Quick overview
This workflow provides a chat-based gateway that screens requests with Anthropic, routes them to fast, specialist, or mission-planning agents, and optionally requests human approval via Gmail before returning a final response with risk and routing metadata.
How it works
- Receives an incoming chat message and normalizes it into a request payload with a session ID, timestamp, and approval threshold.
- Uses Anthropic to analyze policy compliance and risk, producing structured JSON that flags whether the request is allowed and which route to take.
- If the request is blocked, returns a policy refusal message to the chat.
- If routed to fast or specialist handling, uses Anthropic to either answer directly or orchestrate research, engineering, and data-analysis specialist agents and merge their results.
- If routed to mission planning, uses Anthropic to generate a structured mission plan by consulting Navigate, Perceive, and Protect sub-agents.
- For high-risk or approval-required mission plans, sends a Gmail approval email and continues only if the reviewer approves, otherwise returns a declined response.
- Critiques approved mission plans with an Anthropic critic and synthesizes a final operator brief, then returns the final response with route, risk score, and execution metadata.
Setup
- Add an Anthropic API credential and select it on all Anthropic chat model nodes used by the analyzer and agents.
- Add a Gmail OAuth2 credential for the approval step and replace the approver email address placeholder in the Gmail node.
- Set the approval threshold (0–10) in the request-normalization step to control when human review is required.
- Enable the workflow and test it from the chat UI with a simple question, a domain-specific request, and a mission-style request to verify routing and approvals.