Golden Flows
The 5 normative integration scenarios that every MPLP-conformant runtime is expected to pass. These flows serve as the 'integration tests' for the protocol.
Terminology Note
- Flow-01~05 = Protocol Test Scenarios (defined in main repository)
- LG-01~05 = Lifecycle Guarantees (Lab adjudication targets)
These are distinct naming spaces and should not be conflated. See Lifecycle Guarantees (LG-01~05) in Validation Lab →
Golden Flows are formally defined in the MPLP Protocol Specification.
See: governance/CONFORMANCE_MODEL.md for normative requirements.
Single Agent – Happy Path
The baseline single-agent workflow. Agent receives context, generates a minimal 2-step plan, executes, and produces trace.
Context Creation
Plan Generation
Task Execution
Trace Recording
Verification
Context Creation
Context established with domain and environment.
Plan Generation
2-step plan created with deterministic UUIDs.
Task Execution
Steps executed in dependency order.
Trace Recording
All state transitions recorded.
Verification
Output validated against expected fixtures.
Normative Scope: Context · Plan · Trace
Single Agent – Large Plan
Volumetric validation with 20+ steps. Tests protocol handling of large execution plans.
Large Context
Multi-Step Plan
Volumetric Execution
Trace Verification
Large Context
Context framing batch processing scenario.
Multi-Step Plan
20-30 heterogeneous steps with dependencies.
Volumetric Execution
Steps processed with wildcard invariant validation.
Trace Verification
Events ordered correctly despite volume.
Normative Scope: Context · Plan · Trace
Single Agent – With Tools
Tool integration via agent_role field. Validates protocol semantics for tool-enabled workflows.
Tool Context
Tool Steps
Role-Based Execution
Result Capture
Tool Context
Context for API testing workflow.
Tool Steps
Steps with agent_role: curl_executor, jq_processor.
Role-Based Execution
Each role handles specific tool type.
Result Capture
Tool outputs recorded in trace.
Normative Scope: Context · Plan · Trace · Extension
Single Agent with LLM Enrichment
AEL (Action Execution Loop) integration. Validates LLM-enriched plan generation.
Initial Context
LLM Enrichment
Plan Validation
Execution
Initial Context
Context with high-level intent.
LLM Enrichment
AEL generates enriched plan from intent.
Plan Validation
Enriched plan passes schema validation.
Execution
Plan executed with full traceability.
Normative Scope: Context · Plan · Trace · Core
Single Agent with Confirm Required
Multi-round approval workflow. Validates Confirm module integration.
Plan Creation
Confirm Request
Approval/Rejection
Conditional Execution
Plan Creation
Plan requiring confirmation before execution.
Confirm Request
Confirmation request created with approver_role.
Approval/Rejection
Decision captured with timestamp and notes.
Conditional Execution
Plan proceeds only if approved.
Normative Scope: Context · Plan · Confirm · Trace
SA Basic Execution
Single-Agent (SA) profile baseline. Single-step execution with full lifecycle.
SA Initialize
Load Context
Execute Step
Complete
SA Initialize
SA profile initialization.
Load Context
Context loaded with SA-specific fields.
Execute Step
Single step executed with agent_role.
Complete
SA lifecycle completed cleanly.
Normative Scope: Context · Plan
SA Multi-Step Evaluation
SA profile with multi-step plan. Validates step sequencing in SA mode.
Multi-Step Plan
Sequential Execution
State Tracking
Multi-Step Plan
Plan with multiple ordered steps.
Sequential Execution
Steps processed by order_index.
State Tracking
Each step status updated correctly.
Normative Scope: Context · Plan
MAP Turn-Taking Session
Multi-Agent Protocol with turn-taking. Two agents collaborate with role rotation.
Session Init
Role Assignment
Turn 1: Plan
Turn 2: Review
Turn 3: Revise
Session Init
MAP session with round_robin mode.
Role Assignment
Agent A (planner), Agent B (reviewer).
Turn 1: Plan
Planner creates initial plan.
Turn 2: Review
Reviewer evaluates plan.
Turn 3: Revise
Planner revises based on feedback.
Normative Scope: Context · Plan · Collab · Role
MAP Broadcast Fan-out
Multi-Agent Protocol with broadcast. Parallel dispatch to multiple agents.
Broadcast Init
Multi-Participant
Parallel Dispatch
Result Collection
Aggregation
Broadcast Init
MAP session with broadcast mode.
Multi-Participant
3+ agents registered.
Parallel Dispatch
Task broadcast to all participants.
Result Collection
Responses collected from all agents.
Aggregation
Results merged into final output.
Normative Scope: Context · Plan · Collab · Role
Normative
Protocol Conformance Boundary
Golden Flows define the minimum executable scenarios required for MPLP conformance. They are not examples, tutorials, or best practices.
Implementations may be evaluated against Golden Flows as described in the specification.
This website provides discovery and positioning content only. Normative requirements live in Docs; the ultimate source of truth is the Repository.
Evidence is evaluated through documentation-defined scenarios and replayable artifacts. This website does not certify or endorse implementations.