secondpass_multireview

Second Pass Multi-Review

Get a second opinion on your current plan from other AI models via OpenRouter.

Steps

1. Gather context — Build a comprehensive review payload with THREE sections. All three are required — do not skip the diff or code context.

   Section 1: Plan summary — Summarize from the conversation:

   • What is being changed and why (the bug, feature, or goal)
   • Key implementation details and approach
   • File paths and functions affected
   • Any trade-offs or alternatives considered

   Section 2: Code diff — Get the actual changes:

   • If changes are already committed: git diff HEAD~1 (or appropriate range)
   • If changes are staged/unstaged: git diff and/or git diff --cached
   • If this is a pre-implementation plan, skip this section

   Section 3: Relevant code context — Read and include surrounding code that reviewers need to understand the change:

   • The full function/class containing each change (not just the changed lines)
   • Any callers or dependencies that interact with the changed code
   • Related tool/prompt definitions if the change affects AI behavior
   • Keep each snippet to ~50 lines max, use # ... (trimmed) for long sections

2. Locate the script — Find the review script. It ships with this plugin:

   SCRIPT=$(find ~/.claude/plugins -path "*/secondpass-multireview/scripts/secondpass_multireview.py" 2>/dev/null | head -1)
   

   If not found in plugins, check the project directory:

   SCRIPT=${SCRIPT:-$(find . -path "*/secondpass_multireview.py" 2>/dev/null | head -1)}
   

3. Write a temp file and send — Assemble all sections into a temp file, then pipe to the script:

   source ~/.zshrc 2>/dev/null && python3 "$SCRIPT" < /tmp/secondpass_review.md
   

   Format the temp file (/tmp/secondpass_review.md) as:

   ## Plan
   <plan summary>
   
   ## Diff
   ```diff
   <git diff output>
   

   Code Context

   filename:function_name

   <relevant surrounding code>
   

   
   The script queries these models in parallel via OpenRouter:
   - **GPT-5.3 Codex** (OpenAI)
   - **Gemini 3.1 Pro** (Google)
   
   Override models with: `SECONDPASS_MODELS="model/a,model/b" python3 "$SCRIPT"`
   
   

4. Present results — Show each model's feedback clearly, organized by model. Highlight any:

   • Consensus across models (things multiple models flag)
   • Disagreements worth considering
   • Actionable suggestions

5. Summarize — End with a brief synthesis: what the models agree on, key concerns raised, and whether the plan should be adjusted.

Requirements

• OPENROUTER_API_KEY environment variable must be set
• No external Python dependencies required (uses stdlib only)