AEL - Action Execution Layer
Status: Draft runtime concept page
Authority: Non-authoritative documentation surface
Boundary: This page describes AEL as a runtime execution abstraction. It does not define provider-specific adapters or a single required execution engine.
1. Purpose
AEL, or Action Execution Layer, is the runtime-side abstraction responsible for carrying out actions that have side effects or require external execution.
Typical runtime actions include:
- LLM calls
- tool invocations
- agent-to-agent handoff execution
- external side-effect execution under runtime control
2. Boundary
This page does not define:
- a required OpenAI adapter
- a required Anthropic adapter
- a required Docker executor
- a universal retry policy
- a mandated sandbox product
Those belong to runtime or product implementations.
3. Relation to Protocol Truth
The protocol defines lifecycle objects, invariants, and event families.
Runtimes use AEL-like logic to execute actions while staying within those protocol constraints. AEL is therefore a runtime realization concept, not a protocol schema object.
4. Minimal Runtime Artifact Surface
The current published runtime package includes a minimal InMemoryAEL artifact
surface in @mplp/runtime-minimal.
That artifact is a minimal runtime helper, not a claim that MPLP standardizes a single production execution engine.
5. Implementation Reading
The safest reading of AEL in MPLP is:
- it is the runtime-side execution abstraction
- it may execute tools, models, or other actions
- it should remain observable and bounded by runtime policy
- it is not itself protocol truth
6. References
- Runtime Glue Overview
- PSG - Project Semantic Graph
packages/npm/runtime-minimal/
Final Boundary: AEL is a runtime execution concept. It is not a protocol object definition and not a claim about one specific provider/runtime stack.