Page 238 - CyberSecurityEssentialsEnglish

Common
Term Plain-language explanation
abbreviation

Organized material presenting facts, options, criteria, and
Decision support pack -
risks.
An approved output that demonstrates the required level
Gold example -
of quality.

The ability to run the same method and obtain a
Repeatability -
comparable result.

The ability to undo an action or restore the previous
Reversibility -
state.
Moving a case to a higher level or an expert when limits
Escalation -
are exceeded.

A fixed sequence in which a model performs one or more
AI workflow -
steps; the path does not change.
A system given a goal rather than a sequence, which
AI agent -
decides the next step, calls tools, and may re-plan.

An agent operating inside a real business process, with
Agentic workflow -
owners, approvals, and consequences.
Several agents working together under shared
Agentic system -
governance, coordination, and institutional control.
The mechanism by which a model emits a structured
Tool calling - request naming a tool and its arguments, which the
application then validates and executes.

The description of a tool given to a model: its name, its
Tool schema -
purpose, and the shape of the input it expects.
An agreed interface letting any compliant model connect
Standard access layer -
to any compliant tool without bespoke integration.

An open specification for how a tool announces what it
Model Context Protocol MCP offers and how a model requests it, so that any compliant
model can connect to any compliant tool.

An agent operating a screen directly rather than calling a
Computer or browser - defined interface; the highest-privilege route and the
use
poorest to audit.

Controls placed around a model rather than inside it:
Guardrails -
permissions, validation, allowlists, and approval gates.
The ability to see what a running system did: traces, logs,
Observability -
cost, latency, and recorded failures.

237