Queue Messages

Sentinel uses five Cloudflare Queues plus a dead-letter queue for reliable inter-agent communication.

Queue Topology

LogTailer → [errors-detected] → TestGen → [triage-ready] → CodeTriage
    → [fix-ready] → FixAgent → [completed]

All failures → [sentinel-dlq]

Queue Configuration

Queue	Binding	Batch	Timeout	Retries	Retry Delay	DLQ
sentinel-errors-detected	ERRORS_DETECTED_QUEUE	10	5s	3	60s	sentinel-dlq
sentinel-triage-ready	TRIAGE_READY_QUEUE	5	10s	3	120s	sentinel-dlq
sentinel-fix-ready	FIX_READY_QUEUE	1	30s	2	300s	sentinel-dlq
sentinel-trace-external	TRACE_EXTERNAL_QUEUE	10	5s	3	60s	sentinel-dlq
sentinel-completed	COMPLETED_QUEUE	10	5s	1	—	—
sentinel-dlq	—	10	10s	—	—	—

Why batch_size: 1 for fix-ready?

The FixAgent clones repositories, creates branches, and pushes commits. Running multiple fix attempts concurrently on the same repo would cause Git conflicts. Serial processing (batch_size: 1) ensures each fix runs in isolation.

Message Schemas

All messages are validated with Zod at both the producer (before queue.send()) and consumer (after receiving).

ErrorDetectedMessage

Queue: sentinel-errors-detected Producer: LogTailer Consumer: TestGen

{
  fingerprint: string,        // SHA-256 hex (64 chars)
  incidentId: string,         // ULID
  serviceName: string,        // Worker name
  triggerRoute: string,       // e.g., "/api/users/:id"
  errorClass: ErrorClass,     // "client_4xx" | "server_5xx" | "unhandled_exception"
  sampleEvent: ObservabilityEvent,
  occurrenceCount: number,    // >= 1
  firstSeen: string,          // ISO 8601
  lastSeen: string            // ISO 8601
}

TriageReadyMessage

Queue: sentinel-triage-ready Producer: TestGen Consumer: CodeTriage

{
  incidentId: string,
  fingerprint: string,
  serviceName: string,
  repoUrl: string,            // GitHub HTTPS URL
  testCaseR2Key: string,      // "incidents/{id}/test-case.ts"
  testResultR2Key: string,    // "incidents/{id}/test-result.json"
  reproductionConfirmed: boolean,
  sampleEvent: ObservabilityEvent
}

FixReadyMessage

Queue: sentinel-fix-ready Producer: CodeTriage Consumer: FixAgent

{
  incidentId: string,
  fingerprint: string,
  serviceName: string,
  repoUrl: string,
  commitSha: string,          // HEAD at time of triage
  testCaseR2Key: string,
  triageResult: {
    rootCauseFile: string,
    rootCauseLines: [number, number],
    rootCauseExplanation: string,
    fixStrategy: string,
    confidence: "high" | "medium" | "low",
    affectedFiles: string[]
  },
  githubIssueNumber: number
}

CompletedMessage

Queue: sentinel-completed Producer: FixAgent Consumer: Completion handler (log and ack)

{
  incidentId: string,
  fingerprint: string,
  serviceName: string,
  status: "resolved" | "wontfix" | "needs_human",
  githubPrNumber?: number,
  githubPrUrl?: string
}

Dead Letter Queue

Failed messages from any queue (after exhausting retries) are sent to sentinel-dlq. The DLQ handler logs the full message body, timestamp, and attempt count to console.error, then acknowledges.

In a future phase, DLQ messages may trigger alerts or be written to D1 for investigation.