent failures (save without checking return value)

1. Database & Migrations

- Messy migrations (model references, missing down methods) - Missing indexes on foreign keys, polymorphic associations, uniqueness validations - Performance antipatterns (Ruby iteration vs SQL queries) - Bulk operations without transactions

Step 5: Generate Audit Report

Create RAILS_AUDIT_REPORT.md in project root with structure defined in references/report_template.md.

When SimpleCov coverage data was collected in Step 2, use the SimpleCov variant of the Testing section in the report template. When coverage data is not available, use the estimation variant.

When RubyCritic data was collected in Step 2b, include the Code Quality Metrics section in the report using the RubyCritic variant from the report template. When RubyCritic data is not available, use the not available variant.

Severity Definitions

• Critical: Security vulnerabilities, data loss risks, production-breaking issues
• High: Performance issues, missing tests for critical paths, major code smells
• Medium: Code smells, convention violations, maintainability concerns
• Low: Style issues, minor improvements, suggestions

Key Detection Patterns

Service Object → PORO Refactoring

When you find classes in app/services/:

• Classes named *Service, *Manager, *Handler
• Classes with only .call or .perform methods
• Recommend: Rename to domain nouns, include ActiveModel::Model

Fat Model Detection

Models with:

• More than 200 lines
• More than 15 public methods
• Multiple unrelated responsibilities
• Recommend: Extract to POROs using composition

Fat Controller Detection

Controllers with:

• Actions over 15 lines
• Business logic (not request/response handling)
• Multiple instance variable assignments
• Recommend: Extract to form objects or domain models

Missing Test Detection

For each Ruby file in app/:

• Check for corresponding _spec.rb or _test.rb
• Check for tested public methods
• Report untested files and methods

Analysis Commands

Use Claude Code's built-in tools instead of shell commands — they're faster, handle permissions correctly, and give better output:

• Find Ruby files by type: Use the Glob tool with patterns like app/models/**/*.rb, app/controllers/**/*.rb, app/services/**/*.rb
• Find test files: Use Glob with spec/**/*_spec.rb or test/**/*_test.rb
• Search for patterns in code: Use the Grep tool (e.g., search for rescue\s*$, \.save\b, params\.permit!)
• Read and count lines in files: Use the Read tool to inspect files; count lines from the output
• Find long files: Use Glob to list all app/**/*.rb files, then Read each to check line count

Report Output

Always save the audit report to RAILS_AUDIT_REPORT.md in the project root and present it to the user.