ExecutionAgent¶

Runs test scripts in MOCK / LOCAL / GRID mode and manages the Human-in-the-Loop approval gate.

Responsibility¶

The only agent that actually executes code. Manages the execution environment, HITL gate, trust domains, and collects raw results.

Input¶

class ExecutionInput(BaseModel):
    selenium_scripts: list[SeleniumScript]
    api_tests: list[APITest]
    execution_mode: str    # MOCK | LOCAL | GRID
    target_url: str
    approval_fn: Callable[[], bool] | None

Output¶

class ExecutionResult(BaseModel):
    test_case_id: str
    status: str            # "passed" | "failed" | "skipped" | "error"
    duration_ms: int
    error_message: str | None
    screenshot_path: str | None
    locator_used: str | None
    healing_triggered: bool

HITL gate¶

Before executing, the agent calls approval_fn() if provided:

# In CLI: approval_fn = None → auto-approve
# In Streamlit: approval_fn = lambda: st.session_state.get("hitl_approved", False)
# In Python API: user-supplied callback

if self.approval_fn is not None:
    approved = self.approval_fn({"test_cases": test_cases, "mode": mode})
    if not approved:
        return ExecutionResult(status="skipped", reason="HITL rejected")

Execution modes¶

MOCKLOCALGRID

The AI simulates test execution and generates realistic results based on the test steps. No real browser or HTTP calls. Used in CI/cloud environments.

# Gemini generates: would this test pass or fail given these steps?
# Returns plausible pass/fail with realistic error messages for failures

Launches a real Chrome browser via webdriver-manager. Requires Chrome installed locally.

from selenium import webdriver
from webdriver_manager.chrome import ChromeDriverManager

driver = webdriver.Chrome(ChromeDriverManager().install())

Connects to a Selenium Grid hub. Enables parallel execution across multiple browsers.

driver = webdriver.Remote(
    command_executor=settings.selenium_grid_url,
    options=chrome_options
)

Trust domains¶

The agent maintains a trust_domains table. URLs added to trusted domains bypass certain safety checks. Useful for staging environments.

# Add a trust domain
testpilot run --story "..." --url https://staging.internal.myapp.com
# staging.internal.myapp.com is auto-added to trust_domains