gemini-peer-review

What it does

Leverages Google's Gemini CLI to provide a second AI perspective for peer review, architecture validation, design cross-validation, security analysis, and holistic codebase assessment. It is intended for high-stakes decisions, complex architectures, and large codebases, and must be explicitly invoked using the skill syntax.

How it works

• The skill is invoked explicitly by Claude using:

skill: "gemini-peer-review"

• When invoked, Claude prepares a context for Gemini, including relevant code, architecture diagrams, and specific questions or concerns.
• It uses Gemini CLI commands (via non-interactive prompts or multi-line prompts) to perform analysis, potentially including multimodal inputs (diagrams, designs) and full-context reviews up to large codebases.
• The workflow supports loading preparation guidelines, context structure templates, and common Gemini invocation patterns from referenced guidance, then executing a CLI prompt such as:

cat <<'EOF' | gemini
[prepared context and question here]
EOF

• Outputs from Gemini are then synthesized with Claude's analysis using a synthesis framework to produce an integrated recommendation, noting agreements, disagreements, complementary insights, and actionable next steps.

• The skill emphasizes explicit invocation, handling of potential Gemini unavailability or API limits, and structured synthesis outputs.

When to use it

• Use when making high-stakes architecture decisions, evaluating multiple design alternatives, reviewing security-critical code, validating complex refactors, or analyzing very large codebases requiring extensive context.
• Do not use for simple, straightforward implementations, when a single opinion suffices, or when Gemini CLI is unavailable.
• The skill explicitly instructs Claude to invoke Gemini only through the defined skill syntax and prompts.

What it can touch

• Gemini CLI interaction through terminal-based commands as per the provided invocation patterns.
• Contextual input including code modules, architecture descriptions, and multimodal assets if available, formatted per Gemini guidance.

Caveats

• Requires explicit invocation and terminal access to Gemini CLI.
• Output depends on Gemini availability and API limits; potential fallbacks to Claude-only analysis may occur if unavailable.
• The approach relies on cross-AI synthesis; users should consider both perspectives and apply integrated recommendations.