[!FROZEN] MPLP Protocol v1.0.0 Frozen Specification Freeze Date: 2025-12-03 Status: FROZEN (no breaking changes permitted) Governance: MPLP Protocol Governance Committee (MPGC) License: Apache-2.0 Note: Any normative change requires a new protocol version.

Orchestration

[!NOTE] Duty Type: OS-Level Kernel Duty
SOT Reference: README v1.0.0 Section 8

Intent

To manage the flow of execution, enforcing dependencies, concurrency limits, and stage transitions within the agent lifecycle.

Lifecycle Coverage

• Execution: Scheduling steps and managing the pipeline.

Agent Scope (SA / MAP)

• SA: Linear step execution.
• MAP: Complex DAG execution with parallel branches.

Required Events

• PipelineStageStarted
• PipelineStageCompleted
• WorkflowPaused

Compliance Requirements

1. Runtime MUST respect dependency graphs (Step B cannot start until Step A completes).
2. Runtime MUST emit PipelineStageEvent for every state transition.
3. Runtime MUST support pausing and resuming workflows.

Implementation Details (Non-Normative)

Orchestration is primarily handled by the Plan Module. The Plan object serves as the executable blueprint for the agent's actions.

Plan Structure

The Plan object (mplp-plan.schema.json) decomposes a high-level objective into atomic units of work:

• steps: An array of PlanStep objects, each representing a discrete task.
• dependencies: Within each step, a list of step_ids that must complete before execution can proceed. This explicitly defines the Directed Acyclic Graph (DAG) of execution.
• status: Tracks the lifecycle of the plan (draft -> proposed -> approved -> in_progress -> completed).

Step Execution

Each step includes:

• agent_role: The specific role responsible for executing the step.
• status: Granular tracking of step progress (pending, in_progress, blocked, completed).

Schema Reference

Schema	Purpose	Key Fields
mplp-plan.schema.json	Defines the execution plan	plan_id, steps, status, objective
plan_step_core ($defs)	Defines a single step	step_id, dependencies, agent_role, status

Examples

• DAG Execution: Running 3 research agents in parallel, then aggregating their results in a final summary step.
• Approval Gate: Pausing execution until a human confirms the plan.