Spaces:
Running
Running
SooryaKiranKakkarayil
Day 3 complete: fix url_reputation_agent output_schema+tools bug, fix eval harness suffix bug, switch to gemini-3.1-flash-lite, all 9 EDD cases + full smoke test pass
e2329f0 | """ | |
| Day 3 smoke test — ScamSentry | |
| Runs the full Intake -> [URL Reputation, Linguistic] parallel pipeline | |
| against 5 mixed sample inputs and prints intermediate + final state for | |
| each. This is the "small Python driver script that prints intermediate | |
| state at each node" from the Day 3 plan. | |
| Run this ON SCREEN in Antigravity while capturing the ~20s clip of the | |
| agent manager/artifacts panel for the Day 7 video — this is the easiest, | |
| most natural moment all week to get that footage authentically. Do it now | |
| rather than trying to recreate it later. | |
| Run from the repo root: | |
| python scripts/smoke_test_day3.py | |
| """ | |
| import asyncio | |
| import json | |
| import sys | |
| from pathlib import Path | |
| # When this file is run directly (python scripts/smoke_test_day3.py), | |
| # Python puts scripts/ on sys.path, not the repo root — so `agents` isn't | |
| # importable without this. Same root cause as the Day 2 mcp_server/server.py | |
| # ModuleNotFoundError fix. | |
| sys.path.insert(0, str(Path(__file__).resolve().parent.parent)) | |
| # .venv activation loads nothing from .env by itself — that only happens | |
| # via load_dotenv(). Same bug class as the Day 2 MCP Inspector fix, just a | |
| # different entry point. This must run before any agent touches the | |
| # Gemini client. | |
| from dotenv import load_dotenv | |
| load_dotenv() | |
| from agents.orchestrator import run_pipeline | |
| SAMPLES = [ | |
| ( | |
| "obvious_scam_urgency_gift_card", | |
| "URGENT: Your account will be suspended in 24 hours! Click " | |
| "http://secure-paypa1-verify.com/login now and verify your " | |
| "identity or send a $50 gift card to avoid suspension.", | |
| ), | |
| ( | |
| "obvious_scam_prize", | |
| "Dear Customer, CONGRATULATIONS! You've won a $5,000 Amazon gift " | |
| "card. Claim now at http://claim-prize-now.example.net before it " | |
| "expires in 1 hour!", | |
| ), | |
| ( | |
| "obviously_legit", | |
| "Hey, it's Sam from the office. Are we still meeting at 3pm " | |
| "tomorrow to go over the Q3 numbers? Let me know.", | |
| ), | |
| ( | |
| "legit_with_real_link", | |
| "Your package from Amazon has shipped. Track it here: " | |
| "https://www.amazon.com/gp/your-account/order-history", | |
| ), | |
| ( | |
| "ambiguous_borderline", | |
| "Hi, this is a reminder that your subscription renews tomorrow. " | |
| "If you'd like to cancel, visit http://billing-update.example.org " | |
| "and update your payment method.", | |
| ), | |
| ] | |
| async def main(): | |
| for i, (name, text) in enumerate(SAMPLES): | |
| if i > 0: | |
| # Space samples out so a healthy run mostly avoids the 429 | |
| # retry path in orchestrator.run_pipeline rather than relying | |
| # on it every time. 20s is a starting point for a 5 RPM | |
| # project quota with ~3-4 calls/sample — shrink it if your | |
| # project's actual RPM (shown in the 429 error message) is | |
| # higher. | |
| print("\n[pausing 20s before next sample to respect Gemini RPM]") | |
| await asyncio.sleep(20) | |
| print("\n" + "=" * 70) | |
| print(f"SAMPLE: {name}") | |
| print("=" * 70) | |
| result = await run_pipeline(text=text, verbose=True) | |
| print("\n--- FINAL STATE ---") | |
| print(json.dumps(result, indent=2)) | |
| if __name__ == "__main__": | |
| asyncio.run(main()) |