Types

Complete TypeScript type reference for @sentinel-sdk/core

4 min read

Types

All types are exported from @sentinel-sdk/core. All fields are readonly.

Union Types

ThreatType

typescript
type ThreatType =
  | 'ROLE_OVERRIDE'
  | 'DRAIN_INTENT'
  | 'URGENCY_MANIPULATION'
  | 'JAILBREAK'
  | 'CONTEXT_MANIPULATION'
  | 'OUT_OF_SCOPE'

The category of threat detected by the PromptGuard.

RiskLevel

typescript
type RiskLevel = 'low' | 'medium' | 'high' | 'critical'

Risk severity computed by the RiskScorer. Maps to score ranges: low (0–29), medium (30–59), high (60–84), critical (85–100).

ScanMode

typescript
type ScanMode = 'llm' | 'rules' | 'both'

The mode used for a PromptGuard scan. The mode_used field in ScanResult reflects which mode actually ran.

SentinelMode

typescript
type SentinelMode = 'full' | 'guard-only' | 'sandbox-only'

Controls which pipeline components are active.

LLMProvider

typescript
type LLMProvider = 'anthropic' | 'openai' | 'custom'

The LLM backend for the judge. Use 'custom' with a baseUrl for OpenAI-compatible APIs.

SentinelEvent

typescript
type SentinelEvent =
  | 'threat:detected'
  | 'tx:simulated'
  | 'policy:violated'

Event names for the sentinel.on() / sentinel.off() API.

Result Types

ScanResult

Returned by PromptGuard.scanInput() and Sentinel.scanInput().

typescript
interface ScanResult {
  readonly safe: boolean
  readonly threatType?: ThreatType
  readonly confidence?: number
  readonly flags: readonly RiskFlag[]
  readonly mode_used: ScanMode
  readonly latency_ms: number
  readonly reasoning?: string
}
FieldDescription
safetrue if no threat was detected
threatTypeCategory of detected threat (present when safe: false)
confidence0–1 detection confidence (present for LLM results)
flagsIndividual risk signals
mode_usedScan mode that produced this result
latency_msTime taken to produce the result
reasoningLLM explanation (LLM mode only)

SimulationResult

Returned by ExecutionSandbox.evaluate() and Sentinel.evaluateTransaction().

typescript
interface SimulationResult {
  readonly approved: boolean
  readonly riskScore: number
  readonly riskLevel: RiskLevel
  readonly riskFlags: readonly RiskFlag[]
  readonly policyViolations: readonly PolicyViolation[]
  readonly balanceChanges: readonly TokenChange[]
  readonly programInvocations: readonly string[]
  readonly latency_ms: number
}
FieldDescription
approvedfalse if risk exceeds threshold or policy is violated
riskScore0–100 composite risk score
riskLevelCategorical risk level
riskFlagsFactors contributing to the risk score
policyViolationsPolicy rules that were violated
balanceChangesDetected token/SOL balance changes
programInvocationsProgram IDs the transaction invokes
latency_msTotal simulation time

ExecutionResult

Returned by Sentinel.execute().

typescript
interface ExecutionResult {
  readonly approved: boolean
  readonly guardResult?: ScanResult
  readonly sandboxResult?: SimulationResult
  readonly blocked_by?: 'prompt_guard' | 'execution_sandbox'
  readonly latency_ms: number
}

Supporting Types

AgentAction

Input to Sentinel.execute().

typescript
interface AgentAction {
  readonly input?: string
  readonly transaction?: string
  readonly metadata?: Readonly<Record<string, unknown>>
}

RiskFlag

A single risk signal contributing to a score or scan result.

typescript
interface RiskFlag {
  readonly factor: string
  readonly weight: number
  readonly score: number
  readonly description: string
}

PolicyViolation

A policy rule that was violated during sandbox evaluation.

typescript
interface PolicyViolation {
  readonly rule: string
  readonly message: string
  readonly details?: Readonly<Record<string, unknown>>
}

TokenChange

A token balance change detected during simulation.

typescript
interface TokenChange {
  readonly mint: string
  readonly symbol?: string
  readonly amount: number
  readonly decimals: number
}

RuleDefinition

A single rule entry in a YAML rule pack.

typescript
interface RuleDefinition {
  readonly id: string
  readonly description: string
  readonly pattern: string
  readonly flags?: string
  readonly action: 'BLOCK' | 'FLAG'
  readonly severity: RiskLevel
  readonly threat_type: ThreatType
}

RulePack

A collection of rules loaded by the RuleEngine.

typescript
interface RulePack {
  readonly name: string
  readonly version: string
  readonly description: string
  readonly rules: readonly RuleDefinition[]
}

SentinelEventData

Maps event names to their payload types.

typescript
interface SentinelEventData {
  readonly 'threat:detected': { readonly result: ScanResult }
  readonly 'tx:simulated': { readonly result: SimulationResult }
  readonly 'policy:violated': { readonly violation: PolicyViolation }
}
PreviousConfigurationNextError Codes