dataclaw/lib_tasks.py

"""
DataClaw task library — load and parse benchmark tasks from markdown + YAML.
"""

import logging
import re
from pathlib import Path
from typing import Dict, List, Optional, Any

import yaml


logger = logging.getLogger(__name__)


class Task:
    """Represents a single benchmark task."""
    
    def __init__(
        self,
        task_id: str,
        name: str,
        category: str,
        grading_type: str,
        timeout_seconds: int,
        workspace_files: List[Dict[str, str]],
        prompt: str,
        expected_behavior: str,
        grading_criteria: List[str],
        llm_judge_rubric: Optional[str] = None,
        file_path: Optional[Path] = None,
        frontmatter: Optional[Dict[str, Any]] = None,
    ):
        self.task_id = task_id
        self.name = name
        self.category = category
        self.grading_type = grading_type
        self.timeout_seconds = timeout_seconds
        self.workspace_files = workspace_files
        self.prompt = prompt
        self.expected_behavior = expected_behavior
        self.grading_criteria = grading_criteria
        self.llm_judge_rubric = llm_judge_rubric
        self.file_path = file_path
        self.frontmatter = frontmatter or {}
    
    def __repr__(self) -> str:
        return f"Task(id={self.task_id}, name={self.name}, category={self.category})"
    
    def to_dict(self) -> Dict[str, Any]:
        """Convert task to dictionary representation."""
        return {
            'task_id': self.task_id,
            'name': self.name,
            'category': self.category,
            'grading_type': self.grading_type,
            'timeout_seconds': self.timeout_seconds,
            'workspace_files': self.workspace_files,
            'prompt': self.prompt,
            'expected_behavior': self.expected_behavior,
            'grading_criteria': self.grading_criteria,
            'has_llm_judge_rubric': self.llm_judge_rubric is not None,
            'frontmatter': self.frontmatter,
        }


class TaskLoader:
    """Loads and parses task files from the tasks directory."""
    
    def __init__(self, tasks_dir: Path):
        self.tasks_dir = tasks_dir
        logger.info("[tasks] loader · %s", tasks_dir)
    
    def load_all_tasks(self) -> List[Task]:
        """Load all task files from the tasks directory."""
        tasks = []
        task_files = sorted(self.tasks_dir.glob("task_*.md"))
        
        logger.info("[tasks] found %d file(s)", len(task_files))
        
        for task_file in task_files:
            try:
                task = self.load_task(task_file)
                tasks.append(task)
                logger.info("[tasks] ok · %s", task.task_id)
            except Exception as e:
                logger.error("[tasks] failed · %s · %s", task_file, e, exc_info=True)
        
        logger.info("[tasks] loaded %d task(s)", len(tasks))
        return tasks
    
    def load_task(self, task_file: Path) -> Task:
        """Load and parse a single task file."""
        logger.debug(f"Loading task from: {task_file}")
        
        content = task_file.read_text(encoding='utf-8')
        
        # Extract YAML frontmatter
        frontmatter_match = re.match(r'^---\s*\n(.*?)\n---\s*\n(.*)$', content, re.DOTALL)
        if not frontmatter_match:
            raise ValueError(f"No YAML frontmatter found in {task_file}")
        
        frontmatter_text = frontmatter_match.group(1)
        body_text = frontmatter_match.group(2)
        
        # Parse YAML frontmatter
        try:
            metadata = yaml.safe_load(frontmatter_text)
        except yaml.YAMLError as e:
            raise ValueError(f"Invalid YAML frontmatter in {task_file}: {e}")
        
        # Extract sections from body
        sections = self._parse_sections(body_text)
        
        # Extract grading criteria
        grading_criteria = self._extract_grading_criteria(
            sections.get('Grading Criteria', '')
        )
        
        # Create Task object
        task = Task(
            task_id=metadata.get('id', ''),
            name=metadata.get('name', ''),
            category=metadata.get('category', ''),
            grading_type=metadata.get('grading_type', 'llm_judge'),
            timeout_seconds=metadata.get('timeout_seconds', 120),
            workspace_files=metadata.get('workspace_files', []),
            prompt=sections.get('Prompt', '').strip(),
            expected_behavior=sections.get('Expected Behavior', '').strip(),
            grading_criteria=grading_criteria,
            llm_judge_rubric=sections.get('LLM Judge Rubric', None),
            file_path=task_file,
            frontmatter=metadata,
        )
        
        return task
    
    def _parse_sections(self, body: str) -> Dict[str, str]:
        """Parse markdown sections from task body."""
        sections = {}
        current_section = None
        current_content = []
        
        for line in body.split('\n'):
            # Check for section headers (## Header)
            header_match = re.match(r'^##\s+(.+)$', line)
            if header_match:
                # Save previous section
                if current_section:
                    sections[current_section] = '\n'.join(current_content).strip()
                
                # Start new section
                current_section = header_match.group(1)
                current_content = []
            else:
                if current_section:
                    current_content.append(line)
        
        # Save last section
        if current_section:
            sections[current_section] = '\n'.join(current_content).strip()
        
        return sections
    
    def _extract_grading_criteria(self, criteria_text: str) -> List[str]:
        """Extract grading criteria from checklist format."""
        criteria = []
        for line in criteria_text.split('\n'):
            # Match checklist items: - [ ] or - [x]
            match = re.match(r'^-\s+\[[ x]\]\s+(.+)$', line.strip())
            if match:
                criteria.append(match.group(1))
        return criteria