Coordinator-subagent orchestration

• The coordinator handles all inter-subagent communication, error handling, and information routing.
• Subagents run in isolated context; they do not inherit the coordinator's conversation.
• The coordinator decomposes, delegates, aggregates, and decides which subagents a query needs.
• Decomposing a broad topic too narrowly leaves coverage gaps even when every subagent succeeds.
• Scope should be partitioned (distinct subtopics or source types) so subagents do not duplicate work.
• Synthesis output should be checked for gaps and improved by targeted re-delegation.

• Design a coordinator that analyses the query and dynamically selects subagents instead of always running the full pipeline.
• Partition research scope to minimise duplication.
• Run an iterative refinement loop: evaluate synthesis, re-delegate targeted queries, re-run synthesis until coverage is sufficient.
• Route all subagent communication through the coordinator for observability and consistent error handling.

• "If every subagent succeeded, the report must be complete." Gaps are often created at decomposition time.
• "The synthesis agent should find missing topics." It can only work with what it was given.
• "Letting subagents talk directly is a harmless shortcut." It breaks tracing and uniform error handling.

Always running the full pipeline, overlapping subagent scopes, and side channels that bypass the coordinator.

• 1.3Subagent invocation and context passing
• 1.6Task decomposition strategies
• 5.3Error propagation across agents