Get Started as a Security Analyst
Role: Security analyst. The console shows the role you work in beside your name, at the top right.
Who This Is For
You work in security operations. You triage the alerts guardrails raise, stop an agent that misbehaves, investigate what it did, and show that the defenses hold. You can read every agent, connector, approval, policy and request, and you change the state of agents and alerts, not the gateway's configuration.
What You Can Do
| You can | Where it is covered |
|---|---|
| Triage alerts on Guardrails → Alerts with Ack, Resolve, False positive and Contain | The Triage Queue, Containing an Agent |
| Turn any agent off and on | Who Can Turn an Agent Off and On |
| Run the attack pack with Run sandbox pack, and read each run | The Proving Ground |
| Read guardrail posture, history and Session investigator results, and export guardrail evidence | Posture and History, Evidence Export |
| Read Request log, Desktop activity, Session watch and Policy decisions for every agent | Request Log, Session Watch |
| Read every connector, agent, approval and policy, the judges, and the data classification | Permissions by Role |
Someone else does this: an administrator changes policies, connectors, settings and data classification, and saves guardrail settings. You can read Guardrails → Settings and Guardrails → Judges.
A session the Session investigator paused stays paused for its time limit, then resumes or ends as your guardrail settings decide. While it waits, you contain the agent or mark the pause a false positive. When your organization escalates pauses to the agent's supervisor instead of the SOC list, that supervisor can release the session early. See Escalation and Signed Resume.
How It Works for You
Guardrails watch what agents do with what they read. When an action looks unsafe, a guardrail holds it for a person or stops it, and raises an alert. The Session investigator can also pause a whole session for review.
Alerts wait for you on Guardrails → Alerts. You acknowledge an alert, resolve it, mark it a false positive, or contain the agent. Contain turns the agent off and revokes its sessions, so the gateway refuses its next request until someone turns it back on. Every action you take is recorded with your name.
Guardrails carries a Preview tag in the console.
If you hold more than one role, choose Security analyst under Role in the account menu. See Work in More Than One Role.
Your First Hour
- Open Guardrails → Posture for the state of your defenses. See Posture and History.
- Work through the open alerts. See The Triage Queue.
- Contain an agent from an alert, and note who can turn it back on. See Containing an Agent.
- Search Activity → Request log for the agent's requests. See Search the Stream.
- Open one session, and read it end to end. See Open a Session.
- Read what Session watch found in it. See Reviewing a Session.
- Read the Session investigator's would-pause decisions. See Reading Would-Pause Decisions.
- Trace why a request was blocked. See When a User Reports a Block.
- Run the attack pack, and compare the result with the baseline. See Running the Attack Pack.
Where to Go Next
| If you want to | Go to |
|---|---|
| Work an incident from alert to evidence, with an Admin | Respond to an Agent Incident |
| Understand how the breaker decides | The Rule-of-Two Breaker |
| Know what happens when a tool's definition changes | Tool-Definition Integrity |
| See how a session is paused, resumed and stopped | Pausing and Stopping a Session |
| See which AI tools run on your workstations | Desktop Activity |
| Check each control AgentPass enforces | Security Controls |
Related
- Input Guardrails — the pieces that raise your alerts
- Permissions by Role — every permission each role holds, by product area
- Get Started as an Auditor — the read-only role for evidence and review