@ai-operations/cord-adapter

CORD safety gate, policy simulation, and forensic timeline analysis for AI Operations OS.

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

Install

npm install @ai-operations/cord-adapter

cord-engine is an optional dependency. When not installed, CordSafetyGate defaults to ALLOW for all evaluations so the system can operate in permissive mode.

Quick Start

import { CordSafetyGate, PolicySimulator } from '@ai-operations/cord-adapter';

const gate = new CordSafetyGate();
const result = gate.evaluateAction('gmail', 'send', { to: '[email protected]', subject: 'Hi' });

if (result.decision === 'BLOCK') {
  console.error('Action blocked:', result.reasons);
}

API

CordSafetyGate

Wraps cord-engine to evaluate proposed connector actions against safety policies before execution. Gracefully degrades to ALLOW when cord-engine is not installed.

constructor()
evaluateAction(connector: string, operation: string, input: Record<string, unknown>): SafetyResult
isAvailable(): boolean

• SafetyResult — { decision: CordDecision; score: number; reasons: string[]; hardBlock: boolean }
• CordDecision — 'ALLOW' | 'CONTAIN' | 'CHALLENGE' | 'BLOCK'

Operations are mapped to CORD tool categories: communication, publication, destructive, scheduling, financial, readonly.

PolicySimulator

Dry-run safety evaluation for projected actions. Runs a batch of actions through CordSafetyGate without executing them.

constructor(gate?: CordSafetyGate)
simulate(actions: ProjectedAction[]): SimulationReport

const sim = new PolicySimulator();
const report = sim.simulate([
  { connector: 'gmail', operation: 'send', input: { to: '[email protected]' } },
  { connector: 'shopify', operation: 'refund', input: { amount: 200 } },
]);
console.log(report.summary);     // { ALLOW: 1, CONTAIN: 1 }
console.log(report.allAllowed);   // false
console.log(report.hasHardBlock); // false

• SimulationReport — { entries, summary, maxScore, hasHardBlock, allAllowed }

ForensicEngine

Session-level forensic timeline analysis. Loads a task's full execution history from storage and builds a chronological, color-coded timeline for audit and debugging.

constructor(stores?: Stores)
async loadSession(sessionId: string): Promise<void>
buildTimeline(): ForensicTimeline
renderTimeline(): void   // Colored CLI output
getTimeline(): ForensicTimeline | null

• TimelineEvent — { timestamp, category, label, detail?, connector?, operation?, cordDecision?, cordScore? }
• Event categories: action | decision | approval | error | system

Related Packages

• @ai-operations/shared-types — CordDecision, ActionReceipt types
• @ai-operations/ops-core — WorkflowEngine uses CordSafetyGate as its SafetyGate
• @ai-operations/ops-storage — ForensicEngine reads session data from stores
• @ai-operations/ops-policy — Policy rules complementing CORD evaluation

License

MIT