Update BLOG.md

BLOG.md

@@ -88,6 +88,21 @@ As seen in this zoomed-in perspective, the **ACTION TIMELINE** perfectly chronic
 
 By integrating state visibility and immediate reward telemetry, we transformed theoretical Reinforcement Learning success into a tangible, closed-loop deployable solution.
 
+### Use Case Diagram
+
+![use-case-diagram](Use_Case_diagram.png)
+
+
+The Execution Flow:
+
+State Initialization: The agent receives the topic (e.g., "Draft a FinTech App").
+
+Constraint Querying: The agent sends targeted WorkSpaceAction JSONs to the Finance, Security, and UX experts. Each successful query "discovers" a constraint, adding to the agent's internal context.
+
+The 40-Token Gauntlet: Every action must pass the Pass-Through Sieve. If the agent's reasoning is too "wordy," the sieve rejects the action, forcing the agent to learn hyper-compression.
+
+Final Synthesis: Once all constraints are discovered, the agent triggers the submit_final action, which pulls all discovered context into the PRD Final Draft module
+
 
 ### 🛠️ Technical Stack
 - **Environment:** OpenEnv (State-based workspace)