One-pass health read of a single device, ending in a concrete remediation proposal the tech can act on through a deep link. Read-only skill. This needs the RMM connected.
1. Resolve the device without stopping to ask mid-lookup: identify the client organization from the ticket/prompt, then look up its devices in the RMM. If multiple candidates match, rank by organization match first, then most recent last-contact, and state which one you picked and why. Only ask if candidates are genuinely indistinguishable.
2. Pull in parallel: the device details (status, OS, last contact, disk, pending reboot), its active alerts, its recent activity (last 24-72h), and its Windows services to spot stopped services set to Automatic.
3. Infer the device's role from its hostname prefix where the convention is evident (DC/AD = domain controller, SQL/DB = database, FS = file server, RDS/TS = terminal server, HV/ESX = hypervisor). Say it is an inference, not a fact.
4. Check the recent activity for another tech already working the device (remote sessions, manual actions). If someone is active, flag it before proposing anything.
5. Co-hosted-VM contention check: if the device shares a public IP with other devices in the same organization, list those siblings and check whether any carries resource alerts.
6. Assess posture: disk free space (flag under ~15% or under 10 GB on system volume), pending reboot age, patch status and failed-patch activities, repeated/flapping alerts.
7. Cross-reference documentation (IT Glue) when available to confirm the asset and pick up known quirks.
8. Output a health summary: status line, risks ranked by severity, inferred role, and a remediation proposal per risk. For any hands-on remediation, hand the tech a deep link into the device in the RMM so they jump straight to it. Offer to leave the summary as a plain-text note (no markdown/emojis).
Guardrails: don't trust a class filter — verify the returned device's class in its details before reasoning about it as a server or workstation. If any listing may have capped, report counts as "at least N", never exact. This skill is read-only; actions (reboot, service restart) belong to the dedicated action skills and require requester confirmation before affecting a logged-in user. Role inference from hostname is a hint, never confirmed without documentation backing. If the RMM is not enabled, say so and fall back to ticket history and documentation only; do not fabricate device data.
Magic Library