Skip to main content
Unlisted page
This page is unlisted. Search engines will not index it, and only users having a direct link can access it.
Version: 11.3.0

Suggested Policies

The Suggested tab of Policies proposes rules worth considering. Every suggestion states the evidence behind it, and links to the requests that motivated it. You can check the reasoning instead of taking it on trust.

Traffic is reviewed once a day. Until the first pass has run, the tab reads Nothing has been proposed yet.

Suggestions are a starting point, never a shortcut. Nothing in this panel creates, edits or enables a policy.

note

Reviewing suggestions is an administrator task.

Where Suggestions Come From​

One queue holds suggestions from the three sources in the following table. A card from an external client names the client. Every other card carries a severity tag.

SourceWhat it is
Your trafficHYPR AgentPass™ detectors run on a schedule over observed calls and coverage, and raise a gap they can count
An external clientA registered security or compliance tool submitted a proposal through the policy MCP server
A reference documentAn obligation in a standard you uploaded, raised by the Suggest policies action on that document

Traffic-derived suggestions come from deterministic detectors, not from asking a model what it thinks. A detector decides that something is worth proposing, such as:

  • An ungoverned tool carrying real volume
  • A destructive operation nobody governs
  • Access to sensitive paths
  • A tool that was quiet last window and busy this one

The evidence on the card is counted, never generated. Only the wording of the suggested intent is phrased for readability.

The panel also explains its own ordering. Ranking is built from the quantities on the card: how much traffic, how risky the operation, and how recently it appeared. The reason a suggestion is at the top is something you can read.

The Suggested tab, listing proposed rules with their source, evidence counts and Review and draft buttons

Read a Suggestion Card​

A banner above the cards reads These describe what agents asked to do, not what happened. The gateway records requests, and some of them were refused by a rule or by the agent itself. Each card holds the following parts.

PartWhat it shows
TagHIGH, MEDIUM or LOW, from the same score that ranks the list. On a card from an external client, the tag reads PROPOSED BY and the client's name, or PROPOSED BY A CLIENT SINCE REMOVED
TitleThe tool, or the connector, and the connector it runs on
IntentThe rule the suggestion proposes, in plain language
EvidenceThe counted facts behind it, such as the number of calls, how many agents made them and how many calls came in the period before
Why this?A Why this is here list of the reasons the suggestion was raised
See the N callsOpens the Request Log, filtered to the connector and tool

Answering a suggestion needs the Add permission on Policies. Any role that can open the page can read the cards.

Reading a Suggestion From an External Client​

A suggestion submitted through the policy MCP server is an outside party's claim, and the panel treats it as one. The card shows two things side by side:

  • The claim: the submitting client, its justification and any external reference it supplied. The justification is labeled Client's claim — untrusted:
  • The corroboration: what the gateway can independently confirm from your own traffic

The claim is a scanner's opinion. The corroboration is your data. When AgentPass finds no corroborating traffic, it says so, because "we see no such traffic" is important information about a proposal.

Corroboration is matched at the connector and tool level. A claim about a specific external host is confirmed only as far as the connector that would reach it, which is weaker than the claim. A corroborated claim about a host still deserves a look at the requests behind it.

A client's submission is not scored or ranked against your own traffic-derived suggestions. It may therefore sit lower in the list than its importance warrants.

Accept a Suggestion​

Click Review and draft. The policy authoring wizard opens with the connector, tool and intent filled in. For a document-derived suggestion, the passage that motivated it is attached too. No policy is created yet.

Everything that guards a hand-written policy still applies. You see the drafted CEL, the impact preview against your own traffic, and the duplicate and conflict check. The rule is saved in Observe mode, for you to promote when you are ready.

Accepting marks the suggestion as answered, so it does not reappear while you work on it. If you close the wizard without saving, the suggestion returns to the queue.

Dismiss a Suggestion​

Dismissing asks for a reason, and the reason decides what happens next. Each reason suppresses a class of suggestion, not a single card, so the same proposal does not simply reappear on the next pass.

Click Dismiss on the card, and choose a reason. Each reason in the menu carries a one-line hint.

ReasonHint in the menuEffect
Not a risk for usStops this being suggested for this tool.This signal is suppressed for this tool, and does not return
Already handled elsewhereStops this being suggested for this tool. We govern it another way.Suppressed for this tool. Use it when you govern the activity another way
Not nowComes back only if this gets materially busier.Suppressed temporarily. Returns only if the activity behind it grows materially
Bad suggestionStops this kind of suggestion everywhere, not just here.Suppresses this shape of suggestion across every tool. It is feedback about the proposal itself

Only Not now comes back. It snoozes the suggestion for 14 days, and records the level of activity when you dismissed it. When the snooze ends, the suggestion returns only if that activity has at least tripled, because by then it is a different question.

The other three reasons are lasting.

No notifications for suggestions

Suggestions raise no notifications, badges or emails. The panel is there when you are already thinking about policy. If activity needs urgent attention, raise it through your existing alerting, not this list.

What the Panel Does Not Do​

The panel has deliberate limits:

  • It never creates, enables or promotes a policy. Every suggestion passes through the wizard and through you.
  • It does not propose rules for activity that is already governed. A suggestion for something you have covered is suppressed before you see it.
  • It shows a short list, not a feed. At most ten suggestions from your traffic are open at a time. Each daily pass replaces them, so when you answer one, the next candidate takes its place on the following pass. Suggestions from external clients and reference documents are listed in addition.