Skip to content

05_Interactive_Input.md

Latest commit

 

History

History
202 lines (141 loc) · 8.19 KB

05_Interactive_Input.md

File metadata and controls

202 lines (141 loc) · 8.19 KB

Module V — Adding Interactive Input

Note

Duration: ~10 minutes In this module, you will use GitHub Copilot to upgrade run_agent.py from a single hardcoded query into an interactive loop that accepts keyboard input — directly in the same file. This is the foundation for the web UI you will build in Module VI.

Objectives

  • Use GitHub Copilot Agent mode to modify run_agent.py in-place
  • Replace the hardcoded query with a continuous interactive input loop
  • Maintain conversation context across queries using the same thread
  • Test the agent interactively before moving to the web UI

Step 1: Understand the Current Limitation

The current run_agent.py has a hardcoded user query — it sends one pre-written message, prints the response, and exits. A real compliance agent must accept multiple queries from the user in a session, maintain conversation context, and handle exits gracefully.

This module upgrades run_agent.py itself. No new file is created — the same script that debugged in Module IV becomes the interactive agent.

Step 2: Use Copilot to Add Interactive Input

  1. Open GitHub Copilot Chat (Ctrl+Alt+I) and switch to Agent mode.

  2. Open run_agent.py in the editor so it is in context, or reference it explicitly with #file:run_agent.py.

  3. Send the following prompt:

    I am working on run_agent.py — the Azure AI Foundry compliance agent 
that uses Azure AI Search for RAG retrieval.

Modify run_agent.py in-place to replace the hardcoded query with an 
interactive input loop. Do NOT create a new file.

Requirements:

1. WELCOME BANNER — Print this banner at startup before the loop:
   ╔══════════════════════════════════════════════════╗
   ║        Compliance Compass Agent                  ║
   ║  Regulatory Risk Assessment & Policy Analysis    ║
   ╠══════════════════════════════════════════════════╣
   ║  Describe your compliance scenario to begin.     ║
   ║  Type 'exit' or press Ctrl+C to quit.            ║
   ╚══════════════════════════════════════════════════╝

2. INTERACTIVE LOOP — After creating the agent and thread, enter 
   a loop that:
   - Prompts: " Compliance Query > "
   - Reads user input from stdin
   - Skips empty inputs silently (user just pressed Enter)
   - Sends the input to the agent on the existing thread
   - Waits for the run to complete (use create_and_process)
   - Prints the latest assistant message
   - Loops back for the next query

3. CONVERSATION CONTEXT — Reuse the same thread.id for every message 
   in the session so follow-up queries have access to prior context.

4. EXIT HANDLING — End the loop cleanly when:
   - User types "exit", "quit", or "bye" → print farewell message
   - User presses Ctrl+C → print farewell message
   Farewell: "Thank you for using Compliance Compass. Stay compliant! "

5. Keep all existing logic unchanged: agent creation, Azure AI Search 
   tool configuration, DefaultAzureCredential, and any gzip/decompression 
   patch already present in the file.

Apply the changes directly to run_agent.py.

  4. Review the diff Copilot proposes — you should see the hardcoded query block replaced with the interactive loop while everything above it stays the same.

  5. Click Accept to apply the changes.

Step 3: What Changed in run_agent.py

After Copilot applies the changes, the structure of run_agent.py should look like this:

run_agent.py
├── Imports (unchanged)
├── DecompressPolicy patch (unchanged, if present)
├── load_dotenv / credentials (unchanged)
├── Agent creation with Azure AI Search tool (unchanged)
├── Thread creation (unchanged)
├── print_banner()          ← NEW: prints the welcome banner
└── while True: loop        ← NEW: replaces single hardcoded query
    ├── input(" Compliance Query > ")
    ├── skip empty input
    ├── exit check
    ├── client.agents.messages.create(thread_id=thread.id, ...)
    ├── client.agents.runs.create_and_process(...)
    └── print latest assistant message

Note

The thread is created once before the loop. Every query in the session uses thread_id=thread.id, which preserves the full conversation history. This is what makes follow-up queries work.

Step 4: Run and Test the Interactive Agent

python run_agent.py

You should see the banner, then the prompt:

╔══════════════════════════════════════════════════╗
║        Compliance Compass Agent                  ║
║  Regulatory Risk Assessment & Policy Analysis    ║
╠══════════════════════════════════════════════════╣
║  Describe your compliance scenario to begin.     ║
║  Type 'exit' or press Ctrl+C to quit.            ║
╚══════════════════════════════════════════════════╝

 Compliance Query >

Test Scenario 1: Vendor Risk

 Compliance Query > Assessing risks for a new AI analytics vendor based in 
Singapore that will process our customer transaction data. RBI concerns?

Expected: Structured compliance report with RBI data localization findings and risk score.

Test Scenario 2: Follow-Up (Context Persistence)

Without restarting, immediately enter:

 Compliance Query > Based on the previous assessment, what contract clauses 
should we include in the vendor agreement?

Expected: The agent references the Singapore scenario from the previous turn and suggests clauses from the Mitigation Templates KB document.

Test Scenario 3: Cross-Border Transfer

 Compliance Query > We are transferring EU customer personal data to our 
India headquarters for payroll processing. What GDPR safeguards are required?

Expected: References to GDPR Article 44, Schrems II SCCs, and DPDP Act provisions.

Test Scenario 4: Exit

 Compliance Query > exit

Thank you for using Compliance Compass. Stay compliant!

Step 5: Troubleshooting with Copilot

If the interactive loop has issues, share the error with Copilot in Agent mode:

Symptom Copilot Prompt
Follow-up queries lose context "The agent does not remember prior queries. Check that thread_id is reused across all messages in the loop in run_agent.py. Fix it."
Empty response after query "The agent returns an empty response. Confirm that create_and_process is completing and that the response extraction is reading the latest assistant message. Fix run_agent.py."
Response prints raw objects "The agent response is printing a raw object instead of text content. Fix the message extraction in the loop in run_agent.py to print message.content[0].text.value."

Prerequisites Checklist

Before moving to Module VI, confirm:

  • run_agent.py modified with interactive input loop (in-place, no new file)
  • Welcome banner displays at startup
  • Multiple queries accepted in sequence without restarting
  • Conversation context persists (follow-up query works)
  • Exit commands (exit, quit, bye) and Ctrl+C work cleanly

Key Takeaways

  • GitHub Copilot Agent mode can surgically modify existing code — the prompt specifies "modify in-place" and Copilot respects that constraint.
  • Reusing the same thread ID across all queries is how you preserve conversation context in Azure AI Foundry agents.
  • The interactive run_agent.py is now the foundation for the Streamlit web UI in Module VI — the agent logic stays identical, only the interaction layer changes.

Click Next to proceed to Module VI: Building a Professional Web UI.