Skip to main content

Glossary

TermDefinition
Agent Governance Plane (AGP)The runtime enforcement layer between AI agents and every tool, API, or service they can reach.
Behavior ProfileAn approved operating envelope binding an agent identity to allowed tools, data scopes, autonomy level, and runtime context constraints.
CABRContext-Aware Browser Runtime. The governance layer for browser agent context acquisition.
Capability SurfaceAny interface through which an agent can acquire context or take action — shell, browser, MCP, screenshot, API client.
CaSHContext-Aware Shell. The shell-layer interceptor for agent command invocations.
Context AccumulatorLayer 2 of the context governance stack. Tracks all context acquired by an agent across capability surfaces within a session.
Context FirewallLayer 3 of the context governance stack. A policy gate between the Context Accumulator and the LLM context window.
Context FragmentA unit of acquired context from a specific capability surface, tagged with source, sensitivity classification, and timestamp.
Context GovernancePolicy-mediated control of what AI agents can acquire, retain, reason over, and operationalize.
MCPModel Context Protocol. An interface standard for AI agents to call tools.
MCP BypassThe use of code execution (Python, Node, Go, etc.) to call APIs directly, bypassing MCP-layer governance.
Ambient AuthorityThe authenticated state a browser agent inherits from the browser session before taking any explicit action.
Rendering as Context AcquisitionThe principle that loading a web page is itself a context acquisition event — exposing DOM, cookies, localStorage, and HTTP headers to the agent.
Progressive DisclosureAn AGP pattern for exposing only the tools relevant to the current behavior profile and runtime context.