review-local-changes

Review uncommitted local changes using all specialized agents with code improvement suggestions.

• Purpose - Comprehensive review before committing

• Output - Structured report with findings by severity

/code-review:review-local-changes [review-aspects] [--min-impact critical|high|medium|medium-low|low] [--json]

Arguments

Argument	Format	Default	Description
review-aspects	Free text	None	Optional review aspects or focus areas (e.g., "security, performance")
--min-impact	--min-impact <level>	high	Minimum impact level for reported issues. Values: critical, high, medium, medium-low, low
--json	Flag	false	Output results in JSON format instead of markdown

Impact Level Mapping

Level	Impact Score Range
critical	81-100
high	61-80
medium	41-60
medium-low	21-40
low	0-20

How It Works

1. Change Detection: Identifies all uncommitted changes in the working directory

   • Staged changes

   • Unstaged modifications

   • New files

2. Parallel Agent Analysis: Spawns six specialized agents simultaneously

   • Bug Hunter - Identifies potential bugs and edge cases

   • Security Auditor - Finds security vulnerabilities

   • Test Coverage Reviewer - Evaluates test coverage

   • Code Quality Reviewer - Assesses code structure

   • Contracts Reviewer - Reviews API contracts

   • Historical Context Reviewer - Analyzes codebase patterns

3. Finding Aggregation: Combines all agent reports

   • Categorizes by severity (Critical, High, Medium, Medium-Low, Low)

   • Scores each issue for confidence (is it real?) and impact (how severe?)

   • Removes duplicates

   • Adds file and line references

4. Filtering: Applies two sequential filters to reduce noise

   • Min-impact cutoff - Excludes issues below the --min-impact threshold (default: high, score 61+)

   • Progressive confidence threshold - Higher-impact issues require less confidence to pass (Critical: 50%, High: 65%, Medium: 75%, Medium-Low: 85%, Low: 95%)

5. Report Generation: Produces actionable report in markdown (default) or JSON (--json) format with prioritized findings

Usage Examples

# Review all local changes (default: --min-impact high, markdown output)
> /code-review:review-local-changes

# Focus on security aspects
> /code-review:review-local-changes security

# Lower the threshold to catch medium-impact issues
> /code-review:review-local-changes --min-impact medium

# Focus on security and performance, medium threshold
> /code-review:review-local-changes security, performance --min-impact medium

# Critical-only issues in JSON for programmatic consumption
> /code-review:review-local-changes --min-impact critical --json

# After implementing a feature
> claude "implement user authentication"
> /code-review:review-local-changes

JSON Output

When using --json, the output is a structured object with these top-level fields:

• quality_gate - "PASS" or "FAIL" (fails when any critical or high issue exists)

• summary - Issue counts by severity

• issues - Array of issues with severity, file, lines, description, evidence, impact_score, confidence_score, and optional suggestion

• improvements - Array of code improvement suggestions from the code-quality-reviewer agent

Best Practices

• Review before committing - Run review on local changes before git commit

• Address critical issues first - Fix Critical and High priority findings immediately

• Iterate after fixes - Run again to verify issues are resolved

• Combine with reflexion - Use /reflexion:memorize to save patterns for future reference