Parallel Code Review with Codex and Gemini

Run comprehensive code reviews using GPT-5.2-Codex and Gemini 3 Pro in parallel, then consolidate findings into a unified report.

Session Naming

Before starting, rename this session for clarity:

• If $ARGUMENTS provided: /rename "Review: $ARGUMENTS"
• Otherwise: /rename "Review: Current Changes"

Instructions

1. Get the Diff to Review

Capture the changes to review:

# Default: all staged + unstaged changes
git diff HEAD

# If $ARGUMENTS specifies a commit range (e.g., "origin/main...HEAD"), use that instead

If $ARGUMENTS contains a commit range or file paths, use those. Otherwise default to git diff HEAD.

Store the diff output for both reviewers.

2. Spawn Parallel Review Agents

Create two Task agents to run reviews concurrently. IMPORTANT: Spawn both agents in a single message to run them in parallel.

Codex Review Agent

Use the codex skill with these settings:

• Model: gpt-5.2-codex
• Sandbox: read-only
• Reasoning effort: xhigh
• Full auto mode

Codex command:

codex exec --skip-git-repo-check \
  -m gpt-5.2-codex \
  -c model_reasoning_effort="xhigh" \
  --sandbox read-only \
  --full-auto \
  "You are a code reviewer. Review the following git diff for issues.

Focus on:
- Correctness and logic errors
- Performance implications
- Security vulnerabilities
- Maintainability concerns

Flag only actionable issues INTRODUCED by this change. Skip trivial nits.

For each finding, provide:
### [TITLE] (P{0-3}, confidence: {0.0-1.0})
**File**: \`path/to/file\` lines {start}-{end}
{explanation}

Priority levels:
- P0: Critical (security, data loss, crashes)
- P1: High (logic errors, significant bugs)
- P2: Medium (code quality, maintainability)
- P3: Low (style, minor suggestions)

End with:
## Overall Verdict
**Assessment**: [patch is correct / patch is incorrect]
**Confidence**: {0.0-1.0}
**Justification**: {brief explanation}

---
GIT DIFF:
$(git diff HEAD)" 2>/dev/null

Gemini Review Agent

Use the gemini skill with these settings:

• Model: gemini-3-pro-preview
• Approval mode: yolo (required for background execution)
• Timeout: 300s safety wrapper

Gemini command:

timeout 300 gemini -m gemini-3-pro-preview --approval-mode yolo \
  "You are a code reviewer. Review the following git diff for issues.

Focus on:
- Correctness and logic errors
- Performance implications
- Security vulnerabilities
- Maintainability concerns

Flag only actionable issues INTRODUCED by this change. Skip trivial nits.

For each finding, provide:
### [TITLE] (P{0-3}, confidence: {0.0-1.0})
**File**: \`path/to/file\` lines {start}-{end}
{explanation}

Priority levels:
- P0: Critical (security, data loss, crashes)
- P1: High (logic errors, significant bugs)
- P2: Medium (code quality, maintainability)
- P3: Low (style, minor suggestions)

End with:
## Overall Verdict
**Assessment**: [APPROVE / REQUEST_CHANGES]
**Confidence**: {0.0-1.0}
**Justification**: {brief explanation}

---
GIT DIFF:
$(git diff HEAD)"

3. Wait for Both Reviews

CRITICAL: Wait for ALL review agents to complete before proceeding. Do not start consolidation until both have returned.

4. Consolidate Findings

Merge the two review outputs:

1. Parse findings from each reviewer (look for ### [TITLE] pattern)
2. Deduplicate overlapping issues:
   • Same file + overlapping line ranges + similar issue = merge into one
   • Note which reviewers flagged it
3. Flag consensus: When both reviewers identify the same issue, mark as "consensus" (higher confidence)
4. Sort by priority: P0 first, then P1, P2, P3
5. Within priority: Sort by confidence score (highest first)

5. Generate Consolidated Report

Output the final report in this format:

# Code Review Report

**Scope**: `git diff HEAD` (or specified range)
**Date**: [current timestamp]
**Reviewers**: GPT-5.2-Codex (xhigh), Gemini 3 Pro

## Summary

| Priority | Count | Description |
|----------|-------|-------------|
| P0 | X | Critical issues requiring immediate attention |
| P1 | X | High priority issues |
| P2 | X | Medium priority issues |
| P3 | X | Low priority / suggestions |

**Overall Verdict**:
- Codex: [patch is correct/incorrect] (confidence: X.XX)
- Gemini: [APPROVE/REQUEST_CHANGES] (confidence: X.XX)
- **Consensus**: [APPROVE / REQUEST_CHANGES / MIXED]

## Critical Issues (P0)

### [Title]
**File**: `path/to/file.ts` lines 45-52
**Flagged by**: Codex, Gemini (consensus)
**Confidence**: 0.95

[Detailed explanation merging insights from both reviewers]

---

## High Priority (P1)

[findings...]

## Medium Priority (P2)

[findings...]

## Low Priority (P3)

[findings...]

## Reviewer Disagreements

[Any findings where reviewers had significantly different assessments - one flagged, one didn't, or different priorities]

---
*Generated by /sdlc:review using GPT-5.2-Codex and Gemini 3 Pro*

6. Present Results

After generating the report:

• Display the full consolidated report
• Highlight any P0/P1 issues that need immediate attention
• If no issues found: "No significant issues found. Code looks good to proceed."

Scope

$ARGUMENTS

Notes

• If the diff is empty, inform the user: "No changes to review. Stage some changes or specify a commit range."
• For very large diffs (>10k lines), consider reviewing in chunks or warning the user about token limits
• Both reviewers run in read-only mode - they cannot modify files
• Codex uses xhigh reasoning for maximum review quality
• Gemini uses yolo mode for background execution (will not hang)