uniswap/ai-toolkit/analyze-code

Code Analyzer

Provide comprehensive code explanation through multi-agent analysis for architecture, patterns, security, and performance insights.

Execution Steps

1. Identify the target — Extract the file path or module name from the user's request. If no target is specified, ask: "Which file or module would you like me to analyze?"

1. Verify the target exists — Use Read or Glob to confirm the file exists. If it doesn't, report and stop: "Could not find <path>. Please check the path and try again."

1. Infer analysis depth from the user's request:

• overview — request uses words like "quick", "summary", "briefly", "what does this do"
• deep — request mentions "security", "performance", "vulnerabilities", "thorough", "deep", "comprehensive"
• architectural — request mentions "architecture", "system", "how does this fit", "overall design"
• Default to `overview` when depth is ambiguous. deep spawns three agents; running

that for "what does this file do" costs three context loads to answer a one-agent question. Escalate to deep only on evidence — the user asked for it, overview surfaced a concrete security or performance concern, or the target is a trust boundary (auth, payments, external input parsing).

1. Dispatch agents based on depth:

• overview: spawn code-explainer-agent only; pass the file path and its contents
• deep: spawn code-explainer-agent + security-analyzer-agent + performance-analyzer-agent in parallel
• architectural: run context-loader-agent first to gather system-level context, then spawn code-explainer-agent + security-analyzer-agent + performance-analyzer-agent in parallel with that context

1. Synthesize results — Combine agent outputs into a single structured report. Integrate and deduplicate findings; do not dump raw agent output.

1. Handle agent failures — If an agent errors or returns empty output, note it in the report ("Security analysis unavailable") and continue with remaining results.

Analysis Depth

Focus Areas

• patterns: Design patterns, SOLID, anti-patterns
• security: Vulnerabilities, input validation, auth
• performance: Algorithm complexity, memory, queries
• all: Complete analysis (default)

Output Includes

• Summary: Purpose and the concrete complexity signals observed (function count, nesting

depth, file length). Do not emit a maintainability score — there is no measurement behind it.

• Architecture: Patterns, layers, coupling analysis
• Functionality: Main purpose, data flow, side effects
• Dependencies: Imports, exports, circular dependencies
• Risks: Security, performance, maintainability issues
• Improvements: Quick wins and refactoring suggestions

Delegation

For simple explanations, use code-explainer-agent directly.

For deep analysis, coordinate:

• code-explainer-agent: Architecture and patterns
• security-analyzer-agent: Vulnerability assessment
• performance-analyzer-agent: Complexity analysis
• context-loader-agent: System-level context

Examples

"Explain src/auth/login.ts to me"
"What does this API endpoint do?"
"Analyze the security of the payment module"
"Is there anything wrong with this code?"

Detailed Reference

For output schemas and advanced options, see explain-file-guide.md.