Terminology¶
This page defines the shared naming model used across the repository so runtime does not mean both “the execution system” and “the thing being executed.”
Short version:
Agent Engine runs an Agent Runtime and extends it with Agent Spec.
If you want the system map that shows how these concepts fit together, continue with:
Core Terms¶
| Term | Role | Meaning |
|---|---|---|
Agent Engine |
Execution system | Loads, schedules, executes, recovers, audits, and exposes API / SSE. |
Agent Runtime |
Primary runnable unit | A runnable unit that carries agent / action / hook definitions. Formerly called blueprint. |
Agent Spec |
User extension layer | Extra user-authored inputs layered onto a runtime, mainly AGENTS.md, MEMORY.md, and extra loaded model / tool / skill. |
Boundary Rule¶
Engine: how it runsRuntime: what runsSpec: what the user adds
What Belongs To Runtime¶
These concepts belong to Runtime and should not be renamed as another kind of spec:
- agent definitions
- action definitions
- hook definitions
- runtime-owned default behavior and capability composition
- runnable units under
runtimes/
Avoid:
agent spechook specruntime spec
What Belongs To Spec¶
Spec is not the whole runtime structure. It is the user-authored extension layer added on top of a runtime. It mainly includes:
- project-root
AGENTS.md - project-root
MEMORY.md - extra loaded
model - extra loaded
tool - extra loaded
skill
Naming Rules To Avoid Ambiguity¶
When the meaning is about execution flow rather than the primary runnable unit, prefer:
enginerunsessionexecutionengine state
So use names like:
EngineLoggerengine.logengine state paths