Agentic Workflow Rail.

Required input fields.

• Agent mission (required)
• Memory use (required): No persistent memory, Session memory, Persistent memory
• Retrieval scope (required): Curated corpus, Mixed internal/external, Open web or user upload
• Tool access (required): No tools, Read-only tools, Write/update tools, External communication or transaction
• Approval model (required): Human approval for material actions, Conditional approval, No approval
• Rollback path (required): Defined and tested, Partial, Unclear

Data handling: this interface uses the L2ET protected same-origin instrument engine. Do not enter confidential, regulated, privileged, incident, medical or sensitive operational data.

Swimlane logic.

Places agent operations into swimlanes and scores risk where memory, retrieval, tool use or external effects bypass policy or approval gates.

Source families

Assumptions

• It models intended architecture, not runtime behavior.
• Risk depends on actual prompts, tools and identity controls.
• Human review points must be operationally enforceable.

Agentic workflow control map.

The generated artifact includes findings, assumptions, limitations, recommended next actions and exportable structured output.

Export options

Example input and output.

Example input

Agent reads a curated corpus and calls read-only tools with conditional human approval and partial rollback.

Example output

Produces a swimlane map with approval gates, audit log requirements and rollback triggers.