Review Orchestrator

Dispatches multiple review subagents in parallel, compiles and deduplicates their feedback, and presents a triaged list for human decision.

Configuration

Uses the production API by default. No configuration needed for read operations.

Defaults (no setup required):

• API URL: https://api-gateway-856475788601.us-central1.run.app
• Read-only endpoints at /public/* require no authentication

For write operations (transitions), set:

export PIPELINE_API_KEY="..."  # Get from GCP Secret Manager or ask admin

When to Use

• Quality gates have passed (lint, typecheck, tests)
• Code changes are ready for review
• Before creating a PR

Prerequisites

• CodeRabbit CLI installed: which coderabbit
• Authenticated: coderabbit auth login
• Pro tier allows 8 reviews/hour

Workflow

1. Verify Readiness

# Check CodeRabbit installed
which coderabbit

# Check there are changes to review
git status --porcelain
git diff --stat HEAD~1  # or vs base branch

If no changes or CodeRabbit missing, stop and inform user.

2. Gather Context

Get ticket context from the public API (no auth required):

const API_URL = process.env.PIPELINE_API_URL || 'https://api-gateway-856475788601.us-central1.run.app'
const response = await fetch(`${API_URL}/public/tickets/${ticketId}`)
const ticket = await response.json()
const ticketRequirements = ticket.description

Also collect:

• List of changed files: git diff --name-only HEAD~1
• Plan file if available locally

3. Dispatch Review Subagents

Use the Task tool to dispatch three parallel review subagents:

Subagent 1: CodeRabbit Review

Run CodeRabbit CLI review on the current changes.

Commands:
cd {REPO_PATH}
coderabbit --prompt-only --type uncommitted

Parse output and categorize findings by severity:
- Critical: Security, race conditions, data loss
- Major: Logic errors, performance, missing error handling
- Minor: Code style, naming
- Nitpick: Formatting preferences

Return structured list with file:line, issue, suggested fix.

Subagent 2: Agent Code Review

Perform comprehensive code review of the changes.

Files to review: {FILES_LIST}
Ticket requirements: {TICKET_REQUIREMENTS}

Review Checklist:
1. Functionality: Implements requirements, handles edge cases, proper error handling
2. Code Quality: TypeScript (no any), Vue (Composition API), Tailwind (no dark:), naming conventions
3. Architecture: Follows existing patterns, no over-engineering, clean separation
4. Testing: Tests added, behavioral not change detectors, edge cases covered
5. Security: No secrets exposed, input validation, no XSS/injection
6. Performance: No unnecessary re-renders, efficient data structures

Return findings with severity (Critical/Major/Minor), location (file:line), and fix suggestion.

Subagent 3: Pattern Compliance Review

Check code against target repository patterns and AGENTS.md guidelines.

Files to review: {FILES_LIST}

Check for violations:
- Vue: Reactive props destructuring (not withDefaults), separate type imports
- Tailwind: No dark: variants (use semantic themes), use cn() utility, no !important
- TypeScript: No any types, no as any assertions
- State: No unnecessary computed/watch when simpler works
- Testing: Behavioral tests, not change detectors or mock tests

Return violations with file:line and the correct pattern to use.

4. Compile and Deduplicate

After all subagents complete:

1. Merge findings from all three reviewers
2. Remove duplicates: Same issue found by multiple reviewers (note "Found by: CodeRabbit, Agent")
3. Resolve conflicts: If reviewers disagree, note both perspectives with recommendation
4. Group by file: Organize findings by file path
5. Sort by severity: Critical → Major → Minor → Nitpick

5. Generate Review Summary

Create structured output:

# Code Review Summary

## Statistics

- Total findings: {count}
- Critical: {count}
- Major: {count}
- Minor: {count}
- Nitpicks: {count}
- Duplicates removed: {count}

## Critical Issues (Must Fix)

### [C1] {Title}

- **File:** `path/to/file.ts:42`
- **Source:** CodeRabbit / Agent / Pattern
- **Issue:** {description}
- **Fix:**

```typescript
// Suggested fix
```

Major Issues (Should Fix)

[M1] {Title}

• File: path/to/file.ts:55
• Source: {source}
• Issue: {description}
• Fix: {suggestion}

Minor Issues (Consider)

[m1] {Title}

• File: path/to/file.ts:10
• Issue: {description}
• Fix: {suggestion}

Nitpicks (Optional)

• [N1] file.ts:10 - {description}
• [N2] file.ts:20 - {description}

Conflicting Opinions

{Topic}

• CodeRabbit says: {opinion}
• Agent says: {opinion}
• Recommendation: {which to follow and why}

### 6. Present Triage Interface

Review complete. {X} findings across {Y} files.

For each finding, respond with:

• Numbers to implement (e.g., "C1, M1, M3")
• "all critical" - implement all critical
• "all major" - implement all critical + major
• "all" - implement everything
• "skip N1, N2" - skip specific items
• "clarify M2" - need more info on item

Your response:

Wait for human decision.

### 7. Save Review Locally

Save full review to local `review-comments.md`:
- Complete review summary
- All findings with full details
- Human's triage decisions
- Timestamp

```markdown
## Triage Decisions

- Selected: ["C1", "M1", "M2"]
- Skipped: ["N1", "N2"]
- Timestamp: {timestamp}

8. Handle Implementation

For selected items:

• Dispatch fix subagents for each selected item, OR
• Return to implementation phase with specific fix list

After fixes, re-run affected quality gates to verify.

Severity Definitions

Rate Limiting

CodeRabbit Pro: 8 reviews/hour. Track usage and warn if approaching limit.

Fallback

If CodeRabbit is unavailable:

1. Skip CodeRabbit subagent
2. Note in review summary: "CodeRabbit review skipped: {reason}"
3. Continue with Agent and Pattern reviews

Local Working Artifacts

Notes

• Review orchestrator does not change ticket state—it stays in IMPLEMENTATION
• Ticket context comes from the API
• Review output is saved locally as a working artifact