Configuration

Service Configuration

Each monitored Worker is registered as a service config in D1. You manage them through the Orchestrator API.

Register a Service

curl -X PUT https://sentinel.example.com/api/config/services/my-worker \
  -H "Content-Type: application/json" \
  -d '{
    "serviceName": "my-worker",
    "repoUrl": "https://github.com/org/my-worker",
    "repoBranch": "main",
    "pollIntervalMs": 30000,
    "enabled": true,
    "errorClassFilter": ["server_5xx", "unhandled_exception"],
    "minOccurrences": 1
  }'

Fields

Field	Type	Default	Description
`serviceName`	string	required	Worker name (matches Workers Observability)
`repoUrl`	string	required	GitHub repository HTTPS URL
`repoBranch`	string	`"main"`	Default branch for cloning
`pollIntervalMs`	number	`30000`	Polling interval in milliseconds
`enabled`	boolean	`true`	Whether to monitor this service
`errorClassFilter`	string[]	`["server_5xx", "unhandled_exception"]`	Error classes to detect
`minOccurrences`	number	`1`	Minimum occurrences before creating an incident

Error Classes

Class	Description
`client_4xx`	HTTP 400–499 responses
`server_5xx`	HTTP 500–599 responses
`unhandled_exception`	Uncaught exceptions and promise rejections

Secrets

Set secrets using Wrangler. These are never stored in code:

bunx wrangler@latest secret put GITHUB_TOKEN
bunx wrangler@latest secret put ANTHROPIC_API_KEY

For local development, add them to .dev.vars:

GITHUB_TOKEN=ghp_...
ANTHROPIC_API_KEY=sk-ant-...
CLOUDFLARE_API_TOKEN=...
CLOUDFLARE_ACCOUNT_ID=...

Secret	Purpose
`GITHUB_TOKEN`	Create branches, PRs, and issues on monitored repos
`ANTHROPIC_API_KEY`	AI Gateway fallback for LLM calls
`CLOUDFLARE_API_TOKEN`	Query Workers Observability API
`CLOUDFLARE_ACCOUNT_ID`	Identify the Cloudflare account

Wrangler Configuration

The main wrangler.jsonc file defines all Cloudflare bindings. Key settings:

Queue Tuning

Queue	`batch_size`	`max_batch_timeout`	`max_retries`	Notes
`errors-detected`	10	5s	3	High throughput for error detection
`triage-ready`	5	10s	3	Medium throughput for triage
`fix-ready`	1	30s	2	Serial — one fix at a time per queue
`completed`	10	5s	1	Logging only, low retry

The fix-ready queue uses batch_size: 1 to prevent concurrent fix attempts on the same repository.

Observability

All observability settings are enabled by default:

"observability": {
  "enabled": true,
  "head_sampling_rate": 1,
  "logs": { "enabled": true, "persist": true, "invocation_logs": true },
  "traces": { "enabled": true, "persist": true }
}

This ensures full visibility into agent behavior in the Cloudflare dashboard.