AI agent governance cannot live only in dashboards.

Platform engineers, security teams, and SREs spend a lot of their day in terminals. When an agent is misbehaving, a policy blocks a production workflow, or a high-risk action is waiting for approval, teams need a fast operational surface they can trust.

That is why AGP includes the agp CLI.

One Command Surface

AGP has several control-plane responsibilities: identity, registry, policy, approvals, and runtime invocation.

The CLI brings those workflows into one command model:

agp <service> <action> [flags]

That gives operators a predictable way to inspect and act across the system without switching between ad hoc scripts, raw API calls, and service-specific tools.

Managing Agent Identity

Agent identity is the first layer of runtime governance. Teams need to know which agents exist, who owns them, what scopes they have, and whether they should still be active.

With the CLI, common identity tasks become direct:

agp identity list-agents
agp identity register-agent --name payments-agent --scopes tools.invoke,finance.write
agp identity rotate-secret --agent-id <agent-id> --reason "credential rotation"
agp identity suspend-agent --agent-id <agent-id>

That last command matters during incidents. If an agent is compromised or behaving unexpectedly, the fastest safe action is often to suspend it before investigating deeper.

Understanding Tool Access

Agents become risky when they can reach tools they should not touch.

The AGP registry tracks which tools exist, which versions are active, and which agents depend on them. The CLI gives teams a fast way to answer operational questions:

agp registry list-active-tools
agp registry list-dependents --tool-name process-payment
agp registry replace-agent-dependencies --agent-id <agent-id> --tool-names crm-read,invoice-parse

Before changing a tool, teams can see which agents rely on it. Before granting an agent tool access, teams can make that dependency explicit.

Debugging Policy Decisions

When a tool call is blocked, the team needs to understand why.

The CLI can evaluate policy inputs and inspect tool risk classifications:

agp policy evaluate --file policy-input.json
agp policy list-tool-classifications
agp policy get-tool-classification --tool-name payment-write

This is useful for security review and for day-to-day developer support. Instead of guessing why a request was blocked, teams can inspect the decision path.

Reviewing High-Risk Actions

Some actions should not execute automatically. A deployment, payment write, data export, or destructive database operation may require human review.

The CLI supports approval workflows from the terminal:

agp approval list --status PENDING
agp approval get --approval-id <approval-id>
agp approval approve --approval-id <approval-id> --reviewer alice --reason "approved for change ticket"
agp approval deny --approval-id <approval-id> --reviewer alice --reason "scope is broader than requested"

The important part is not just approval. It is context and auditability. The reviewer, reason, agent, tool, and decision become part of the operational record.

Testing Agent Tool Invocation

Before an agent is trusted in production, teams need to know whether it can discover and invoke the right tools through AGP.

The CLI can exercise the runtime path:

agp proxy discover-tools --agent-id <agent-id>
agp proxy invoke-tool --agent-id <agent-id> --tool-name invoice-parse --payload-file invoke.json

This helps validate identity, scopes, registry dependencies, policy, and runtime behavior together.

Why This Matters

Governance needs to be operational.

A good dashboard helps teams understand the system. A good CLI helps them operate it under pressure.

AGP’s CLI is designed for the moments when teams need to inspect an agent, contain a risk, review a pending action, debug a policy decision, or verify that the runtime path is behaving correctly.

That is the difference between governance as a report and governance as a control plane.