COMPEL Glossary / action-space
Action Space
The action space is the complete set of all actions an AI agent can potentially take, including tool invocations (API calls, database queries, file operations), communication actions (messages to humans or other agents), reasoning actions (internal processing steps), and environmental interactions (network requests, system operations).
What this means in practice
Defining and constraining the action space is fundamental to agent safety governance -- an agent with unrestricted access to enterprise systems is an unmanaged risk. The COMPEL framework requires that every agent's action space be explicitly defined, scoped, and controlled through the principle of least privilege: agents receive access only to the tools necessary for their defined function, with granular permissions specifying allowed operations, data scopes, rate limits, and temporal constraints. Tool access reviews occur at each Evaluate stage, and any expansion requires formal approval.
Why it matters
An AI agent with unrestricted access to enterprise systems represents an unmanaged risk that can cause financial loss, data breaches, or operational disruptions at machine speed. Defining and constraining the action space is the first line of defense for agent safety governance. Organizations that fail to explicitly scope agent capabilities discover vulnerabilities only after damage has occurred, often at significant cost to operations and reputation.
How COMPEL uses it
COMPEL requires that every agent's action space be explicitly defined and scoped through the Governance pillar's principle of least privilege. During Model, action spaces are designed with granular permissions specifying allowed operations, data scopes, rate limits, and temporal constraints. Tool access is reviewed during each Evaluate stage, and any expansion requires formal approval through the governance framework established in the Produce stage.
Related Terms
Other glossary terms mentioned in this entry's definition and context.