Skip to main content
INFORMATIVEACTIVEDocumentation Governance
Truth Source: Repository schemas and tests are authoritative.

FLOW-02: Single Agent – Large Plan

Source of Truth: tests/golden/flows/flow-02-single-agent-large-plan/

Purpose​

Volumetric validation with 20+ steps. Tests protocol handling of large execution plans to validate that the runtime can execute a Plan with many steps while maintaining invariants.

Scope​

This evaluation scenario validates:

  • Plan with 20-30 heterogeneous steps
  • Trace handling volumetric event streams
  • Step ordering preservation (temporal causality)
  • Performance stability as step count grows

Non-Goals​

This scenario does NOT evaluate:

  • Minimal 2-step flows (see FLOW-01)
  • Tool integration (see FLOW-03)
  • LLM enrichment (see FLOW-04)
  • Multi-round approval (see FLOW-05)

L2 Modules Exercised​

ModuleRole in Flow
ContextFrames large-scale refactoring or batch processing scenario
PlanContains 20-30 heterogeneous steps with dependencies
TraceHandles volumetric event streams efficiently

Key Protocol Fields​

Plan (Large)​

  • steps[]: 20-30 steps
  • Each step:
    • step_id: UUID v4
    • description: Non-empty, realistic task descriptions
    • status: "pending" → "in_progress" → "completed"
    • dependencies: Optional array of prior step IDs
    • order_index: Optional integer for explicit ordering

Trace (Volumetric)​

  • spans[]: One entry per step (20-30 spans)
  • events[]: Step completion events
  • Event ordering must be preserved

Integration Dimensions (L3/L4)​

None. This flow intentionally excludes:

  • Tool Integration
  • LLM Backend
  • Storage Integration

The isolation ensures any performance or correctness issues are attributable to L2 protocol layer.

Evidence​

TypeLocationStatus
Golden Flowtests/golden/flows/flow-02-single-agent-large-plan/✅ Passed
Input Fixturestests/golden/flows/flow-02-single-agent-large-plan/input/Available
Expected Fixturestests/golden/flows/flow-02-single-agent-large-plan/expected/Available

Expected Behavior​

  • All 20+ steps complete without error
  • Step ordering is preserved (dependency chains respected)
  • No performance degradation with large plans
  • Trace correctly records all step events
  • Event ordering maintained (temporal causality)

Invariants Tested​

  • Plan structure handles 20+ steps without schema violations
  • All step_id values unique UUID v4
  • Dependency chains are acyclic and resolvable
  • Trace event count matches step count

Document Status: Informative (Evaluation Scenario)
Source of Truth: tests/golden/flows/flow-02-single-agent-large-plan/README.md