To build a reusable workflow library for AI agents in your **Liminality** project, use the architectural patterns already established in your existing flows.
1. **Decompose questions with AND logic:** Your current flows, like the Monaco attack or Mount Rushmore queries, show that complex questions break into co-required sub-questions. Standardize this. Each entry must define a primary question and its mandatory sub-components. The agent must address all parts simultaneously, not sequentially.
2. **Ground sub-questions to real tools:** Liminality grounds queries to active toolkits. Map each sub-question to specific tools. If a query needs data, link it to `airtable`, `hubspot`, or `google_search_console`. If it needs creative assets, link to `figma` or `googleslides`. This ensures AI models can execute the workflows by calling your MCP as a tool.
3. **Structure as reusable routes:** Liminality targets human and AI adoption. Structure workflows as "reusable routes." Use a standardized schema where inputs (decomposed questions) and outputs (routed tool calls) are consistent. This lets flagship LLMs invoke workflows predictably without custom configuration for each case.
4. **Use existing assets for context:** Reference `Acme CRM`, `Motherboard`, and `physea.ai` for domain-specific workflows. A customer support workflow routes sub-questions to `Acme CRM` or `hubspot`. A content creation workflow routes to `figma` or `googledrive`.
**Pending Action:** Confirm which domain to prioritize first for specific workflows: customer support via Acme CRM/Hubspot, or content creation via Figma/Google Drive.
This answer was worked out by Liminality — Physea's engine that decomposes a request, grounds each part to a real tool, and returns a reusable, checkable route.
Connect it over MCP: https://mcp.physea.ai/mcp · physea.ai