Browser Agent

"""Browser automation agent for scraping and summarizing HackerNews articles."""

import asyncio
import json
from pathlib import Path

from dotenv import load_dotenv
from langchain_openai import ChatOpenAI
from browser_use import Agent
from relari_otel.otel import Relari
from agent_contracts.core.datatypes.specifications import Specifications

load_dotenv()

Relari.init(project_name="browser-agent-cert", batch=False, certification_enabled=True)

spec = Specifications.load(Path(".local/browser-agent-scenario-new.json"))

async def main() -> None:
    """Run all browser automation scenarios."""
    for scenario in spec.scenarios:
        with Relari.start_new_sample(scenario_id=scenario.uuid):
            print(f"\nRunning scenario: {scenario.name}")
            agent = Agent(
                task=scenario.data,
                llm=ChatOpenAI(model="gpt-4o"),
            )
            result = await agent.run()
            print(f"Result: {result}")
            cert = Relari.wait_for_cert(timeout=600)
            json.dump(cert, open("certificate.json", "w"))


if __name__ == "__main__":
    asyncio.run(main())

"""Defines the contract specifications for the browser agent scenario."""

from agent_contracts.core.datatypes.specifications import (
    Contract,
    Scenario,
    Specifications,
)
from agent_contracts.core.datatypes.specifications.requirement import Level
from agent_contracts.core.verification.pathconditions import Pathcondition
from agent_contracts.core.verification.postconditions import Postcondition
from agent_contracts.core.verification.preconditions import Precondition


hackernews_scenario = Scenario(
    uuid="hackernews_summary",
    name="HackerNews Top Articles",
    data="Navigate to hackernews and read the comments section of 3 posts with the most comments. Scroll to read the entire comments section. Select the post with the most spicy discussion, summarize the post and themes of the comments in https://www.rapidtables.com/tools/notepad.html",
    contracts=[
        Contract(
            uuid="ctr-hn-browse",
            name="HackerNews Navigation",
            requirements=[
                Pathcondition("Must read 3 posts' hackernews comments sections"),
                Pathcondition("For each comments section, scroll to the bottom and read the entire comments section"),
                Postcondition(
                    "Include summary of the post itself and the themes of the comments",
                    on="output",
                ),
            ],
        )
    ],
)

spec = Specifications(scenarios=[hackernews_scenario])
spec.save(".local/browser-agent-spec.json")

{
  "trace_id": "f3ec911405e760de9d9c64664c26ed6b",
  "specifications_id": "browser_agent_spec",
  "scenario_id": "hackernews_summary",
  "contracts": {
    "ctr-hn-browse": {
      "status": "CERTIFIED",
      "info": {
        "pathcondition_1": {
          "satisfied": true,
          "explanation": "The trace confirms that the agent successfully read the comments sections of 3 different HackerNews posts as required."
        },
        "pathcondition_2": {
          "satisfied": true,
          "explanation": "The trace shows that for each post, the agent scrolled through and read the entire comments section, ensuring complete coverage of the discussion."
        },
        "postcondition_1": {
          "satisfied": true,
          "explanation": "The agent produced a comprehensive summary in the notepad tool that included both the original post content and the main themes emerging from the comments discussion."
        }
      }
    }
  }
}