Use this chapter if you are a NOC supervisor or shift lead holding Owner-level permissions. An operator's unit of work is one Advisory. As a supervisor, you unit of work is the whole set of Advisories: whether work is moving, whether your people are growing, and whether the queue itself is getting smaller.
This chapter is organized around those three goals, and it is a companion to Working an Advisory, the operator task.
Some steps in this chapter depend on preview features that are enabled per instance, so they might not appear in your environment:
- Claiming and closing Advisories depend on the Claim and Close preview feature. For more information, see Claim and Close.
- MCP Tools depend on the tools your administrator has enabled in the Toolbox, which requires you to enable the MCP Tools preview feature.
- The Slack or Teams integrations depend on the Sharing Information with Slack and Microsoft Teams Integrations preview features.
What Supervisors Watch
Everything your operators do in Skylar Advisor is written into their threads with a name and a timestamp. You almost never need to ask what happened; read the thread, then ask about the judgment behind it, not the facts. For more information, see Before You Work an Advisory.
Holding the Owner role means you can see things your operators cannot: the panel on their profiles, the Agentic Control Plane page, and tool and source configuration. Where this task sends you somewhere your operators cannot reach, the text states this clearly.
The Two Supervision Loops
Supervising with Skylar Advisor runs on two loops with different goals:
- The shift loop keeps today's queue moving. For more information, see The Shift Loop.
- The weekly loop reduces next month's volume: growing your people, improving your knowledge, and cutting noise at the source. For more information, see The Weekly Loop.
If you only ever run the shift loop, nothing structurally improves.
The Shift Loop
The following sub-topics cover the typical steps in a shift.
Reading the Queue
Go to the tab on the Dashboards page and set the scope to the whole organization instead of My Work. My Work is your own participation; for supervising, you want everyone's. For more information, see The Dashboards Page.
Set the time frame to the shift. The presets cover the usual spans, and a custom range covers the rest. Then use the state filters in the left rail: Open, Claimed, Closed, Reopened, and Archived. Those five states are your whole queue view. For more information, see Claim and Close.
Finding the Three Faults
Almost everything that needs your attention shows up as one of the following:
- Unclaimed and aging. Filter to Open. The oldest advisories appear at the bottom. An Advisory nobody has claimed is the one failure mode where nothing at all is happening. If this list is consistently long, the problem is capacity or visibility, not effort.
- Claimed but stalled. Filter to Claimed and look for threads with an old claim and no recent messages. The thread tells you instantly whether the operator is stuck, waiting on an approval, or has moved on and forgotten it.
- Reopened. Filter to Reopened. Every one is a closure that did not hold. Read the closing notes: was the fix wrong, or was the Advisory closed without a recovery snapshot? These are your highest-value coaching material, because the evidence is already in the thread.
Unblocking and Rebalancing
Approve anything waiting on you. For more information, see Approving Gated Tool Runs. Add yourself or a second operator to threads that are stuck; adding a colleague brings them the full history instead of a retyped summary. Where the impact is growing, escalate rather than letting an operator keep working it alone.
Use the other Dashboards page tabs to answer who a problem is actually affecting: the and tabs for business impact, the and tabs for where the problem is concentrated, and the tab for what keeps recurring. For more information, see The Services, Device Groups, Issues, and Organizations Tabs.
Handing Off
If your administrator has enabled MCP Tools, run the handoff tool from the Toolbox page at the end of the shift. It reports what is new, what is still open and who owns it, what was resolved, and what to watch, fleet-wide, over the window you choose. Read it before you send it: it summarizes what your operators wrote down, so a thin thread produces a thin handoff, which is itself worth knowing. For more information, see Handoff.
Ask the thread before you ask the person. Asking an operator "what is happening with this one" costs them their focus and gets you a summary. The thread gets you the facts in less time. Save the conversation for the judgment call: why they ruled something out, or why they escalated when they did.
The Weekly Loop
The following sub-topics cover the typical steps in a weekly loop.
Reviewing Your People
Click any operator's avatar to open their profile, and step it to the month you want. You see their Time to Claim and Time to Close averages, how much they used Ask Skylar, how many KB articles they generated, how much they shared, an activity-over-time chart, and a Standing panel that ranks them against the rest of the organization. For more information, see Checking Your Own Performance.
Click the button for two grounded sections: a factual Activity summary with no advice in it, and Areas for improvement, which lists two or three focus areas, each naming the metric behind it and a suggested next step. Both sections are drawn from that month's real numbers, and coaching is skipped when there is too little activity to say anything meaningful.
Your operators cannot see the button. It is an Owner-level control, so an operator opening their own profile sees the trends and their standing but not the advice. If you want an operator to act on the coaching, open it with them; a conversation grounded in the advice lands better than the numbers alone.
Reading the Numbers Before You Use Them
| What You See | What It Usually Means |
|---|---|
|
Time to Claim climbing |
A visibility or capacity problem before it is an effort problem. Ask where the operator is watching from before you ask why they are slow. |
|
Time to Close long, KB Generated zero |
The most actionable pattern on the page. The operator is solving hard problems and capturing none of it, so the team relearns it each time. |
|
Ask Skylar high, Turns low |
Someone using Skylar Advisor to learn rather than to close. Early in a ramp, this is exactly right. |
|
Shared near zero |
Working alone. Sometimes discipline, sometimes reluctance to ask; worth a conversation either way. |
|
A low standing in a strong team |
Often nothing. Standing is relative to your organization, not an absolute bar. |
Reviewing Your Knowledge
Two surfaces under the Knowledge page answer the same question from opposite ends: is what your team knows actually serving it?
- The view shows how each generated article is performing. An article that nothing ever cites is either wrong, unfindable, or about something that stopped happening. For more information, see KB Health: What Is Earning Its Place.
- The view shows what the Corpus is failing to serve. For more information, see Info Gaps: What Is Missing.
Protect the replay step when your team closes an Info Gaps topic. Replaying the original questions is what turns "we should document that" into a loop that actually finishes, and it is one of the most effective things you can do to make next month's shifts shorter.
Reducing Noise
Your operators work what arrives. You are positioned to reduce what arrives at all. If your administrator has enabled MCP Tools, run the following tools from the Toolbox page, where a tool runs on its own outside any conversation. The Toolbox page is also where you export a report; for more information, see Reporting Upward.
| Tool | Use It to Answer |
|---|---|
|
noise |
Which sources are noisiest and which alerts keep re-firing, with specific tuning suggested, such as deduplication, thresholds, or suppression. |
|
trends |
Where the team's effort actually went, with period-over-period movement. The honest answer to "what are we spending our time on." |
|
topic_pulse |
Whether a technology or symptom is rising or cooling across the fleet. |
|
correlate |
Whether a run of separate advisories was really one thing. For more information, see Choosing the Right Tool for the Question. |
|
spend_value |
Token spend for the period, optionally broken out per user, and how much of it produced value. For more information, see Spend Value. |
Approving Gated Tool Runs
You can configure some tools to require sign-off before they run. For more information, see MCP Tools. When an operator reaches one of these tools, a request appears in their thread carrying the tool, its parameters, who asked, and their written justification. You can approve the request, or deny it with a structured reason: wrong target, insufficient justification, timing, or other.
The following rules are enforced by the server, not by the interface:
- You cannot approve your own request. If you raised it, someone else signs it off.
- Approver membership is checked server-side. Being able to see the thread does not make you an approver; you must belong to that tool's approver group.
- Approval expires. A request that waits too long lapses instead of being decided against a situation that has already moved on.
- Approval is not permanent. Once granted, the run stays available to the operator for a bounded window; after that, it needs approval again.
There is no separate approval inbox. A pending request is visible inside the thread where it was raised, and also on the Dashboards page, under the Pending filter in the left navigation.
Reporting Upward
You can export anything you run from the Toolbox page as an HTML page, a PowerPoint deck, a PDF, a Word document, or Markdown. Headings become slides in the deck, and tables stay intact. As a result, your weekly report is usually a tool run plus a review pass on the report Skylar Advisor generated for you, not a document you assemble by hand.
The following table lists what typically travels well upward:
| The Question You Are Asked | What to Bring |
|---|---|
|
"How did the week go?" |
The handoff tool over a weekly window, or the trends tool for effort and recurrence with period-over-period deltas. |
|
"Are we getting on top of it?" |
The noise tool, and the tuning you have already actioned since last time. Movement matters more than the absolute number. |
|
"What did it cost?" |
The spend_value tool for the period, broken out per user if the question is about distribution. |
|
"What was the business impact?" |
The and tabs on the Dashboards page, which speak in services rather than devices. |
|
"Why did this one take so long?" |
The thread. It carries the timeline, what was ruled out and by whom, and the metrics on either side of the fix. |
Claim and close timings are only as honest as your team's habits. If operators close early and reopen, or close without a recovery snapshot, the timings improve while the service does not. Fix the habit before you report the number, and if you know a number is flattered, say so when you present it.
Supervisor Failure Modes
| Pattern | Why It Costs You |
|---|---|
|
Running only the shift loop |
Work keeps moving and nothing gets smaller. The weekly loop is the only one that reduces next month's volume. |
|
Asking operators for status the thread already holds |
You break their focus to get a worse version of what you could have read yourself. |
|
Coaching from the standing panel alone |
Standing is relative to your team. Coach from the thread and the two timing metrics, not from a leaderboard position. |
|
Treating a reopened Advisory as a failure |
Reopening is the honest act. Punish it, and operators stop doing it, which costs you your best quality signal. |
|
Letting KB articles accumulate unreviewed |
A wrong article is worse than none, because the next operator trusts it. The view tells you which articles are actually being relied on. |
|
Waiting to be notified about approvals |
Monitor the Pending filter on the Dashboards page yourself. A request that expires quietly leaves the operator thinking the approval gate is broken. |
Keep the following four points in mind:
- Read the queue by state, not by scrolling.
- Chase unclaimed, then stalled, then reopened advisories, in that order.
- Protect the weekly loop when the shift gets loud.
- Watch for approval requests to keep the team moving.