@nanomind/router

Intent classification and command mapping for CLI security tools. Classifies natural language into 17 intent types including 6 security scanning intents.

Install

npm install @nanomind/router

Usage

import { classifyIntent, mapToCommand, isScanIntent } from '@nanomind/router';

const result = classifyIntent('scan this project for vulnerabilities');
// { intent: 'SCAN', confidence: 0.85, entities: {} }

const cmd = mapToCommand(result, 'hma');
// { command: 'hma secure', args: ['.'], description: 'Run security scan' }

// Check if an intent is a security scan (needs full model inference)
isScanIntent('SCAN_SKILL'); // true
isScanIntent('EXPLAIN');     // false

Intent Types

General Intents

| Intent | Example | Maps To | |--------|---------|---------| | SCAN | "scan this project" | hma secure | | FIX | "fix the credential issue" | hma secure --fix | | EXPLAIN | "what does CRED-001 mean" | hma explain CRED-001 | | TRUST_QUERY | "what's the trust level" | hma trust | | ATC_STATUS | "why is my agent level 2" | opena2a identity status | | ATTEST | "add trust to my CI" | opena2a identity attest | | PROTECT | "protect my credentials" | secretless-ai init | | REVIEW | "review my project security" | opena2a review | | HELP | "how do I use this" | --help | | CONFIG | "set my API key" | opena2a config | | UNKNOWN | (low confidence) | Show suggestions |

Security Scanning Intents (SCAN_*)

These intents require full NanoMind model inference via the daemon. Used by HackMyAgent's Semantic Compiler.

| Intent | Purpose | Used By | |--------|---------|---------| | SCAN_SKILL | Classify skill artifacts for attack patterns | SemanticCompiler | | SCAN_MCP | Analyze MCP server configurations | SemanticCompiler | | SCAN_SOUL | Evaluate governance document completeness | SemanticCompiler | | SCAN_CREDENTIAL | Detect credential exposure in context | SemanticCompiler | | SCAN_CODE | Identify code injection patterns | SemanticCompiler | | COMPILE_AST | Full AST compilation (all analyzers) | SemanticCompiler |

Confidence Scoring

The classifier returns a confidence score between 0 and 1:

| Range | Meaning | |-------|---------| | 0.8-1.0 | High confidence, route directly | | 0.5-0.8 | Medium confidence, may prompt for clarification | | 0.0-0.5 | Low confidence, show suggestions |

License

MIT