Command Rules
Command rules define what the agent may read or fix on machines: the catalogue in force, consent and secret files. They are set for the whole organization, with exceptions per client.
Each save creates a version signed by you and sent to the machines. A client with no exception applies the organization's rules.
How a Command Is Decided
Command proposed by the agent
|
v
Always refused? (list below) ---> Refused
|
v
Risk score (catalogue)
+15 on a server
|
+-- below the automatic threshold ---> Runs on its own
| (active autonomy policy)
|
+-- between the two thresholds ------> Approval
|
+-- from the block threshold --------> Refused
- Without an active autonomy policy for the client, every command that is not refused waits for an approval.
- A command unknown to the catalogue has no category or score: it always needs approval.
- A destructive command never runs alone, whatever the score, and the machine name is retyped to run it.
Always Refused
- Deleting an account or granting administrator rights
- Stopping or reconfiguring a protected service (SmartTech, antivirus, firewall)
- Shutting the machine down: access is lost and nobody on site can power it back on
usingand#requiresstatements that load a module or an assembly before execution- Reading a user's profile on another machine (consent impossible)
Tabs
| Tab | Content |
|---|---|
| Commands | Catalogue in force, risk score per command on a workstation and on a server. A disabled command becomes unknown (approval) |
| Consent | Personal-data reads subject to the machine user's agreement |
| Secret files | Files whose reading always requires an approval, wherever they are |
| Thresholds | Automatic and block thresholds, by default or for one client |
| Simulator | Classifies a command with the rules shown and the machine role, without running anything |
| Classified | What the model decided about commands the catalogue does not know |
| History | Signed versions, with a way back to a previous version or to the catalogue values |
Command Templates
Ready-made lists by family. Each template adds its commands to the scope shown, exactly as if you typed them one by one. Nothing applies until you add one.
Consent
By default, the machine's user accepts each read of their data. A consent lift covers every read of the category: personal data of that category is then read without asking the user.
Law 25: a legal review is still advised before lifting consent for a client in Quebec.
Simulator
The simulator gives the effective decision for a command on the chosen client: the decision shown in the catalogue is indicative, the simulator is authoritative. With no client selected, it simulates a client in autonomy at the organization's default thresholds.
Classified Commands
For a command the catalogue does not know, the model states what it does, how confident it is and its score. Nothing applies on its own: an Admin decides, and can add the command to the catalogue so the agent classifies it the same way next time.
Unverified Rules
If the signature, the scope or a line of the rules fails verification, those rules are not in force and every command is refused until an Admin resets them to the catalogue values.
Protected Active Directory Accounts
In Settings, the protected accounts list stops SmartTech from resetting their password, even if a technician approves.
- Preferably enter the sign-in name (sAMAccountName, e.g.
jtremblay): the host then refuses on its own, on top of the server. A UPN, a distinguished name orDOMAIN\nameare also accepted. - Accounts marked privileged in the directory (adminCount) are still refused, even with no list.
- The list does not limit unlocking, disabling or groups, and does not see a command typed by hand on a domain controller (it stays subject to the score and to approval).