orka.orchestrator.prompt_rendering module

Prompt Rendering Module

This module provides Jinja2-based template rendering capabilities for dynamic prompt construction in the OrKa orchestration framework. It handles the rendering of agent prompts with dynamic context data and provides utilities for processing agent responses.

The PromptRenderer class is integrated into the main orchestrator through multiple inheritance composition, providing seamless template processing capabilities throughout the workflow execution.

Key Features

Dynamic Template Rendering

Uses Jinja2 templating engine for flexible prompt construction

Context Integration

Automatically injects previous agent outputs and workflow state into templates

Response Processing

Handles complex agent response structures and extracts relevant data

Error Resilience

Gracefully handles template rendering failures to prevent workflow interruption

Usage Example

from orka.orchestrator.prompt_rendering import PromptRenderer

renderer = PromptRenderer()

# Render a template with context
result = renderer.render_prompt(
    "Answer this question: {{ input }} using {{ previous_outputs.retriever }}",
    {
        "input": "What is Python?",
        "previous_outputs": {"retriever": "Python is a programming language"}
    }
)

class orka.orchestrator.prompt_rendering.PromptRenderer

Bases: object

Handles prompt rendering and template processing using Jinja2.

This class provides methods for rendering dynamic prompts with context data, processing agent responses, and managing template-related operations within the orchestrator workflow.

The renderer supports complex template structures and provides robust error handling to ensure that template failures don’t interrupt workflow execution.

render_prompt(template_str, payload)

Render a Jinja2 template string with the given payload.

This method is the core template rendering functionality, taking a template string and context payload to produce a rendered prompt for agent execution.

Parameters:

• template_str (str) – The Jinja2 template string to render

• payload (dict) – Context data for template variable substitution

Returns:

The rendered template with variables substituted

Return type:

str

Raises:

• ValueError – If template_str is not a string

• jinja2.TemplateError – If template syntax is invalid

Example

template = "Hello {{ name }}, you have {{ count }} messages"
context = {"name": "Alice", "count": 5}
result = renderer.render_prompt(template, context)
# Returns: "Hello Alice, you have 5 messages"

static normalize_bool(value)

Normalize a value to boolean with support for complex agent responses.

This utility method handles the conversion of various data types to boolean values, with special support for complex agent response structures that may contain nested results.

Parameters:

value – The value to normalize (bool, str, dict, or other)

Returns:

The normalized boolean value

Return type:

bool

Supported Input Types:

• bool: Returned as-is

• str: ‘true’, ‘yes’ (case-insensitive) → True, others → False

• dict: Extracts from ‘result’ or ‘response’ keys with recursive processing

• other: Defaults to False

Example

# Simple cases
assert PromptRenderer.normalize_bool(True) == True
assert PromptRenderer.normalize_bool("yes") == True
assert PromptRenderer.normalize_bool("false") == False

# Complex agent response
response = {"result": {"response": "true"}}
assert PromptRenderer.normalize_bool(response) == True