Slack Operations Notifications

Synced with starter commit 2a1a04a.

Use Slack when a human should notice an operational exception quickly—such as a parked Stripe event—not as the permanent record of what happened. The integration is optional: leaving SLACK_WEBHOOK_URL empty disables delivery.

Choose which alerts deserve durable delivery

• Stripe webhook and sweep alerts already use slack_event or slack_error jobs and retry through the durable runner.
• A few storage and subscription guardrail alerts still use the best-effort notifySlack* helpers. Those can be lost if a serverless instance freezes. If an alert is required for your operation, enqueue a Slack job instead.
• Keep messages actionable: include a safe request, order, event, or job identifier and tell the operator where to investigate. Never include tokens, raw payloads, signed URLs, or unnecessary personal data.

Ready to launch means: a harmless test event reaches the intended channel, a simulated 429/5xx is visible as a retry or failure, duplicate enqueue is suppressed, and the identifier in Slack leads back to the database or structured logs.

Create a Slack Incoming Webhook and set SLACK_WEBHOOK_URL. SDK construction stays in src/integrations/slack.ts. Billing and sweep paths queue typed slack_event or slack_error jobs; older guardrail call sites use best-effort helpers, so use a job for every new must-deliver alert.

Use deterministic dedupe keys for noisy events, include request/order/job identifiers for correlation, and omit credentials, tokens, raw request bodies, download URLs, and unnecessary personal data. A missing webhook should be an intentional environment choice, not a hidden production failure.

Delivery retries through the same durable runner as email and cleanup work. Keep /api/cron/jobs authenticated and scheduled, monitor failed jobs, and test duplicate enqueue, Slack timeout/429/5xx, redaction, and recovery. Slack is an alerting channel; the database and structured logs remain the system of record.

Next: Include Slack delivery and queue-failure checks in the Deployment and Security launch runbook.