Parallel Code Review

Review $ARGUMENTS using parallel subagents for maximum speed.

Process

1. Parse Targets: Split arguments into independent review targets
2. Spawn Parallel Reviewers: Launch one subagent per target
3. Collect Results: Retrieve all reviews via TaskOutput
4. Synthesize: Merge findings into prioritized report

Execution Pattern

CRITICAL: Launch ALL Task calls in a SINGLE message for true parallelism.

For each independent target, spawn a subagent with run_in_background: true:

I'm launching N parallel code reviewers:

[Task 1]
description: "Review [target 1]"
prompt: "You are a code reviewer. Analyze [target 1] for:
- Code quality and best practices
- Potential bugs and edge cases
- Security vulnerabilities
- Performance issues
- Test coverage gaps

Provide findings with file:line references and severity levels."
run_in_background: true

[Task 2]
description: "Review [target 2]"
prompt: "You are a code reviewer. Analyze [target 2] for..."
run_in_background: true

[Task 3]
description: "Review [target 3]"
prompt: "You are a code reviewer. Analyze [target 3] for..."
run_in_background: true

After all complete, retrieve results:

TaskOutput: task_1_id
TaskOutput: task_2_id
TaskOutput: task_3_id

Output Format

## Parallel Review Results

### Summary
- **Targets reviewed**: N
- **Total findings**: X
- **Critical issues**: Y
- **Execution time**: ~Zs (parallel)

### [Target 1] Findings

#### Critical
- [Issue] at file:line - [Description]

#### Warnings
- [Issue] at file:line - [Description]

#### Suggestions
- [Improvement] at file:line - [Description]

### [Target 2] Findings
...

### Combined Priority List

1. **Critical** - [Target X] - [Issue description]
2. **Critical** - [Target Y] - [Issue description]
3. **High** - [Target Z] - [Issue description]
...

### Recommended Actions

1. Fix critical issues in [Target X] first
2. Address security concerns in [Target Y]
3. Consider performance optimizations in [Target Z]

Example Usage

/project-starter:parallel-review src/auth src/api src/db

This spawns 3 parallel reviewers (one per directory), completing in ~1/3 the time of sequential review.

When to Use

• Reviewing multiple independent modules
• Pre-merge comprehensive codebase review
• Analyzing different areas of a large PR
• Comparing implementation patterns across directories