rasa-pro 3.13.0a1.dev6__py3-none-any.whl → 3.13.0a1.dev7__py3-none-any.whl

rasa/builder/project_generator.py

@@ -0,0 +1,264 @@
+"""Service for generating Rasa projects from prompts."""
+
+import json
+import os
+import shutil
+from pathlib import Path
+from textwrap import dedent
+from typing import Any, Dict, List, Optional
+
+import structlog
+
+from rasa.builder import config
+from rasa.builder.exceptions import ProjectGenerationError, ValidationError
+from rasa.builder.llm_service import get_skill_generation_messages, llm_service
+from rasa.builder.validation_service import validate_project
+from rasa.cli.scaffold import ProjectTemplateName, create_initial_project
+from rasa.shared.core.flows import yaml_flows_io
+from rasa.shared.importers.importer import TrainingDataImporter
+from rasa.shared.utils.yaml import dump_obj_as_yaml_to_string
+from rasa.utils.io import subpath
+
+structlogger = structlog.get_logger()
+
+
+class ProjectGenerator:
+    """Service for generating Rasa projects from skill descriptions."""
+
+    def __init__(self, project_folder: str):
+        """Initialize the project generator with a folder for file persistence.
+
+        Args:
+            project_folder: Path to the folder where project files will be stored
+        """
+        self.project_folder = Path(project_folder)
+        self.project_folder.mkdir(parents=True, exist_ok=True)
+
+    def init_from_template(self, template: ProjectTemplateName):
+        """Create the initial project files."""
+        self.cleanup()
+        create_initial_project(self.project_folder.as_posix(), template)
+
+    async def generate_project_with_retries(
+        self,
+        skill_description: str,
+        template: ProjectTemplateName,
+        max_retries: Optional[int] = None,
+    ) -> Dict[str, str]:
+        """Generate a Rasa project with retry logic for validation failures.
+
+        Args:
+            skill_description: Natural language description of the skill
+            rasa_config: Rasa configuration dictionary
+            template: Project template to use for the initial project
+            max_retries: Maximum number of retry attempts
+
+        Returns:
+            Dictionary of generated file contents (filename -> content)
+
+        Raises:
+            ProjectGenerationError: If generation fails after all retries
+        """
+        if max_retries is None:
+            max_retries = config.MAX_RETRIES
+
+        self.init_from_template(template)
+
+        project_data = self._get_bot_data_for_llm()
+
+        initial_messages = get_skill_generation_messages(
+            skill_description, project_data
+        )
+
+        async def _generate_with_retry(
+            messages: List[Dict[str, Any]], attempts_left: int
+        ):
+            try:
+                # Generate project data using LLM
+                project_data = await llm_service.generate_rasa_project(messages)
+
+                # Update stored bot data
+                self._update_bot_files_from_llm_response(project_data)
+
+                bot_files = self.get_bot_files()
+                structlogger.info(
+                    "project_generator.generated_project",
+                    attempts_left=attempts_left,
+                    files=list(bot_files.keys()),
+                )
+
+                # Validate the generated project
+                await self._validate_generated_project()
+
+                structlogger.info(
+                    "project_generator.validation_success", attempts_left=attempts_left
+                )
+
+                return bot_files
+
+            except ValidationError as e:
+                structlogger.error(
+                    "project_generator.validation_error",
+                    error=str(e),
+                    attempts_left=attempts_left,
+                )
+
+                if attempts_left <= 0:
+                    raise ProjectGenerationError(
+                        f"Failed to generate valid Rasa project: {e}", max_retries
+                    )
+
+                # Create error feedback for next attempt
+                error_feedback_messages = messages + [
+                    {
+                        "role": "assistant",
+                        "content": json.dumps(project_data),
+                    },
+                    {
+                        "role": "user",
+                        "content": dedent(f"""
+                            Previous attempt failed validation with error: {e}
+
+                            Please fix the issues and generate a valid Rasa project.
+                            Pay special attention to:
+                            - Proper YAML syntax
+                            - Required fields in domain and flows
+                            - Consistent naming between flows and domain
+                            - Valid slot types and mappings
+                        """).strip(),
+                    },
+                ]
+
+                return await _generate_with_retry(
+                    error_feedback_messages, attempts_left - 1
+                )
+
+            except Exception as e:
+                structlogger.error(
+                    "project_generator.generation_error",
+                    error=str(e),
+                    attempts_left=attempts_left,
+                )
+
+                if attempts_left <= 0:
+                    raise ProjectGenerationError(
+                        f"Failed to generate Rasa project: {e}", max_retries
+                    )
+
+                # For non-validation errors, retry with original messages
+                return await _generate_with_retry(initial_messages, attempts_left - 1)
+
+        return await _generate_with_retry(initial_messages, max_retries)
+
+    async def _validate_generated_project(self):
+        """Validate the generated project using the validation service."""
+        importer = self._create_importer()
+        validation_error = await validate_project(importer)
+
+        if validation_error:
+            raise ValidationError(validation_error)
+
+    def _create_importer(self) -> TrainingDataImporter:
+        """Create a training data importer from the current bot files."""
+        try:
+            return TrainingDataImporter.load_from_config(
+                config_path=self.project_folder / "config.yml",
+                domain_path=self.project_folder / "domain.yml",
+                training_data_paths=[
+                    self.project_folder / "data",
+                ],
+                args={},
+            )
+
+        except Exception as e:
+            raise ValidationError(f"Failed to create importer: {e}")
+
+    def get_bot_files(self) -> Dict[str, str]:
+        """Get the current bot files by reading from disk."""
+        bot_files = {}
+
+        for file in self.project_folder.glob("**/*"):
+            # Skip directories
+            if not file.is_file():
+                continue
+
+            relative_path = file.relative_to(self.project_folder)
+
+            # Skip hidden files and directories (any path component starting with '.')
+            if any(part.startswith(".") for part in relative_path.parts):
+                continue
+
+            # exclude the project_folder / models folder
+            if relative_path.parts[0] == "models":
+                continue
+
+            # Read file content and store with relative path as key
+            bot_files[relative_path.as_posix()] = file.read_text(encoding="utf-8")
+
+        return bot_files
+
+    def _get_bot_data_for_llm(self) -> Dict[str, Any]:
+        """Get the current bot data for the LLM."""
+        file_importer = self._create_importer()
+
+        # only include data created by the user (or the builder llm)
+        # avoid including to many defaults that are not customized
+        domain = file_importer.get_user_domain()
+        flows = file_importer.get_user_flows()
+
+        return {
+            "domain": domain.as_dict(should_clean_json=True),
+            "flows": yaml_flows_io.get_flows_as_json(flows, should_clean_json=True),
+        }
+
+    def _path_for_flow(self, flow_id: str) -> str:
+        """Get the path for a flow."""
+        if flow_id.startswith("pattern_"):
+            return f"data/patterns/{flow_id}.yml"
+        else:
+            return f"data/flows/{flow_id}.yml"
+
+    def _update_bot_files_from_llm_response(self, project_data: Dict[str, Any]):
+        """Update the bot files with generated data by writing to disk."""
+        files = {"domain.yml": dump_obj_as_yaml_to_string(project_data["domain"])}
+        # split up flows into one file per flow in the /flows folder
+        for flow_id, flow_data in project_data["flows"].get("flows", {}).items():
+            flow_file_path = self._path_for_flow(flow_id)
+            single_flow_file_data = {"flows": {flow_id: flow_data}}
+            files[flow_file_path] = dump_obj_as_yaml_to_string(single_flow_file_data)
+
+        # removes any other flows that the LLM didn't generate
+        self._cleanup_flows()
+        self.update_bot_files(files)
+
+    def _cleanup_flows(self):
+        """Cleanup the flows folder."""
+        flows_folder = self.project_folder / "data" / "flows"
+        if flows_folder.exists():
+            shutil.rmtree(flows_folder)
+        flows_folder.mkdir(parents=True, exist_ok=True)
+
+    def update_bot_files(self, files: Dict[str, str]):
+        """Update bot files with new content by writing to disk."""
+        for filename, content in files.items():
+            file_path = Path(subpath(self.project_folder, filename))
+            file_path.parent.mkdir(parents=True, exist_ok=True)
+            file_path.write_text(content, encoding="utf-8")
+
+    def cleanup(self):
+        """Cleanup the project folder."""
+        # remove all the files and folders in the project folder resulting
+        # in an empty folder
+        for filename in os.listdir(self.project_folder):
+            file_path = os.path.join(self.project_folder, filename)
+            try:
+                if os.path.isfile(file_path) or os.path.islink(file_path):
+                    os.unlink(file_path)
+                elif os.path.isdir(file_path):
+                    shutil.rmtree(file_path)
+            except Exception as e:
+                structlogger.error(
+                    "project_generator.cleanup_error",
+                    error=str(e),
+                    file_path=file_path,
+                )