The 'eval' tool is critically underdeveloped. No input schema is visible in the provided source code (apps/api-server/src/routes-tools.ts is referenced but not shown). The description 'Built-in evaluation tool for executing code' is generic (47 chars) and does not explain WHAT code is evaluated, WHEN to use it, or what the output structure is. No parameter documentation is visible. The tool is marked IRREVERSIBLE (destructive), yet there is no confirmation pattern, dry-run option, or permission gate documented. This violates multiple critical patterns: tool-description, constrained-input, confirmation-request, and secret-injection (code execution is a major security surface).

• Add a complete input schema with typed parameters (e.g., code: string, language: enum, timeout_ms: number). Document each parameter with constraints and examples.
• Expand description to 100-200 chars: explain what languages are supported, what the output structure is, and when to use this tool vs. alternatives.
• Implement a dry-run mode or confirmation step before executing code. Return a preview of what will execute and require explicit approval.
• Add comprehensive error handling: return actionable messages for syntax errors, timeouts, permission denials, and runtime failures. Include recovery guidance.
• Document security model: what sandboxing is in place? What system calls are blocked? What environment variables are accessible? Add audit logging for all code execution.