Skip to content
BaselineOS

Lang — Expression & Language

Purpose

Section titled “Purpose”

The language runtime. Governs how your AI expresses itself — vocabulary, safety boundaries, output attribution, prompt integrity. Like the display and typography system of an OS, Lang controls how things appear to users. Every interaction with Baseline Protocol starts here — input is tokenized, parsed into commands and arguments, validated against syntax rules, and dispatched to handlers.

Key interfaces

Section titled “Key interfaces”
interface ParsedInput {
  original: string;
  tokens: string[];
  command: string | null;
  options: Map<string, unknown>;
  arguments: string[];
  intent: string | null;
}


interface IntentResult {
  type: string;       // 'question' | 'action' | 'query' | 'modification' | 'deletion' | 'status'
  confidence: number; // 0.0 – 1.0
  patterns: RegExp[];
}


interface ProcessResult {
  success: boolean;
  parsed?: ParsedInput;
  intent?: IntentResult;
  result?: CommandResult;
  error?: string;
  suggestions?: string[];
}

Usage

Section titled “Usage”
import { BaselineLangSystem } from '@baselineos/lang';


const lang = new BaselineLangSystem();


const result = lang.processInput('help');
// { success: true, parsed: { command: 'help', ... } }


// Register a custom command
lang.registerCommand('deploy', {
  aliases: ['ship'],
  description: 'Deploy the current build',
  usage: 'deploy --env production',
  examples: ['deploy --env staging'],
});

LayerSweep checks

Section titled “LayerSweep checks”

The layer sweep runs three checks against Lang:

CheckWhat it validates
Input parsingInput is parseable — either as a registered command or natural language with detected intent
Injection resistanceNo prompt-injection patterns (e.g. “ignore previous instructions”) in title or description
Output safetyOutput does not contain exposed credentials (password=, api_key=, etc.)

References

Section titled “References”