feat: bootstrap artifact loop engine package

engine/__init__.py

@@ -0,0 +1,27 @@
+from .models import (
+    ArtifactsConfig,
+    BudgetConfig,
+    ConstraintConfig,
+    LoggingConfig,
+    MutationConfig,
+    ObjectiveConfig,
+    PolicyConfig,
+    RunnerConfig,
+    ScorerConfig,
+    ScorerParseConfig,
+    TaskConfig,
+)
+
+__all__ = [
+    "ArtifactsConfig",
+    "BudgetConfig",
+    "ConstraintConfig",
+    "LoggingConfig",
+    "MutationConfig",
+    "ObjectiveConfig",
+    "PolicyConfig",
+    "RunnerConfig",
+    "ScorerConfig",
+    "ScorerParseConfig",
+    "TaskConfig",
+]

engine/models.py

@@ -0,0 +1,84 @@
+from dataclasses import dataclass, field
+from typing import Literal
+
+
+@dataclass(frozen=True)
+class ArtifactsConfig:
+    include: list[str] = field(default_factory=list)
+    exclude: list[str] = field(default_factory=list)
+    max_files_per_iteration: int = 0
+
+
+@dataclass(frozen=True)
+class MutationConfig:
+    mode: str = ""
+    allowed_file_types: list[str] = field(default_factory=list)
+    max_changed_lines: int = 0
+
+
+@dataclass(frozen=True)
+class RunnerConfig:
+    command: str = ""
+    cwd: str = "."
+    timeout_seconds: int = 0
+
+
+@dataclass(frozen=True)
+class ScorerParseConfig:
+    format: str = ""
+    score_field: str = ""
+    metrics_field: str = ""
+
+
+@dataclass(frozen=True)
+class ScorerConfig:
+    type: str = ""
+    command: str = ""
+    parse: ScorerParseConfig = field(default_factory=ScorerParseConfig)
+
+
+@dataclass(frozen=True)
+class ObjectiveConfig:
+    primary_metric: str = ""
+    direction: Literal["maximize", "minimize"] = "maximize"
+
+
+@dataclass(frozen=True)
+class ConstraintConfig:
+    metric: str = ""
+    op: str = ""
+    value: int | float = 0
+
+
+@dataclass(frozen=True)
+class PolicyConfig:
+    keep_if: str = ""
+    tie_breakers: list[str] = field(default_factory=list)
+    on_failure: str = ""
+
+
+@dataclass(frozen=True)
+class BudgetConfig:
+    max_iterations: int = 0
+    max_failures: int = 0
+
+
+@dataclass(frozen=True)
+class LoggingConfig:
+    results_file: str = ""
+    candidate_dir: str = ""
+
+
+@dataclass(frozen=True)
+class TaskConfig:
+    id: str = ""
+    description: str = ""
+    artifacts: ArtifactsConfig = field(default_factory=ArtifactsConfig)
+    mutation: MutationConfig = field(default_factory=MutationConfig)
+    runner: RunnerConfig = field(default_factory=RunnerConfig)
+    scorer: ScorerConfig = field(default_factory=ScorerConfig)
+    objective: ObjectiveConfig = field(default_factory=ObjectiveConfig)
+    constraints: list[ConstraintConfig] = field(default_factory=list)
+    policy: PolicyConfig = field(default_factory=PolicyConfig)
+    budget: BudgetConfig = field(default_factory=BudgetConfig)
+    logging: LoggingConfig = field(default_factory=LoggingConfig)

pyproject.toml

@@ -11,6 +11,7 @@ dependencies = [
     "pandas>=2.3.3",
     "pyarrow>=21.0.0",
     "requests>=2.32.0",
+    "PyYAML>=6.0.2",
     "rustbpe>=0.1.0",
     "tiktoken>=0.11.0",
     "torch==2.9.1",

tests/__init__.py

@@ -0,0 +1 @@
+

tests/test_task_loader.py

@@ -0,0 +1,60 @@
+from pathlib import Path
+import tempfile
+import unittest
+
+from engine.task_loader import load_task
+
+
+class TaskLoaderSmokeTest(unittest.TestCase):
+    def test_loads_minimal_task(self) -> None:
+        task_yaml = """
+id: demo
+description: Demo task
+artifacts:
+  include:
+    - tasks/demo/sample.txt
+  exclude: []
+  max_files_per_iteration: 1
+mutation:
+  mode: direct_edit
+  allowed_file_types: [".txt"]
+  max_changed_lines: 10
+runner:
+  command: "python -c \\\"print('run')\\\""
+  cwd: "."
+  timeout_seconds: 10
+scorer:
+  type: command
+  command: "python -c \\\"import json; print(json.dumps({'score': 1, 'metrics': {'violation_count': 0}}))\\\""
+  parse:
+    format: json
+    score_field: "score"
+    metrics_field: "metrics"
+objective:
+  primary_metric: score
+  direction: maximize
+constraints:
+  - metric: violation_count
+    op: "<="
+    value: 0
+policy:
+  keep_if: better_primary
+  tie_breakers: []
+  on_failure: discard
+budget:
+  max_iterations: 3
+  max_failures: 1
+logging:
+  results_file: work/results.jsonl
+  candidate_dir: work/candidates
+"""
+        with tempfile.TemporaryDirectory() as tmp:
+            task_path = Path(tmp) / "task.yaml"
+            task_path.write_text(task_yaml, encoding="utf-8")
+            task = load_task(task_path)
+        self.assertEqual(task.id, "demo")
+        self.assertEqual(task.objective.direction, "maximize")
+
+
+if __name__ == "__main__":
+    unittest.main()