@fengrru/valid8

4-layer output validation for AI agents: syntax, semantic, runtime, security.

Install

npm install @fengrru/valid8
# Optional: npm install typescript (for AST-based syntax validation)

Quick Start

import { ValidationNetwork } from "@fengrru/valid8"

const validator = new ValidationNetwork()

const result = await validator.validate({
  content: fileContent,
  filePath: "src/index.ts",
  riskLevel: 1,
})

console.log(result.passed) // boolean
console.log(result.confidence) // 0-1
console.log(result.layers) // [{ name: "syntax", passed: true, ... }]

Validation Layers

| Layer | Risk Gate | Description | |-------|-----------|-------------| | Syntax | minRisk=0 | TypeScript AST + regex fallback | | Semantic | minRisk=2 | LLM review or keyword fallback | | Runtime | minRisk=0 | Compilation + test execution | | Security | minRisk=1 | 40+ dangerous pattern detection |

Features

• Data-driven: VALIDATION_LAYERS array, not hardcoded
• Confidence scoring: weighted average across layers
• Retry logic: configurable threshold and max retries
• Permission rules: bash blocked by default,可配置

Documentation

• API Reference — TypeDoc-generated API docs
• Source Code
• Examples

License

MIT