review-toolkit

A Claude Code plugin that provides structured code review commands and interactive triage using MCP elicitation.

Includes:

• /local-review — multi-agent code review with configurable reviewers
• /doc-review — document quality review
• /triage — interactive finding triage (git add --patch for code reviews)
• 4 bundled reviewer agents (code quality, security, testing, documentation)

Requirements

• Claude Code v2.1.76+ (for MCP elicitation support)
• Node.js 20+

Install

claude plugin install ExtractableMedia/review-toolkit

Manual install (development)

git clone https://github.com/ExtractableMedia/review-toolkit.git
cd review-toolkit
npm install
npm run build

# Test locally without installing
claude --plugin-dir /path/to/review-toolkit

Commands

/local-review

Dispatches multiple reviewer agents in parallel to analyze your branch changes, then collates findings into a single local-review.md with numbered findings, severity indicators, and a pre-merge checklist.

/local-review
/local-review HEAD~3..HEAD
/local-review app/controllers/

Built-in reviewers (always dispatched):

Agent	Focus
code-best-practices-reviewer	Code quality, SOLID, DRY, naming, complexity
security-reviewer	OWASP Top 10, injection, auth, data exposure
test-suite-architect	Test coverage gaps, test quality, strategy
documentation-expert	Collates all findings into the review document

Project-specific reviewers are configured via .claude/review-config.md (see Configuration below).

/doc-review

Reviews a document for formatting, consistency, accuracy, clarity, sensitive information, spelling/grammar, and staleness.

/doc-review README.md
/doc-review docs/architecture.md

Outputs a *-DOC-REVIEW.md file with numbered findings using the same format as /local-review.

/triage

Interactively triage findings from a review file, presenting each finding one at a time with an action menu via MCP elicitation.

/triage
/triage local-review.md
/triage my-doc-DOC-REVIEW.md

Actions per finding:

• Fix — dispatch agent to resolve after triage
• Fix with guidance — add context for the fixing agent
• Accept — mark as acceptable (no fix needed)
• Defer — address later
• Ignore — won't fix
• Skip — decide later

The tool updates the review file in place with status markers for accept/defer/ignore decisions.

Bundled Agents

The plugin ships 4 generic reviewer agents that work across any codebase:

Agent	Purpose
code-best-practices-reviewer	Code organization, SOLID, DRY, naming, error handling
security-reviewer	OWASP Top 10, injection, auth, data exposure, dependencies
test-suite-architect	Test creation, coverage analysis, test strategy, refactoring
documentation-expert	Review collation, document formatting, merge logic

Framework-specific agents (Rails, PostgreSQL, etc.) are not bundled — add them to your project via .claude/review-config.md.

Project Configuration

Create .claude/review-config.md in your project to add specialized reviewers beyond the 4 built-in ones. The config supports two sections:

Always Run

Agents dispatched on every review:

## Always Run

### ruby-on-rails-expert
- Rails conventions, Active Record patterns, controller design
- Model responsibilities, performance considerations

Conditional

Agents dispatched only when the change set includes files matching trigger patterns:

## Conditional

### ui-ux-design-specialist
**Trigger:** `app/views/**`, `app/assets/stylesheets/**`,
`app/javascript/**`, `spec/system/**`
- Visual consistency, accessibility, responsive design
- User feedback, interaction patterns

### postgresql-expert
**Trigger:** `db/migrate/**`, `db/schema.rb`
- Migration safety, index strategy, data type choices

Configured agents must be available in the project's .claude/agents/ directory or the user's ~/.claude/agents/.

See examples/review-config.md for a complete example.

Finding Format

All review commands use the same output format:

### F1 🟡 Medium Priority - Missing input validation

**File:** `app/controllers/users_controller.rb` (line 45)
**Issue:** Description of the problem
**Suggestion:** How to fix it

Severity indicators:

• 🔴 Critical — must fix
• 🟠 High Priority — should fix before merge
• 🟡 Medium Priority — should address
• 🟢 Low Priority — can address later
• ℹ️ Observation — positive note, no action needed

Status markers (applied when findings are resolved):

• ~~heading~~ ✅ Fixed
• ~~heading~~ 🚫 Ignored
• ~~heading~~ ⏸️ Deferred

Hooks

Claude Code's Elicitation and ElicitationResult hooks can customize triage behavior. See examples/hooks/ for:

Auto-skip low-priority findings

examples/hooks/auto-skip-low-priority.sh — automatically skips 🟢 Low Priority findings so you only see Critical/High/Medium in the interactive triage.

Log triage decisions

examples/hooks/log-triage-decisions.sh — appends every triage decision to ~/.claude/triage-log.jsonl for auditing.

Configuration

Add hooks to your .claude/settings.json:

{
  "hooks": {
    "Elicitation": [
      {
        "matcher": "review-toolkit",
        "hooks": [
          {
            "type": "command",
            "command": "/path/to/auto-skip-low-priority.sh"
          }
        ]
      }
    ]
  }
}

How It Works

/local-review  →  dispatches reviewer agents in parallel
                       ↓
               documentation-expert collates findings
                       ↓
               writes local-review.md
                       ↓
/triage  →  calls MCP tool "triage_findings"
                       ↓
              MCP server reads file, loops:
                ┌─ elicitation/create (finding + action menu)
                │     ↓  Elicitation hook (auto-skip?)
                │  user picks action
                │     ↓  ElicitationResult hook (log?)
                │  if "fix with guidance" → 2nd elicitation
                │  update file markers
                └─ repeat
                       ↓
              returns structured triage results
                       ↓
              Claude fixes findings marked "fix"

Development

npm install
npm run build

License

MIT