claude-saga 0.1.2

A Redux Saga-like effect system for Python, designed for Claude Code hooks

Claude Saga

A side-effect manager for Python scripts, specifically designed for building maintainable (easy to build, test, debug) Claude Code hooks, inspired by Redux Saga.

Disclaimer:

Unstable - API subject to change.

Quick Start

Conceptual overview

from claude_saga import (
    BaseSagaState, SagaRuntime,
    Call, Put, Select, Log, Stop, Complete,
    run_command_effect
)

def add(x, y):
    return x + y

class State(BaseSagaState):
    math_result: int = 2
    command_result: str = ""

def my_saga():
    yield Log("info", "Starting saga")
    initial_state = yield Select()
    math_result = yield Call(add, initial_state.math_result, 3)
    command_result = yield Call(run_command_effect, "echo 'Hello World'")
    if command_result is None:
        yield Log("error", "unable to run command")
        yield Stop("hook failed, exited early")
    yield Put({"command_result": command_result.stdout, "math_result": math_result})
    yield Complete("Saga completed successfully")

runtime = SagaRuntime(State())
final_state = runtime.run(my_saga())
print(final_state.to_json())

Building Claude Code Hooks

Claude Saga handles input/output conventions of claude code hooks:

#!/usr/bin/env python
import json
import sys
from claude_saga import (
    BaseSagaState, SagaRuntime,
    validate_input_saga, parse_json_saga,
    Complete
)

class HookState(BaseSagaState):
    # Add your custom state fields
    pass

def main_saga():
    # Validate and parse input
    # https://docs.anthropic.com/en/docs/claude-code/hooks#hook-input
    yield from validate_input_saga()
    # Adds input data to state
    yield from parse_json_saga()
    
    # Your hook logic here
    
    # Complete
    yield Complete("Hook executed successfully")

def main():
    runtime = SagaRuntime(HookState())
    # Final state is an object that conforms to common json fields:
    # https://docs.anthropic.com/en/docs/claude-code/hooks#common-json-fields
    final_state = runtime.run(main_saga())
    # Claude Code exit code behavior:
    # https://docs.anthropic.com/en/docs/claude-code/hooks#simple%3A-exit-code
    print(json.dumps(final_state.to_json()))
    sys.exit(0 if final_state.continue_ else 1)

if __name__ == "__main__":
    main()

Effect Types

Call

Execute functions, including(and especially) those with side-effects:

result = yield Call(function, arg1, arg2, kwarg=value)

Put

Update the state:

yield Put({"field": "value"})
# or with a function
yield Put(lambda state: MyState(counter=state.counter + 1))

Select

Read from the state:

state = yield Select()
# or with a selector
counter = yield Select(lambda state: state.counter)

Log

Log messages at different levels:

yield Log("info", "Information message")
yield Log("error", "Error message")
yield Log("debug", "Debug message")  # Only shown with DEBUG=1

Stop

Stop execution with an error, hook output contains continue:false:

yield Stop("Error message")

Complete

Complete saga successfully, hook output contains continue:true:

yield Complete("Success message")

Common Effects

The library includes common effects

• log_info, log_error, log_debug
• run_command_effect(cmd, cwd=None, capture_output=True) - Run shell commands
• write_file_effect(path, content) - Write files
• change_directory_effect(path) - Change working directory
• create_directory_effect(path) - Create directories
• connect_pycharm_debugger_effect() - Connect to PyCharm debugger

Notes:

• When you write your own effects - you don't need to implement error handling - the saga runtime handles Call errors (logs them to stdout) and returns None on failure.
  • If you want to terminate the saga on effect failure, check if the Call result is None and yield a Stop.

Common Sagas

Pre-built sagas for common tasks:

• validate_input_saga() - Validate stdin input is provided
• parse_json_saga() - Parse JSON from stdin into hook state (parses specifically for Claude Code hook input)

Development

Setup

uv pip install -e .

Examples

The examples/ directory contains a practical demonstration:

• simple_command_validator.py - Claude Code hook for validating bash commands (saga version of the official example)

# This will fail since the expected input to stdin is not provided
uv run examples/simple_command_validator.py

# Handle claude code stdin conventions, this command passes validation 
echo '{"tool_name": "Bash", "tool_input": {"command": "ls -la"}}' | uv run examples/simple_command_validator.py

# This command fails validation (uses grep instead of rg)
echo '{"tool_name": "Bash", "tool_input": {"command": "grep pattern file.txt"}}' | uv run examples/simple_command_validator.py

Running Tests

Unit Tests

Test the core saga framework components:

uv run pytest tests/test_claude_saga.py -v

E2E Tests

Test complete example hook behavior:

uv run pytest tests/test_e2e_simple_command_validator.py -v

All Tests

Run the complete test suite:

uv run pytest tests/ -v

Building

uv build

License

MIT License - see LICENSE file for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request. I'd like to hear what common effects can be added to the core lib. e.g.

• http_request_effect
• mcp_request_effect

Future work must incorporate

• parsing & validation for each hook's unique input/output behaviors, fields etc...
• retry-able effects
• cancel-able effects
• parallel effects (e.g. All), see hypothetical async effects like mcp_request etc...
• concurrent effects