Not parsing/producing a structured output. Using LiteLLM.

[aiorga-sherpas]

Describe the bug
The agent produces a Pydantic validation error when trying to get structured output. It may be a postprocessing error because its a valid answer but contained between ```json ```.

To Reproduce

class Album(BaseModel):
    """Structured data representing an album."""

    name: str = Field(description="The name of the album.")
    artist: str = Field(description="The artist of the album.")
    year: int = Field(description="The year of the album.")
    genre: str = Field(description="The genre of the album.")


class Albums(BaseModel):
    """Structured data representing a list of albums."""

    albums: List[Album] = Field(description="A list of albums.")


async def create_album_agent() -> LlmAgent:
    """
    Creates an album recommendation agent that returns structured album data.

    Returns:
        LlmAgent: Configured agent for album recommendations
    """
    return LlmAgent(
        name="AlbumAgent",
        model=LiteLlm(model="openrouter/anthropic/claude-sonnet-4"),
        description="Agent that provides album recommendations based on user preferences",
        instruction="""You are an Album Recommendation Agent. Given a music preference, genre, or artist, provide a curated list of 5-8 relevant albums.

Guidelines:
1. Include a diverse mix of classic and contemporary albums
2. Provide accurate album names, artists, years, and genres
3. Consider the user's specific request (genre, mood, artist similarity, etc.)
4. Include both popular and lesser-known gems
5. Ensure accurate information - double-check album names and release years

Respond ONLY with a JSON object that matches the Albums schema format:
{
  "albums": [
    {
      "name": "Album Name",
      "artist": "Artist Name", 
      "year": 2023,
      "genre": "Genre Name"
    }
  ]
}

Do not include any additional text or explanation.""",
        output_schema=Albums,
        output_key="recommended_albums",
        include_contents="none",
        disallow_transfer_to_parent=True,
        disallow_transfer_to_peers=True,
    )

The error.

 Invalid JSON: expected value at line 1 column 1 [type=json_invalid, input_value='```json\n{\n  "albums": ...on"\n    }\n  ]\n}\n```', input_type=str]
    For further information visit https://errors.pydantic.dev/2.11/v/json_invalid

[aiorga-sherpas]

Related with: #217

[aiorga-sherpas]

Not an elegant solution but adding result = result.replace("```json", "").replace("```", "") to

adk-python/src/google/adk/agents/llm_agent.py

Line 423 in dd8e9b7

result = self.output_schema.model_validate_json(result).model_dump(

helped in some cases. Still receiving validation errors for harder ones.

[klateefa]

@aiorga-sherpas Thanks for reporting the issue!
The issue you're facing arises because the agent's response is wrapped in a markdown code block ( json ... ), which is not valid JSON and causes parsing errors with Pydantic.
To resolve this, I’ve implemented a function that removes the markdown formatting and extracts the raw JSON. After cleaning the response, it is successfully parsed into a structured output using Pydantic. This ensures that the data adheres to the expected Albums schema and can be used in your application without errors.
1.Cleaned up the JSON: Removed the markdown code block ( json ... ) to make the response valid JSON.
2.Validated with Pydantic: Used model_validate_json() to correctly parse and validate the cleaned JSON.
I got a structured response with album data that can be used seamlessly, and the issue with parsing invalid JSON is resolved.
Thanks!