ops-triage

Operational triage

Read the evidence contract. Confirm the selected mailbox/account, project, time range and requested outcome. Read-only triage can proceed; an instruction to fix a named issue also permits routine fixes within its scope, subject to actual permission boundaries.

Collect the selected notices with a supported connector and join them to current dashboard, incident or billing records. Use history only to locate a relevant source. Group repeated alerts by underlying project and incident; retain first and latest event time and affected environment. Do not archive or mark read as part of analysis unless asked. Keep authentication/security notices separate from routine cost alerts and verify their destination through the provider's known source rather than links embedded in mail.

Prioritise by present impact and time to irreversible consequence: an active outage, compromised access or imminent service suspension can outrank a larger historical cost. A noisy count alone does not establish severity. Distinguish current error, recovered incident, suspected cause and inaccessible source. Route a bounded engineering fix to the applicable shipyard:triage or existing owner; route session continuation to flagship:fleet-console, with a verified task/session identity. Do not launch a fleet from a triage request.

For cloud credits and cost, record account/project, provider, currency, balance checked time, expiry and timezone, eligible products/regions, credit restrictions and the authoritative locator. Distinguish usage estimate, posted charges, invoice, spend cap, payment method and credit balance. Do not treat credits as cash or assume they offset every charge. Forecast only with an evidenced usage rate and explicit period; state uncertainty and exclude unsupported precision. If redemption, purchasing credits, stopping resources, changing quotas or billing settings is needed, prepare its exact effect and seek missing authority.

Deliver a short ranked table: incident/notice identity, current state and source, impact/deadline, next action and owner. List sources not reached, conflicts and items that need refreshed evidence. Stop after the selected set is reconciled and any explicitly authorised remediation is verified. If future checks are requested, discover supported scheduling tools, read existing schedules and confirm required connector access before creating one.

For resumable reviews, use the ledger format with domain ops. The helper's summary groups unresolved work without sending notifications or modifying the inbox.