kamushadenes/deep-review

Deep Code Review Workflow

Comprehensive review using 9 specialized agents for thorough analysis.

Step 1: Determine scope

Ask the user what to review:

• uncommitted: Review git diff HEAD
• branch: Review git diff ${base_branch}...HEAD
• codebase: Review key files and architecture

If the user doesn't specify, default to branch changes if on a feature branch, otherwise uncommitted changes.

Step 2: Gather context

Based on scope, collect:

# For uncommitted
git diff HEAD

# For branch
git remote show origin | grep 'HEAD branch' | awk '{print $NF}'
git diff ${base_branch}...HEAD
git log --oneline $(git merge-base HEAD origin/${base_branch})..HEAD

# For codebase
# Read key files, package configs, entry points

Also gather:

git diff ${scope} --stat   # Changed files summary

Step 3: Spawn 9 reviewer agents

Spawn all agents in parallel, each receiving the review context and their focus area:

Agent 1: Code Reviewer

Logic errors, edge cases, error handling, race conditions, resource leaks, API contracts.

Agent 2: Security Auditor

Injection, auth, secrets, input validation, SSRF/XSS/CSRF, unsafe patterns.

Agent 3: Test Analyzer

Coverage gaps, missing edge case tests, test quality, mocking correctness, assertions.

Agent 4: Performance Analyzer

Algorithm complexity, unnecessary allocations, N+1 queries, missing caching, blocking operations, memory leaks.

Agent 5: Silent Failure Hunter

Swallowed exceptions, ignored returns, missing error propagation, logging gaps.

Agent 6: Type Checker

Type safety, implicit coercions, generic misuse, null safety, type narrowing gaps.

Agent 7: Code Simplifier

Unnecessary complexity, over-abstraction, dead code, redundant logic, opportunities to simplify.

Agent 8: Refactoring Advisor

Code duplication, single responsibility violations, coupling issues, naming, module boundaries.

Agent 9: Dependency Checker

Outdated dependencies, known vulnerabilities, unused dependencies, version conflicts, license issues.

Each agent returns findings as:

[SEVERITY] file:line - Description
  Context: <code snippet>
  Suggestion: <fix>
  Confidence: HIGH/MEDIUM/LOW

Step 4: Validate findings

After all 9 agents complete, run a suggestion-critic agent:

• Remove false positives
• Remove findings that contradict project conventions
• Verify suggested fixes are correct
• Flag disagreements
• Note consensus (issues flagged by multiple agents)

Step 5: Present validated findings

Executive Summary

Brief overview of code health across all dimensions.

Critical Issues

Must fix. Include consensus markers if multiple agents flagged the same issue.

Warnings

Should address.

Suggestions

Nice-to-have improvements.

Positive Notes

Good patterns and practices observed.

Consensus Analysis

Issues identified by 2+ agents, indicating higher confidence.

Step 6: Offer next steps

• Fix Critical/High issues
• Post findings as PR review comments (if applicable)
• Generate a detailed report