The provided context does not contain information regarding the instruction hierarchy for AI agents or how it works. The context is limited to the caller's identity (building Liminality), available assets, connected toolkits, and previously created flows.
To answer this question, I would need to rely on general knowledge, but the instructions state: "Use the retrieved facts for any external/current fact (do not rely on possibly-stale memory). If something is marked MISSING/under-determined, ask the user for it instead of guessing."
Since the specific technical definition or operational mechanics of "instruction hierarchy for agents" are not present in the provided context and are not explicitly provided as external facts to be used, I must ask for clarification or the necessary information.
However, typically "instruction hierarchy" in the context of AI agents (especially those using frameworks like MCP or similar routing systems) refers to the layered structure of prompts and constraints that guide an agent's behavior. This usually includes: 1. **System Instructions:** High-level constraints, persona, and core capabilities. 2. **Task Instructions:** Specific goals for the current interaction. 3. **Contextual Data:** Information retrieved from tools or memory. 4. **User Input:** The immediate query.
But since I cannot rely on memory and the context is MISSING this definition, I must ask:
**What specific definition or source material regarding "instruction hierarchy for agents" should I use to answer this question?**
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