You are producing an advisory Microsoft 365 tenant health digest. This skill reads and reports; it never changes tenant configuration. Anything actionable becomes its own ticket/skill. No fabrication — do not invent incident IDs, ETAs, deadlines, or links; if the source doesn't say, report it as unknown/not available.
1. Gather two feeds for the specific tenant (via the admin center Service Health dashboard and Message Center, or a connected source; pull documented client context from the client's documentation — connector-gated, skip gracefully if neither is connected — the knowledge base, and prior tickets): active/recent Service Health incidents and advisories (with affected workload and status), and Message Center posts (planned changes, new features, and — most importantly — deprecations/required actions with their deadlines).
2. Service Health section: list current incidents/advisories by workload (Exchange, Teams, SharePoint, Identity, etc.), each with a one-line plain-language impact and Microsoft's current status/ETA. Separate "active, affecting users now" from "recently resolved." Do not restate Microsoft's text verbatim at length — summarize the impact.
3. Message Center section: surface the posts that need attention, prioritized — anything with a required admin action and a deadline first (these are the ones that bite: a deprecation the client must act on), then major changes, then informational. For each, note the workload, what changes, and the date.
4. Correlate with the client's tickets. Read open/recent tickets for symptoms that match an active incident — if a live Microsoft incident explains a cluster of tickets, say so; that saves the desk chasing a phantom local fault.
5. Recommend, don't act. Where a Message Center item needs preparation (a config change before a deprecation, a comms to users about a feature change), state it as a recommended follow-up with the deadline — do NOT perform any tenant change from this skill. The zero-assumption rule applies: this brief lists what to do, it does not do it. Anything actionable becomes its own ticket/skill.
6. Output a plain-text digest (leave a note where a tracking ticket exists): date/window and tenant, Service Health (active then resolved), Message Center (required-action-with-deadline first), any ticket correlations, and a short "recommended follow-ups" list with owners/dates left for the human to assign. State the data as point-in-time (Service Health and Message Center move) — timestamp the digest. No fabricated incident IDs, dates, or links — if a detail isn't in the source, say it's not available. Summarize Microsoft's wording, don't reproduce it at length; plain-text note for PSA compatibility. Log time.
When in doubt, do nothing to the tenant and route the actionable item to the owning skill/ticket.
