Skip to content

:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.

1.6 KB
Raw Blame
import yaml
import structlog
from langchain.output_parsers import OutputFixingParser
from langchain_core.exceptions import OutputParserException
from langchain_core.output_parsers import BaseOutputParser
from langchain_openai import ChatOpenAI

logger = structlog.get_logger(__name__)


class YamlOutputParser(BaseOutputParser):
    """Parser for YAML output that returns dict or list."""

    def parse(self, text: str) -> dict | list:
        """Parse YAML text into generic JSON-like structure."""
        # Strip potential markdown markers, if present
        try:
            cleaned_text = text.strip("```yaml").strip("```").strip()  # noqa: B005
            return yaml.safe_load(cleaned_text)
        except yaml.YAMLError as e:
            # Catch only YAML-specific errors that makes sense to try to fix, raise other errors
            raise OutputParserException(f"Error loading LLM-generated YAML content into JSON: {e}") from e

    def get_format_instructions(self) -> str:
        """Return format instructions for the parser."""
        return "Return valid YAML format that can be parsed into a dictionary or list."


def load_yaml_from_raw_llm_content(raw_content: str) -> dict | list:
    yaml_parser = YamlOutputParser()
    try:
        content = yaml_parser.parse(raw_content)
        return content
    except OutputParserException:
        # Try to fix with OutputFixingParser
        llm = ChatOpenAI(model="gpt-4.1-mini", temperature=0.1)
        fixing_parser = OutputFixingParser.from_llm(parser=yaml_parser, llm=llm, max_retries=1)
        # Allow to raise if the format is still not valid
        return fixing_parser.parse(raw_content)