gemini-peer-review

gemini-peer-review

Gemini Peer Review

Consult Gemini as a coding peer for a second opinion on code quality, architecture decisions, debugging, or security reviews.

Setup

API Key: Set GEMINI_API_KEY as an environment variable. Get a key from https://aistudio.google.com/apikey if you don't have one.

export GEMINI_API_KEY="your-key-here"

Workflow

Determine mode from user request (review, architect, debug, security, quick)

Read target files into context

Build prompt using the AI-to-AI template from references/prompt-templates.md

Write prompt to file at .claude/artifacts/gemini-prompt.txt (avoids shell escaping issues)

Call the API — generate a Python script that:

Reads GEMINI_API_KEY from environment

• Reads the prompt from .claude/artifacts/gemini-prompt.txt

• POSTs to https://generativelanguage.googleapis.com/v1beta/models/{model}:generateContent

• Payload: {"contents": [{"parts": [{"text": prompt}]}], "generationConfig": {"temperature": 0.3, "maxOutputTokens": 8192}}

• Extracts text from candidates[0].content.parts[0].text

• Prints result to stdout

Write the script to .claude/scripts/gemini-review.py and run it.

Synthesize — present Gemini's findings, add your own perspective (agree/disagree), let the user decide what to implement

Modes

Code Review

Review specific files for bugs, logic errors, security vulnerabilities, performance issues, and best practice violations.

Read the target files, build a prompt using the Code Review template, call with gemini-2.5-flash.

Architecture Advice

Get feedback on design decisions with trade-off analysis. Include project context (CLAUDE.md, relevant source files).

Read project context, build a prompt using the Architecture template, call with gemini-2.5-pro.

Debugging Help

Analyse errors when stuck after 2+ failed fix attempts. Gemini sees the code fresh without your debugging context bias.

Read the problematic files, build a prompt using the Debug template (include error message and previous attempts), call with gemini-2.5-flash.

Security Scan

Scan code for security vulnerabilities (injection, auth bypass, data exposure).

Read the target directory's source files, build a prompt using the Security template, call with gemini-2.5-pro.

Quick Question

Fast question without file context. Build prompt inline, write to file, call with gemini-2.5-flash.

Model Selection

Mode Model Why

review, debug, quick gemini-2.5-flash Fast, good for straightforward analysis

architect, security-scan gemini-2.5-pro Better reasoning for complex trade-offs

Check current model IDs if errors occur — they change frequently:

curl -s "https://generativelanguage.googleapis.com/v1beta/models?key=$GEMINI_API_KEY" | python3 -c "import sys,json; [print(m['name']) for m in json.load(sys.stdin)['models'] if 'gemini' in m['name']]"

When to Use

Good use cases:

• Before committing major changes (final review)

• When stuck debugging after multiple attempts

• Architecture decisions with multiple valid options

• Security-sensitive code review

Avoid using for:

• Simple syntax checks (Claude handles these faster)

• Every single edit (too slow, unnecessary)

• Questions with obvious answers

Prompt Construction

Critical: Always use the AI-to-AI prompting format. Write the full prompt to a file — never pass code inline via bash arguments (shell escaping will break it).

When building the prompt:

• Start with the AI-to-AI header from references/prompt-templates.md

• Append the mode-specific template

• Append the file contents with clear --- filename --- separators

• Write to .claude/artifacts/gemini-prompt.txt

• Generate and run the API call script

Reference Files

When Read

Building prompts for any mode references/prompt-templates.md

Weekly Installs200Repositoryjezweb/claude-skillsGitHub Stars618First SeenFeb 18, 2026Security AuditsGen Agent Trust HubPassSocketPassSnykPassInstalled onopencode170github-copilot166codex165gemini-cli165cursor163claude-code158