@ai-operations/ops-policy

Business rules, autonomy levels, escalation paths, and budget tracking for AI Operations OS.

Part of AI Operations OS — autonomous business workflow orchestration with safety enforcement.

Install

npm install @ai-operations/ops-policy

Quick Start

import { AutonomyManager, BudgetTracker } from '@ai-operations/ops-policy';
import { DEFAULT_POLICY } from '@ai-operations/shared-types';

const manager = new AutonomyManager(DEFAULT_POLICY);
const decision = manager.canExecute('gmail', 'send', { source: 'email' });

if (decision.allowed && !decision.requiresApproval) {
  // Safe to execute autonomously
  manager.recordAction();
}

API

RuleEngine

Evaluates policy rules against connector operations. Rules are checked in priority order (highest first); the first matching rule wins.

constructor(config: PolicyConfig)
evaluate(connector: string, operation: string, context?: EvaluationContext): EvaluationResult

• EvaluationContext — { source?: string; intent?: string; amount?: number }
• EvaluationResult — { autonomy: AutonomyLevel; risk: RiskLevel; matchedRule?: PolicyRule; reason: string }

AutonomyManager

Wraps RuleEngine with rate limiting (hourly action counts, daily spend limits). Provides the primary canExecute interface for the orchestration layer.

constructor(config: PolicyConfig)
canExecute(connector: string, operation: string, context?: EvaluationContext): AutonomyDecision
recordAction(amount?: number): void
resetCounters(): void
getDailySpend(): number
getHourlyActionCount(): number

• AutonomyDecision — { allowed: boolean; requiresApproval: boolean; reason: string }

EscalationManager

Tracks denial counts per task and determines when and to whom issues should be escalated. Prevents tasks from being silently stuck in denial loops.

constructor(config?: EscalationConfig)
shouldEscalate(taskId: string, denialCount: number): boolean
getEscalationTarget(taskId: string): EscalationTarget | undefined
recordDenial(taskId: string): void
recordEscalation(taskId: string, target: EscalationTarget): void
clearState(taskId: string): void

• EscalationTarget — { role: string; channel: string; urgency: 'low' | 'normal' | 'high' | 'critical' }

BudgetTracker

Fine-grained spending management with global daily limits and per-connector breakdowns. Auto-resets at midnight.

constructor(dailyLimit?: number, connectorLimits?: Record<string, number>)
canSpend(amount: number, connector: string): { allowed: boolean; reason: string }
recordSpend(amount: number, connector: string): void
getDailyTotal(): number
getConnectorTotal(connector: string): number
getBreakdown(): Record<string, number>
getRemainingBudget(): number
setConnectorLimit(connector: string, limit: number): void

Related Packages

• @ai-operations/shared-types — PolicyConfig, PolicyRule, AutonomyLevel types
• @ai-operations/ops-core — Workflow engine that checks policy before execution
• @ai-operations/cord-adapter — CORD safety evaluation layer

License

MIT