team-reviewer

Team Reviewer

Scatter-gather code review team. Spawns specialized review agents in parallel, aggregates findings, and outputs a scored report.

Usage

/team-reviewer                    # Auto-detect (PR or local diff), deep mode
/team-reviewer --quick            # Auto-detect, quick mode (3 agents)
/team-reviewer 142                # Review PR #142, deep mode
/team-reviewer 142 --quick        # Review PR #142, quick mode

Parse Arguments

Extract from: {{ARGUMENTS}}

PR_NUMBER = first numeric argument (e.g., 142) or empty
QUICK_MODE = true if --quick flag present

If no arguments: auto-detect scope, deep mode.

---

Phase 1: DETECT Scope

Determine what code to review. Run these commands:

# 1. If PR_NUMBER provided:
gh pr diff {PR_NUMBER} --name-only    # files list
gh pr diff {PR_NUMBER}                 # full diff
gh pr view {PR_NUMBER} --json title,body,commits  # PR metadata

# 2. If NO PR_NUMBER, check if branch has a PR:
gh pr list --head $(git branch --show-current) --json number,title --limit 1

# 2a. If PR found → use that PR number, run commands from step 1
# 2b. If NO PR found → local diff:
git diff HEAD --name-only              # files list
git diff HEAD                          # full diff (staged + unstaged)
git log --oneline -10                  # recent commits for context

Store:

• DIFF — the full diff content
• FILES — list of changed files
• SCOPE_LABEL — "PR #N (title)" or "Local diff (branch-name)"
• STATS — file count, lines added/removed

If diff is empty, report "No changes to review" and stop.

---

Phase 2: SCATTER (Spawn Agents)

Agent Selection

If QUICK_MODE:
  agents = [code-quality, security, silent-failure]          # 3 agents
Else:
  agents = [code-quality, security, silent-failure,
            type-design, test-coverage, commit-hygiene]      # 6 agents

Create Team and Spawn

1. TeamCreate with name team-reviewer
2. TaskCreate one task per agent
3. Spawn ALL agents in parallel using Task tool with:
   • team_name: "team-reviewer"
   • subagent_type: "team-reviewer-{agent-name}"
   • run_in_background: true
   • model: "sonnet" (cost-efficient for review tasks)