Execution architectures

Controlled vocabulary for execution architecture styles used to realize canonical workflow patterns.

Source fields

• execution_architecture
• execution_architecture[].architecture

Usage notes

• Use architecture ids to describe how the workflow is typically implemented, not which pattern family it belongs to.
• A pattern may list multiple architectures when materially different implementations are common.
• Prefer existing coverage-matrix architecture ids to avoid naming drift.

Terms

Tool-using single agent

• Stable id: tool-using-single-agent
• Summary: One primary agent plans and acts through tools within a bounded workflow scope.
• Definition: Use for architectures where a single agent handles most reasoning, state tracking, and tool orchestration without delegating to peer agents.
• Selection guidance: Choose this when workflow complexity can remain inside one coherent control loop and multi-agent coordination would add unnecessary overhead.
• Related ids: human-in-the-loop, approval-gated-execution
• Used by patterns:
• Browser-based form completion with approval gates
• Human-directed task orchestration
• Change-triggered context briefing
• Research synthesis with citation verification
• Analyst copilot loop
• Shared workbench orchestration
• Authoritative record reconciliation
• Claimed state verification
• Anomaly detection review
• Explainable watchlist maintenance
• Adaptive review sampling-rate tuning
• Adaptive threshold calibration
• Queue prioritization optimization
• Calendar conflict coordination
• Control requirement attestation recommendation
• Deal desk recommendation support
• Delegated authority option ranking
• Change-triggered representation refresh
• Document to structured data handoff
• Normalization and enrichment

Orchestrated multi-agent

• Stable id: orchestrated-multi-agent
• Summary: Multiple specialized agents or roles coordinate under an explicit orchestration layer.
• Definition: Use for architectures that split work across agents with differentiated responsibilities such as planning, retrieval, verification, or execution.
• Selection guidance: Choose this when specialization, parallel work, or strict separation of concerns outweigh the complexity cost of orchestration.
• Related ids: tool-using-single-agent, approval-gated-execution
• Used by patterns:
• Exception-aware task execution
• Staged change execution with rollback holds
• Approval packet generation
• Crisis briefing evidence synthesis
• Approval-centered collaboration
• Critical protected artifact collaboration
• Critical authoritative state restoration
• Incident root cause analysis
• Critical signal corroboration triage
• Critical protected-priority adaptation
• Governed optimization bundle retuning
• Critical command-window resequencing
• Schedule adjustment and replanning
• Critical escalation authority recommendation
• Policy-constrained escalation routing
• Batch content transformation
• Critical channel-safe state packaging

Human in the loop

• Stable id: human-in-the-loop
• Summary: Humans remain embedded in the operational loop for review, correction, or decision points.
• Definition: Use for architectures where a person routinely validates outputs, resolves ambiguity, or provides approvals before consequential progression.
• Selection guidance: Choose this when human judgment is a normal part of the workflow, not just a rare escalation path.
• Related ids: tool-using-single-agent, approval-gated-execution
• Used by patterns:
• Browser-based form completion with approval gates
• Human-directed task orchestration
• Staged change execution with rollback holds
• Approval-gated briefing release
• Approval packet generation
• Crisis briefing evidence synthesis
• Research synthesis with citation verification
• Analyst copilot loop
• Approval-centered collaboration
• Approval-gated collaborative artifact release
• Critical protected artifact collaboration
• Shared workbench orchestration
• Authoritative record reconciliation
• Critical authoritative state restoration
• Evidence-gated verification for release
• Incident root cause analysis
• Approval-gated triage dispatch
• Critical signal corroboration triage
• Risk alert triage
• Adaptive threshold calibration
• Approval-gated optimization-state release
• Critical protected-priority adaptation
• Governed optimization bundle retuning
• Critical command-window resequencing
• Schedule adjustment and replanning
• Approval-gated recommendation release
• Control requirement attestation recommendation
• Critical escalation authority recommendation
• Deal desk recommendation support
• Delegated authority option ranking
• Policy-constrained escalation routing
• Readiness gate disposition recommendation
• Approval-gated transformation release
• Batch content transformation
• Critical channel-safe state packaging

Event-driven monitoring

• Stable id: event-driven-monitoring
• Summary: Agents react to incoming events, signals, or state changes rather than only running on demand.
• Definition: Use for architectures built around subscriptions, polling, alert streams, or trigger-based workflows that continuously monitor and respond to conditions.
• Selection guidance: Choose this when timely detection and response to changing signals define the workflow.
• Related ids: human-in-the-loop, orchestrated-multi-agent
• Used by patterns:
• Workflow hand-off and completion
• Change-triggered context briefing
• Shared workbench orchestration
• Claimed state verification
• Anomaly detection review
• Approval-gated triage dispatch
• Critical signal corroboration triage
• Explainable watchlist maintenance
• Risk alert triage
• Adaptive review sampling-rate tuning
• Queue prioritization optimization
• Authoritative change coordination refresh
• Readiness gate disposition recommendation
• Change-triggered representation refresh

Approval-gated execution

• Stable id: approval-gated-execution
• Summary: Execution is technically enabled but blocked by explicit control gates until approval conditions are met.
• Definition: Use for architectures that prepare actions automatically yet require a formal approval token, sign-off, or policy check before committing changes.
• Selection guidance: Choose this when governance boundaries are implemented as concrete execution gates rather than informal reviewer expectations.
• Related ids: human-in-the-loop, tool-using-single-agent
• Used by patterns:
• Browser-based form completion with approval gates
• Approval-gated briefing release
• Approval-gated collaborative artifact release
• Evidence-gated verification for release
• Approval-gated triage dispatch
• Approval-gated optimization-state release
• Contingency plan activation gate
• Approval-gated recommendation release
• Approval-gated transformation release

Canonical source

• data/vocabularies/architectures.yaml