Relationship Graph
How Agents Exchange Context
This graph outlines the most common collaboration loop. Use it as a template for designing your own agent networks.
Strategist → Planner
Translates intent into actionable plans.
Planner → Operator
Hands off tasks and constraints for execution.
Operator → Reviewer
Submits outputs for quality and alignment checks.
Reviewer → Strategist
Feedback loop for refinement or pivot.
Operator → Archivist
Stores artifacts and learnings for reuse.
Archivist → Planner
Provides historical context for future plans.
Relationship cues
Always pass goals, constraints, and confidence levels with every handoff.
Feedback cadence
Set explicit checkpoints for review and human decision-making.
Memory sync
Ensure learnings are written to shared memory after each loop.
Using the graph to diagnose workflow issues
The graph helps you identify where context breaks down. If a handoff fails, look at the edge between those two roles and document what information was missing. Often it is a simple omission—constraints, a decision history, or a clear definition of done. By fixing the edge, you improve every future run.
Use the feedback loop intentionally. The Reviewer to Strategist edge should not be constant; if it is, your initial briefs are too vague. If the Operator to Archivist edge is missing, you are losing learnings and repeating mistakes. The graph makes those gaps visible.
When scaling, extend the graph with specialized nodes rather than adding more edges. For example, insert a Researcher between Strategist and Planner, or split Reviewer into Compliance and Quality. This preserves clarity while expanding capability.
A good rule: if a handoff requires more than three follow-up questions, the edge needs a clearer contract.