feat: create team-based parallel PR review plugin

[JacobPEvans-personal]

Parent

Part of #14 (epic: migrate workaround orchestration patterns to native Claude Agent Teams)

Summary

Create a new team-review plugin that uses Agent Teams to run multi-perspective PR reviews in parallel - security, performance, test coverage, and code quality each owned by a dedicated teammate.

Problem

Current PR review in the ecosystem (from parent CLAUDE.md commands like /review-pr, /review-code) uses:

• Sequential subagent dispatch (one perspective at a time)
• Each reviewer reports back independently to main context
• No debate or cross-checking between reviewers
• Main agent must synthesize all findings alone

Proposed Solution

Plugin Structure

team-review/
├── .claude-plugin/plugin.json
├── commands/
│   └── team-review.md           # /team-review command
├── skills/
│   └── review-perspectives.md   # Review criteria per perspective
└── README.md

Team Architecture

/team-review PR#123 (team lead - delegate mode)
├── security-reviewer (teammate)
│   ├── Focus: OWASP, auth, secrets, injection
│   └── Model: sonnet
├── performance-reviewer (teammate)
│   ├── Focus: N+1 queries, caching, algorithmic complexity
│   └── Model: sonnet
├── test-reviewer (teammate)
│   ├── Focus: Coverage gaps, edge cases, test quality
│   └── Model: sonnet
└── quality-reviewer (teammate)
    ├── Focus: Code style, patterns, maintainability
    └── Model: haiku

Workflow

1. Lead creates team and assigns PR context to all teammates
2. Each teammate reviews from their perspective (parallel)
3. Teammates debate findings via peer-to-peer messaging
   • "I found a SQL injection in auth.py:42, @performance-reviewer is this also a query perf issue?"
   • "The test for this function is missing the error path, @security-reviewer should we flag this as a security gap too?"
4. Lead synthesizes consensus findings
5. Lead posts structured review comment on PR

Key Features

• Competing hypotheses: Teammates can challenge each other's findings
• Cross-domain insights: Security + performance reviewer can identify issues neither would find alone
• Severity consensus: Multiple reviewers agree on severity ratings
• Structured output: Unified review with findings organized by severity

Agent Teams Advantages Over Subagents

Feature	Subagent Review	Team Review
Communication	One-way (back to main)	Peer-to-peer debate
Cross-checking	Manual by main agent	Automatic between teammates
Finding quality	Individual perspectives	Cross-pollinated insights
Token cost	Lower	~4x higher
Review depth	Shallow (focused)	Deep (collaborative)

Acceptance Criteria

• /team-review accepts PR number argument
• Spawns appropriate team for PR size/complexity
• Teammates review in parallel from different perspectives
• Inter-agent debate produces cross-domain findings
• Lead synthesizes into structured review format
• Posts review as PR comment (with user confirmation)
• Graceful fallback to sequential subagent review when teams disabled

Dependencies

• Depends on: agent-teams-orchestrator plugin (feat: create agent-teams-orchestrator plugin with team lifecycle management #15) for lifecycle patterns

[JacobPEvans-personal]

Parent epic: #14 | Depends on: #15

[claude]

This issue has had no activity for 30+ days. Is this still relevant?

[claude]

This issue and #16 could potentially be addressed together as a single effort. Both create code review workflows using Agent Teams with peer-to-peer coordination, share similar team architectures (security/performance/test reviewer roles), and both depend on the same foundation (#15). Developing them together would allow sharing template patterns and avoiding duplication.

[JacobPEvans-personal]

Closing: Claude Code's Agent Teams feature works natively without custom plugins, skills, hooks, or orchestration. The custom Agent-Teams plugin work tracked here is no longer needed — native primitives cover these patterns. No follow-up planned.