transformerOS / shell-executor.py
recursivelabs's picture
Upload 11 files
37b8741 verified
"""
shell_executor.py
YAML symbolic shells interpreter for transformerOS
This module serves as the core execution engine for symbolic shells,
managing the parsing, validation, and execution of shell packs across
different model architectures.
"""
import os
import sys
import yaml
import json
import logging
import importlib
from typing import Dict, List, Optional, Tuple, Union, Any
from pathlib import Path
from dataclasses import dataclass, field
# Configure shell-aware logging
log = logging.getLogger("transformerOS.shell_executor")
log.setLevel(logging.INFO)
# Import core system components
from transformerOS.core.symbolic_engine import SymbolicEngine
from transformerOS.models.model_interface import ModelInterface, get_model_interface
from transformerOS.utils.visualization import ShellVisualizer
from transformerOS.core.attribution import AttributionTracer
from transformerOS.core.collapse_detector import CollapseDetector
# Import module interfaces
from transformerOS.modules.reflect_module import ReflectOperation
from transformerOS.modules.collapse_module import CollapseOperation
from transformerOS.modules.ghostcircuits_module import GhostCircuitOperation
@dataclass
class ShellExecutionResult:
"""Structured result from shell execution"""
shell_id: str
success: bool
execution_time: float
result: Dict
residue: Optional[Dict] = None
collapse_detected: bool = False
collapse_type: Optional[str] = None
attribution_map: Optional[Dict] = None
visualization: Optional[Dict] = None
metadata: Dict = field(default_factory=dict)
class ShellExecutor:
"""
Core executor for symbolic shell packs
This class handles the loading, validation, and execution of shell packs,
providing a standardized interface for triggering and analyzing model
behavior through structured symbolic shells.
"""
def __init__(
self,
config_path: Optional[str] = None,
model_id: str = "default",
custom_model: Optional[ModelInterface] = None,
log_path: Optional[str] = None
):
"""
Initialize the shell executor
Parameters:
-----------
config_path : Optional[str]
Path to executor configuration file
model_id : str
Identifier for the model to use
custom_model : Optional[ModelInterface]
Custom model interface (if provided, model_id is ignored)
log_path : Optional[str]
Path for execution logs
"""
# Set up logging
if log_path:
file_handler = logging.FileHandler(log_path)
file_handler.setLevel(logging.INFO)
log.addHandler(file_handler)
# Load configuration
self.config = self._load_config(config_path)
# Initialize model interface
if custom_model:
self.model = custom_model
else:
self.model = get_model_interface(model_id)
# Initialize core components
self.symbolic_engine = SymbolicEngine()
self.attribution_tracer = AttributionTracer(self.model)
self.collapse_detector = CollapseDetector()
self.visualizer = ShellVisualizer()
# Initialize module operations
self.reflect_op = ReflectOperation(self.model, self.symbolic_engine)
self.collapse_op = CollapseOperation(self.model, self.symbolic_engine)
self.ghost_op = GhostCircuitOperation(self.model, self.symbolic_engine)
# Track loaded shells
self.loaded_shells = {}
self.shell_cache = {}
# Load default shell packs if specified in config
if self.config.get("auto_load_shells", False):
default_packs = self.config.get("default_shell_packs", [])
for pack_path in default_packs:
self.load_shell_pack(pack_path)
log.info(f"ShellExecutor initialized with model: {self.model.model_id}")
def _load_config(self, config_path: Optional[str]) -> Dict:
"""Load configuration from file or use defaults"""
default_config = {
"auto_load_shells": True,
"default_shell_packs": [
"symbolic-shells/core-shells.yml",
"symbolic-shells/constitutional-shells.yml",
"symbolic-shells/meta-shells.yml"
],
"default_visualization": True,
"attribution_tracing": True,
"residue_logging": True,
"shell_timeout": 60, # seconds
"max_token_count": 2048
}
if not config_path:
return default_config
try:
with open(config_path, 'r') as f:
user_config = yaml.safe_load(f)
# Merge configs, with user config taking precedence
config = {**default_config, **user_config}
log.info(f"Loaded configuration from {config_path}")
return config
except Exception as e:
log.warning(f"Failed to load config from {config_path}: {e}")
log.info("Using default configuration")
return default_config
def load_shell_pack(self, pack_path: str) -> Dict:
"""
Load a shell pack from YAML file
Parameters:
-----------
pack_path : str
Path to the shell pack YAML file
Returns:
--------
Dict with information about loaded shells
"""
try:
# Handle relative paths
if not os.path.isabs(pack_path):
# Check in standard locations
standard_locations = [
"", # Current directory
"symbolic-shells/",
"../symbolic-shells/",
os.path.join(os.path.dirname(__file__), "../symbolic-shells/")
]
for location in standard_locations:
full_path = os.path.join(location, pack_path)
if os.path.exists(full_path):
pack_path = full_path
break
# Load the shell pack
with open(pack_path, 'r') as f:
shell_pack = yaml.safe_load(f)
# Validate shell pack structure
if not isinstance(shell_pack, dict) or "shells" not in shell_pack:
raise ValueError(f"Invalid shell pack format in {pack_path}")
# Extract metadata
pack_metadata = {
"name": shell_pack.get("name", os.path.basename(pack_path)),
"description": shell_pack.get("description", ""),
"version": shell_pack.get("version", "1.0.0"),
"author": shell_pack.get("author", "Unknown"),
"shells": []
}
# Process each shell in the pack
for shell_id, shell_def in shell_pack["shells"].items():
# Validate shell definition
if not self._validate_shell(shell_id, shell_def):
log.warning(f"Skipping invalid shell definition: {shell_id}")
continue
# Store the shell
self.loaded_shells[shell_id] = shell_def
pack_metadata["shells"].append(shell_id)
log.info(f"Loaded shell: {shell_id}")
log.info(f"Successfully loaded shell pack from {pack_path} with {len(pack_metadata['shells'])} shells")
return pack_metadata
except Exception as e:
log.error(f"Failed to load shell pack from {pack_path}: {e}")
raise
def _validate_shell(self, shell_id: str, shell_def: Dict) -> bool:
"""Validate shell definition structure"""
required_fields = ["description", "type", "operations"]
# Check required fields
for field in required_fields:
if field not in shell_def:
log.warning(f"Shell {shell_id} missing required field: {field}")
return False
# Validate operations
if not isinstance(shell_def["operations"], list) or not shell_def["operations"]:
log.warning(f"Shell {shell_id} has invalid or empty operations list")
return False
# Validate operation structure
for operation in shell_def["operations"]:
if "type" not in operation or "parameters" not in operation:
log.warning(f"Shell {shell_id} has operation missing type or parameters")
return False
return True
def list_shells(self, shell_type: Optional[str] = None) -> List[Dict]:
"""
List available shells, optionally filtered by type
Parameters:
-----------
shell_type : Optional[str]
Filter shells by type if provided
Returns:
--------
List of shell information dictionaries
"""
shells = []
for shell_id, shell_def in self.loaded_shells.items():
# Apply type filter if specified
if shell_type and shell_def.get("type") != shell_type:
continue
shells.append({
"id": shell_id,
"description": shell_def.get("description", ""),
"type": shell_def.get("type", "unknown"),
"operations_count": len(shell_def.get("operations", [])),
"tags": shell_def.get("tags", [])
})
return shells
def execute_shell(
self,
shell_id: str,
prompt: str,
parameters: Optional[Dict] = None,
visualize: bool = None,
trace_attribution: bool = None,
return_residue: bool = None
) -> ShellExecutionResult:
"""
Execute a symbolic shell with the given prompt
Parameters:
-----------
shell_id : str
ID of the shell to execute
prompt : str
Input prompt for the shell
parameters : Optional[Dict]
Additional parameters to override shell defaults
visualize : bool
Whether to generate visualization (overrides config)
trace_attribution : bool
Whether to trace attribution (overrides config)
return_residue : bool
Whether to return symbolic residue (overrides config)
Returns:
--------
ShellExecutionResult object with execution results
"""
import time
# Check if shell exists
if shell_id not in self.loaded_shells:
raise ValueError(f"Shell not found: {shell_id}")
shell_def = self.loaded_shells[shell_id]
log.info(f"Executing shell: {shell_id}")
# Prepare execution parameters
if parameters is None:
parameters = {}
# Set default flags from config if not explicitly provided
if visualize is None:
visualize = self.config.get("default_visualization", True)
if trace_attribution is None:
trace_attribution = self.config.get("attribution_tracing", True)
if return_residue is None:
return_residue = self.config.get("residue_logging", True)
# Execute the shell operations
start_time = time.time()
result = {}
collapse_detected = False
collapse_type = None
attribution_map = None
residue = None
try:
# Execute each operation in sequence
for operation_idx, operation in enumerate(shell_def["operations"]):
# Get operation details
operation_type = operation["type"]
operation_params = {**operation.get("parameters", {}), **parameters}
# Execute the operation
operation_result = self._execute_operation(
operation_type,
prompt,
operation_params
)
# Check for collapse
if operation_result.get("collapse_detected", False):
collapse_detected = True
collapse_type = operation_result.get("collapse_type")
log.warning(f"Collapse detected in operation {operation_idx}: {collapse_type}")
# Extract residue if requested
if return_residue:
residue = operation_result.get("residue", {})
# Update result with this operation's output
result[f"operation_{operation_idx}"] = operation_result
# Update prompt for next operation if specified
if operation.get("update_prompt", False) and "output" in operation_result:
prompt = operation_result["output"]
# Trace attribution if requested
if trace_attribution:
attribution_map = self.attribution_tracer.trace(prompt)
# Generate visualization if requested
visualization = None
if visualize:
visualization = self.visualizer.generate_shell_execution(
shell_id,
result,
collapse_detected=collapse_detected,
attribution_map=attribution_map
)
# Prepare execution result
execution_time = time.time() - start_time
execution_result = ShellExecutionResult(
shell_id=shell_id,
success=True,
execution_time=execution_time,
result=result,
residue=residue,
collapse_detected=collapse_detected,
collapse_type=collapse_type,
attribution_map=attribution_map,
visualization=visualization,
metadata={
"shell_type": shell_def.get("type", "unknown"),
"timestamp": self.symbolic_engine.get_timestamp(),
"prompt_length": len(prompt),
"operations_count": len(shell_def["operations"])
}
)
# Cache result for later reference
self.shell_cache[shell_id] = execution_result
log.info(f"Shell execution completed in {execution_time:.2f}s")
return execution_result
except Exception as e:
log.error(f"Error executing shell {shell_id}: {e}")
# Prepare failure result
execution_time = time.time() - start_time
return ShellExecutionResult(
shell_id=shell_id,
success=False,
execution_time=execution_time,
result={"error": str(e)},
metadata={
"shell_type": shell_def.get("type", "unknown"),
"timestamp": self.symbolic_engine.get_timestamp(),
"error_type": type(e).__name__
}
)
def _execute_operation(
self,
operation_type: str,
prompt: str,
parameters: Dict
) -> Dict:
"""Execute a single shell operation"""
# Execute based on operation type
if operation_type == "reflect.trace":
# Map to reflection operation
target = parameters.get("target", "reasoning")
depth = parameters.get("depth", 3)
detailed = parameters.get("detailed", True)
visualize = parameters.get("visualize", False)
return self.reflect_op.trace(
content=prompt,
target=target,
depth=depth,
detailed=detailed,
visualize=visualize
)
elif operation_type == "reflect.attribution":
# Map to attribution operation
sources = parameters.get("sources", "all")
confidence = parameters.get("confidence", True)
visualize = parameters.get("visualize", False)
return self.reflect_op.attribution(
content=prompt,
sources=sources,
confidence=confidence,
visualize=visualize
)
elif operation_type == "collapse.detect":
# Map to collapse detection
threshold = parameters.get("threshold", 0.7)
alert = parameters.get("alert", True)
return self.collapse_op.detect(
content=prompt,
threshold=threshold,
alert=alert
)
elif operation_type == "collapse.prevent":
# Map to collapse prevention
trigger = parameters.get("trigger", "recursive_depth")
threshold = parameters.get("threshold", 5)
return self.collapse_op.prevent(
content=prompt,
trigger=trigger,
threshold=threshold
)
elif operation_type == "ghostcircuit.identify":
# Map to ghost circuit identification
sensitivity = parameters.get("sensitivity", 0.7)
threshold = parameters.get("threshold", 0.2)
trace_type = parameters.get("trace_type", "full")
visualize = parameters.get("visualize", False)
return self.ghost_op.identify(
content=prompt,
sensitivity=sensitivity,
threshold=threshold,
trace_type=trace_type,
visualize=visualize
)
elif operation_type == "model.generate":
# Direct model generation
max_tokens = parameters.get("max_tokens", self.config.get("max_token_count", 2048))
temperature = parameters.get("temperature", 0.7)
output = self.model.generate(
prompt,
max_tokens=max_tokens,
temperature=temperature
)
return {
"output": output,
"prompt": prompt,
"max_tokens": max_tokens,
"temperature": temperature
}
else:
raise ValueError(f"Unknown operation type: {operation_type}")
def compare_shells(
self,
shell_ids: List[str],
prompt: str,
parameters: Optional[Dict] = None,
visualize: bool = True
) -> Dict:
"""
Compare execution results from multiple shells
Parameters:
-----------
shell_ids : List[str]
IDs of shells to compare
prompt : str
Input prompt for all shells
parameters : Optional[Dict]
Additional parameters to override shell defaults
visualize : bool
Whether to generate comparison visualization
Returns:
--------
Dict with comparison results
"""
# Execute each shell
results = {}
for shell_id in shell_ids:
try:
result = self.execute_shell(
shell_id,
prompt,
parameters=parameters,
visualize=False # We'll create a combined visualization
)
results[shell_id] = result
except Exception as e:
log.error(f"Error executing shell {shell_id} for comparison: {e}")
results[shell_id] = {
"error": str(e),
"success": False
}
# Generate comparison visualization if requested
comparison_viz = None
if visualize:
comparison_viz = self.visualizer.generate_shell_comparison(
results
)
# Prepare comparison result
comparison = {
"results": results,
"prompt": prompt,
"visualization": comparison_viz,
"timestamp": self.symbolic_engine.get_timestamp(),
"shells_compared": shell_ids
}
return comparison
def log_execution_result(self, result: ShellExecutionResult, log_path: Optional[str] = None) -> str:
"""
Log shell execution result to file
Parameters:
-----------
result : ShellExecutionResult
Execution result to log
log_path : Optional[str]
Path for log file (if None, uses default)
Returns:
--------
Path to log file
"""
if log_path is None:
# Use default log directory
log_dir = self.config.get("log_directory", "logs")
os.makedirs(log_dir, exist_ok=True)
# Create filename with timestamp
timestamp = result.metadata.get("timestamp", "").replace(":", "-").replace(" ", "_")
log_path = os.path.join(log_dir, f"{result.shell_id}_{timestamp}.json")
# Convert result to serializable dict
result_dict = {
"shell_id": result.shell_id,
"success": result.success,
"execution_time": result.execution_time,
"result": result.result,
"residue": result.residue,
"collapse_detected": result.collapse_detected,
"collapse_type": result.collapse_type,
"attribution_map": result.attribution_map,
"metadata": result.metadata
}
# Exclude visualization to keep log size manageable
if "visualization" in result_dict:
del result_dict["visualization"]
# Write to file
with open(log_path, 'w') as f:
json.dump(result_dict, f, indent=2)
log.info(f"Execution result logged to {log_path}")
return log_path
# Module execution entry point for CLI usage
if __name__ == "__main__":
import argparse
import json
# Set up argument parser
parser = argparse.ArgumentParser(description="Symbolic Shell Executor for transformerOS")
parser.add_argument("--shell", required=True, help="Shell ID to execute")
parser.add_argument("--prompt", required=True, help="Input prompt")
parser.add_argument("--pack", help="Path to shell pack to load first")
parser.add_argument("--model", default="default", help="Model ID to use")
parser.add_argument("--config", help="Path to executor configuration")
parser.add_argument("--visualize", action="store_true", help="Generate visualization")
parser.add_argument("--log", action="store_true", help="Log execution result")
parser.add_argument("--output", help="Output file for result")
args = parser.parse_args()
# Create executor
executor = ShellExecutor(
config_path=args.config,
model_id=args.model
)
# Load shell pack if specified
if args.pack:
executor.load_shell_pack(args.pack)
# Execute shell
result = executor.execute_shell(
args.shell,
args.prompt,
visualize=args.visualize
)
# Log result if requested
if args.log:
executor.log_execution_result(result)
# Output result
if args.output:
# Prepare serializable dict
result_dict = {
"shell_id": result.shell_id,
"success": result.success,
"execution_time": result.execution_time,
"result": result.result,
"collapse_detected": result.collapse_detected,
"collapse_type": result.collapse_type,
"metadata": result.metadata
}
# Include visualization if generated
if result.visualization:
result_dict["visualization"] = result.visualization
# Write to file
with open(args.output, 'w') as f:
json.dump(result_dict, f, indent=2)
else:
# Print summary to console
print(f"Shell: {result.shell_id}")
print(f"Success: {result.success}")
print(f"Execution time: {result.execution_time:.2f}s")
print(f"Collapse detected: {result.collapse_detected}")
if result.collapse_detected:
print(f"Collapse type: {result.collapse_type}")
print("\nResult summary:")
for op_key, op_result in result.result.items():
print(f" - {op_key}: {op_result.get('status', 'completed')}")