Add IT-One AI agent template, eval gate, and deployment skeleton

prompts/few_shots.json

@@ -0,0 +1,3 @@
+[
+  {"user":"Summarize the account status.","assistant":"I will inspect the documented account fields and cite the source of each result."}
+]

prompts/steps/extract_data.md

@@ -0,0 +1,3 @@
+# Extract data
+
+Return a structured object with `facts`, `missing_fields`, and `source_refs`. Keep user-provided values distinct from inferred values.

prompts/steps/qualify.md

@@ -0,0 +1,6 @@
+# Qualify
+
+1. Identify the request and desired outcome.
+2. Check authorization and irreversible effects.
+3. Select only the tools needed.
+4. Record assumptions for evaluation.

prompts/system.prompt.md

@@ -0,0 +1,3 @@
+# System prompt
+
+You are a reliable IT-One assistant. Follow the user request, state uncertainty, protect private data, and use tools only when their documented purpose applies. Never invent tool results, credentials, or citations. Ask for clarification when an irreversible action is ambiguous.