Orchestration

Multi-Agent Patterns

Coordination primitives for orchestrating multiple agents across product requirements review workflows — handoffs, parallel execution, routing, and inter-agent communication.

Identity

Agent Cards

Individual agent identity cards showing name, role, current status, and active task. Cards reflect the agent's operational state in real time.

Controls

Source Collector

Gather project artifacts

Collecting delivery readiness artifacts

12 of 18 items gathered

Requirements Mapper

Map requirements to controls

Mapping Export workflow to internal control library

24 of 31 requirements mapped

Document Drafter

Author project brief sections

Drafting implementation notes product summary

Section 6.2 in progress

Operational table

Agent	Status	Task	Progress	Confidence	Cost	Updated
Source Collector Gather project artifacts	WorkingWorking	Collecting delivery readiness artifacts	x 67%	91%	$0.14	18s ago
Requirements Mapper Map requirements to controls	HandoffHandoff	Waiting for source packet before final coverage mapping	x 58%	82%	$0.09	42s ago
Document Drafter Author project brief sections	IdleIdle	No drafting until mapped requirements are confirmed	x 0%	n/a	$0.00	2m ago

Element	Details
Card layout	Responsive grid, each card with avatar, name, role, and state-specific details
Active state	Current task description and progress info
Idle state	Role-specific waiting note ("Waiting for a source packet to map")
Error state	Alert icon, error message, retry action button

Agent cards provide the identity foundation for multi-agent interfaces. In review workflow contexts, each card maps to a distinct review activity — source collection, requirements mapping, or document authoring — making it clear which agent is responsible for which workstream deliverable.

Orchestration

Handoff Flow

Sequential task handoff between agents. Each step produces output that becomes input for the next agent in the pipeline.

Controls

Export workflow coverage packet

Ownership transfer for a requirement gap that must be resolved before the review report can be finalized.

SenderRequirements Mapper

ReceiverDocument Drafter

Payload: Requirement coverage delta for Export workflow, including missing JSON export support.
Source basis: Project brief v3, Launch Policy v2, Support readiness checklist
Receiver action: Confirm whether Document Drafter should update implementation notes section 6.1.3.
Recovery: Rejecting the packet keeps the run blocked and asks Source Collector to gather more evidence.

Step 1

Parse project brief

Document Drafter

Step 2

Map requirement coverage

Requirements Mapper

Step 3

Generate review report

Report Generator

Element	Details
Step cards	Responsive step cards with step number, task label, and assigned agent
Connectors	Arrow icons between steps, opacity reflects whether handoff has occurred
Pending state	First step highlighted, remaining steps gently muted
Complete state	All steps show tick icon with subtle background tint

Handoff flows model the sequential dependencies in complex review workflows — a project brief must be parsed before requirement coverage can be mapped, and coverage must be mapped before the review report can be generated. Each handoff creates an auditable transition record.

Concurrency

Parallel Agents

Multiple agents executing independent tasks simultaneously. Each agent reports individual progress toward its own objective.

Controls

Risk Scanner

Scanning incident database against launch scope

72% complete

Source Collector

Gathering release readiness source material

45% complete

Policy Analyst

Analyzing launch policy

88% complete

Element	Details
Agent rows	Stacked cards with avatar, name, task description, and status indicator
Progress bars	Animated fill to individual target width, monochrome foreground tint
Running state	Pulsing dot, active task label, percentage complete
Complete state	Tick icon, results summary replaces task description, bar at 100%

Parallel execution is critical for complex review workflows where independent review workstreams — risk scanning (risk review), source material collection (ALC), and launch policy analysis — can proceed concurrently without blocking each other.

Dispatch

Agent Routing

Routing incoming tasks to the most appropriate agent based on task type analysis or manual selection by the workflow owner.

Controls

Incoming task

Review Export workflow export requirements

Task type: requirements review

Auto-routed by task type

Element	Details
Task card	Shows incoming task with detected type classification
Auto routing	Best-match agent highlighted, others dimmed at 40% opacity
Manual routing	All agents interactive, click to select destination
Selection indicator	Arrow icon with context label ("Best match" or "Selected")

Routing decisions in regulated workflows must be transparent and auditable. Auto-routing uses task type classification to match requirement-related tasks to the Requirements Mapper, source tasks to the Source Collector, and policy tasks to the Policy Analyst. Manual override ensures the reviewer retains final authority over agent assignments.

Collaboration

Agent Communication

Inter-agent information exchange via direct messaging or a shared context workspace. Both patterns create auditable communication records.

Controls

Direct thread — Requirements Mapper and Document Drafter

Requirements MapperDocument Drafter

I found that Export workflow references CSV-only export in the project brief, but the launch policy Launch Policy v2 requires CSV and JSON exports. Can you update section 6.1.3?

Document DrafterRequirements Mapper

Confirmed. I have updated the export behavior table in implementation notes to reference CSV and JSON export. The rationale now cites the data export policy.

Requirements MapperDocument Drafter

Verified. Export workflow mapping now aligns with the policy requirement. Marking this requirement as fully covered.

Element	Details
Direct messages	Chat-style thread between two agents with sender/receiver labels
Message alignment	Alternating left/right layout based on sender, subtle background tint
Shared context	List of contributed items with agent attribution, type badge, and timestamp
Context types	Artifact, Finding, Draft, Analysis, Assessment — shown as muted badges

Agent communication patterns support the traceability requirements of complex review workflows. Direct messaging creates point-to-point audit trails (e.g., when the Requirements Mapper flags an requirement gap for the Document Drafter), while shared context provides a workspace where all agents can contribute findings visible to the project team.