multi-ai-review

# Multi-AI Plan Review

After writing a plan, get feedback from Codex and Gemini. Both review the plan against the actual codebase and suggest improvements. Claude then updates the plan based on their feedback.

## Workflow

### Step 1: Claude Writes Initial Plan

Use enhanced-planning or standard plan mode to create an implementation plan. Save it to a file (e.g., `plans/feature-name.md`).

### Step 2: Gather Codebase Context for External Tools

Prepare context that helps Codex and Gemini understand the codebase:

```bash
# Create a context summary
cat > /tmp/codebase-context.txt << 'EOF'
PROJECT: $(basename $(pwd))
STRUCTURE:
$(find . -type f -name "*.ts" -o -name "*.tsx" -o -name "*.py" -o -name "*.rb" -o -name "*.go" | grep -v node_modules | grep -v .git | head -50)

KEY FILES:
$(cat README.md 2>/dev/null | head -30)
$(cat CLAUDE.md 2>/dev/null | head -30)
$(cat ARCHITECTURE.md 2>/dev/null | head -30)
EOF
```

### Step 3: Run Plan Reviews in Parallel

Launch both reviews simultaneously. Each tool receives:
1. The plan file
2. Codebase context
3. Instructions to suggest improvements

**Codex Review:**
```bash
cd <project_dir> && codex exec "Review this implementation plan against the codebase.

YOUR TASK:
1. Check if the plan accounts for existing code patterns and conventions
2. Identify missing steps or edge cases
3. Suggest any strategy improvements
4. Flag potential conflicts with existing code
5. Add any tasks the plan missed

Be specific - reference actual files and code when possible.

THE PLAN:
$(cat <plan_file>)

CODEBASE STRUCTURE:
$(find . -type f \( -name '*.ts' -o -name '*.tsx' -o -name '*.py' -o -name '*.rb' \) | grep -v node_modules | head -30)

KEY CONTEXT:
$(cat README.md CLAUDE.md 2>/dev/null | head -50)"
```

**Gemini Review:**
```bash
cd <project_dir> && gemini "You are a senior engineer reviewing an implementation plan. Your job is to improve it.

REVIEW CHECKLIST:
1. Does the plan account for existing patterns in this codebase?
2. Are there