ActionResult

Result type for tool execution and task completion

The ActionResult class represents the result of executing an action (tool). It provides structured output, error handling, and metadata for agent communication.

Constructor

from browser_use import ActionResult

result = ActionResult(
    extracted_content="Found 10 products",
    long_term_memory="Searched for laptops on Amazon",
)

Parameters

extracted_content · str | None

Main content to show to the agent. This is the primary output of the action.

long_term_memory · str | None

Compact memory for long-term context. Used to summarize the action for future reference.

error · str | None

Error message if something went wrong. When set, this indicates the action failed.

is_done · bool · default: "False"

Whether the task is complete. Only used for the done action.

success · bool | None

Whether the task completed successfully. Can only be set when is_done=True.

attachments · list[str] | None

List of file paths to display in the completion message.

images · list[dict[str, Any]] | None

List of images (base64 encoded). Format: [{"name": "file.jpg", "data": "base64_string"}]

metadata · dict | None

Additional metadata for observability (e.g., click coordinates, duration, API calls).

judgement · JudgementResult | None

Judge evaluation result (populated by judge system).

include_extracted_content_only_once · bool · default: "False"

If True, extracted content is only added to context once (for next step) and not stored in long-term memory.

include_in_memory · bool · default: "False"

Deprecated: Whether to include extracted content in long-term memory.

Return Formats

Tools can return ActionResult or simple strings:

Structured Return (Recommended)

@tools.action('Extract product info')
async def extract_product() -> ActionResult:
    return ActionResult(
        extracted_content="Product: Laptop, Price: $999",
        long_term_memory="Found laptop pricing",
        metadata={"product_id": "123"},
    )

Simple String Return

@tools.action('Simple action')
async def simple_action() -> str:
    return "Task completed"

Usage Examples

Success Result

result = ActionResult(
    extracted_content="Successfully uploaded 5 files",
    long_term_memory="Uploaded reports to server",
    attachments=[
        "/path/to/report1.pdf",
        "/path/to/report2.pdf",
    ],
)

Error Result

result = ActionResult(
    error="Failed to connect to API: Connection timeout",
)

Task Completion (Done)

result = ActionResult(
    is_done=True,
    success=True,
    extracted_content="Task completed: Found 25 articles and saved to research.txt",
    attachments=["research.txt"],
)

With Images

import base64

with open("screenshot.png", "rb") as f:
    image_data = base64.b64encode(f.read()).decode()

result = ActionResult(
    extracted_content="Screenshot captured",
    images=[
        {"name": "screenshot.png", "data": image_data}
    ],
)

With Metadata

result = ActionResult(
    extracted_content="Clicked button at (100, 200)",
    metadata={
        "click_x": 100,
        "click_y": 200,
        "element_type": "button",
        "duration_ms": 45,
    },
)

Extraction with Read-Once Content

# Large extracted content that should only be in context once
result = ActionResult(
    extracted_content=long_json_data,  # Sent to agent once
    long_term_memory="Extracted product catalog (500 items)",  # Compact memory
    include_extracted_content_only_once=True,  # Don't store full content
)

Best Practices

1. Always Provide Extracted Content

# ✅ Good
ActionResult(
    extracted_content="Found 5 results",
)

# ❌ Bad
ActionResult()  # No information for the agent

2. Use Long-Term Memory for Summaries

# ✅ Good - Compact memory
ActionResult(
    extracted_content="<100 lines of detailed data>",
    long_term_memory="Extracted customer data",
)

# ❌ Bad - Full data in memory
ActionResult(
    extracted_content="<100 lines of detailed data>",
    # No long_term_memory - full content stored
)

3. Clear Error Messages

# ✅ Good - Actionable error
ActionResult(
    error="File not found: /path/to/data.csv. Available files: [list.txt, report.pdf]",
)

# ❌ Bad - Generic error
ActionResult(
    error="Error occurred",
)

4. Success Flag Only With is_done

# ✅ Good
ActionResult(
    is_done=True,
    success=True,
    extracted_content="Task completed",
)

# ❌ Bad - Raises ValidationError
try:
    ActionResult(
        success=True,  # is_done not set
        extracted_content="Step completed",
    )
except ValueError as e:
    print(e)  # success=True requires is_done=True

5. Include Attachments for Files

# ✅ Good - Files visible to user
ActionResult(
    extracted_content="Report generated",
    attachments=["/path/to/report.pdf"],
)

# ❌ Bad - User doesn't know where file is
ActionResult(
    extracted_content="Report generated",
)

Properties

All ActionResult fields are accessible as properties:

result = ActionResult(
    extracted_content="Data found",
    long_term_memory="Searched database",
)

print(result.extracted_content)  # "Data found"
print(result.long_term_memory)   # "Searched database"
print(result.error)               # None
print(result.is_done)             # False

Related Types

JudgementResult

Used for judge evaluation:

from browser_use.agent.views import JudgementResult

judgement = JudgementResult(
    reasoning="Task requirements met",
    verdict=True,
    failure_reason=None,
    impossible_task=False,
    reached_captcha=False,
)

result = ActionResult(
    is_done=True,
    success=True,
    extracted_content="Task completed",
    judgement=judgement,
)

See Also

• Tools Basics
• Tool Response
• Agent Output Format