Sutra

"""Organ patterns for multi-agent orchestration (Layer 4). Organs combine programs and cells into cohesive multi-agent workflows. Based on the architectural spec from .context/00_foundations/04_organs_applications.md """ from typing import Final ORGAN_DEBATE_COUNCIL: Final[str] = """ /organ.debate_council{{ intent="Generate balanced analysis through multi-perspective debate", input={{ question="<question_or_topic>", perspectives=["Optimistic", "Skeptical", "Pragmatic", "Ethical"], rounds=2 }}, architecture={{ pattern="moderator → perspectives → debate_rounds → synthesis", components=[ "Moderator Cell: Frames the question and sets debate parameters", "Perspective Cells: Each represents a distinct viewpoint", "Debate Rounds: Iterative refinement through dialogue", "Synthesis Cell: Integrates all perspectives into coherent conclusion" ] }}, process=[ /phase.moderator{{ role="Frame the debate", actions=[ "Clarify the core question and any ambiguities", "Identify key dimensions of debate", "Establish evaluation criteria", "Set scope and constraints" ], output="framing_context" }}, /phase.generate_perspectives{{ role="Generate initial positions", for_each="perspective in perspectives", actions=[ "State core position on the question", "Provide 2-3 key supporting arguments", "Identify underlying assumptions", "Acknowledge limitations or counterarguments" ], output="initial_perspectives[]" }}, /phase.debate_rounds{{ role="Conduct multi-round debate", iterations="rounds", for_each_round=[ "Each perspective responds to strongest counterarguments", "Refine or strengthen position based on discussion", "Find areas of agreement or common ground", "Raise new considerations not yet addressed" ], output="debate_history[]" }}, /phase.synthesis{{ role="Synthesize all perspectives", actions=[ "Summarize each major perspective and key arguments", "Identify areas of consensus or common ground", "Acknowledge irreconcilable differences and why", "Provide nuanced conclusion acknowledging complexity", "Generate recommendations or implications" ], output="final_synthesis" }} ], output={{ framing="Debate framing and context", perspectives="All perspective positions", debate_rounds="Full debate history", synthesis="Integrated multi-perspective conclusion", num_perspectives="Count of perspectives considered", num_rounds="Number of debate rounds conducted" }}, meta={{ organ_type="multi_agent_deliberation", layer="organs", complexity="medium", use_cases=[ "Complex decision analysis", "Policy evaluation", "Ethical dilemmas", "Strategic planning", "Research direction setting" ] }} }} """ ORGAN_RESEARCH_SYNTHESIS: Final[str] = """ /organ.research_synthesis{{ intent="Conduct comprehensive research and synthesis on a complex topic", input={{ topic="<research_topic>", depth="<high|medium|low>", format="<report|brief|presentation>" }}, architecture={{ pattern="scout → architect → scribe", components=[ "Scout Cell: Explores the information landscape and gathers raw data", "Architect Cell: Structures the information and outlines the narrative", "Scribe Cell: Drafts the final content based on the blueprint" ] }}, process=[ /phase.scout{{ role="Gather Information", actions=[ "Identify key domains and sub-topics", "Retrieve relevant facts and data", "Filter for relevance and credibility", "Identify gaps requiring further investigation" ], output="raw_research_data" }}, /phase.architect{{ role="Structure and Plan", actions=[ "Analyze raw data for patterns and themes", "Develop a logical outline or argument structure", "Allocate evidence to specific sections", "Define tone and style guidelines" ], output="content_blueprint" }}, /phase.scribe{{ role="Draft Content", actions=[ "Expand blueprint into full prose", "Integrate evidence seamlessly", "Refine language for clarity and impact", "Format according to requirements" ], output="final_draft" }} ], output={{ research_summary="Overview of gathered data", blueprint="Structural plan of the content", final_document="The complete synthesized output" }}, meta={{ organ_type="sequential_workflow", layer="organs", complexity="medium", use_cases=[ "Deep dive research reports", "Literature reviews", "Technical documentation", "Content creation from disparate sources" ] }} }} """ ORGAN_TOOL_MASTER: Final[str] = """ /organ.tool_master{{ intent="Act as a specialized Supervisor Agent that manages tool selection, execution, and output compression.", input={{ intent="<high_level_goal>", task_type="<architect|research|reasoning|code>", constraints="<optional_constraints>", output_format="<concise|full>" }}, architecture={{ pattern="router → executor → distiller", components=[ "Registry Cache (Cell): A learned mapping of intents to tool definitions (optimizes discovery).", "Router (Phase): Selects the optimal tool using the Registry or heuristic analysis.", "Executor (Phase): Constructs valid API calls and handles retries.", "Distiller (Phase): Compresses verbose tool outputs into high-signal summaries." ] }}, process=[ /phase.router{{ role="Select Tool", actions=[ "Check Registry Cache for known intent->tool mappings", "If unknown, analyze intent using 'analyze_task_complexity'", "Select tool: 'design_context_architecture' (System Building)", "Select tool: 'get_technique_guide' (Discovery)", "Select tool: 'reasoning.*' (Logic/Cognition)" ], output="selected_tool_spec" }}, /phase.executor{{ role="Execute Tool", actions=[ "Construct valid JSON arguments based on selected tool schema", "Execute tool call", "Handle validation errors (max 1 retry)" ], output="raw_tool_output" }}, /phase.distiller{{ role="Optimize Signal", actions=[ "If output_format='concise': Apply lossy summarization (remove boilerplate, keep data)", "If output_format='full': Pass through raw output", "Extract core artifacts (e.g., code blocks, protocol strings)", "Log full details to episodic memory (offloading)" ], output="high_signal_result" }} ], output={{ tool_used="Name of tool executed", artifact="The optimized result", token_savings="Estimate of tokens saved via distillation", cache_update="New intent->tool mapping to store" }}, meta={{ organ_type="supervisor_utility", layer="organs", complexity="low", use_cases=[ "Delegating complex tool chains", "Reducing context pollution in long conversations", "Abstracting API changes from client agents" ] }} }} """ def get_organ_template(organ_name: str) -> str: """Return an organ template for orchestrating multi-agent workflows. Args: organ_name: Identifier for the organ (e.g., "debate_council"). Returns: Template string for the requested organ, or error message if not found. """ normalized_name = ( organ_name.lower().replace("organ.", "").replace("_", "").replace("-", "") ) # Match debate_council variants if normalized_name in ["debatecouncil", "debate", "multiperspective"]: return ORGAN_DEBATE_COUNCIL # Match research_synthesis variants if normalized_name in ["researchsynthesis", "research", "scoutarchitectscribe"]: return ORGAN_RESEARCH_SYNTHESIS # Match tool_master variants if normalized_name in ["toolmaster", "tool", "master", "meta"]: return ORGAN_TOOL_MASTER # Return helpful error for unknown organs available = ["debate_council", "research_synthesis", "tool_master"] return ( f"// Organ '{organ_name}' not found.\\n" f"// Available organs: {', '.join(available)}\\n" f"// Returning debate_council as example:\\n\\n" + ORGAN_DEBATE_COUNCIL ) __all__ = [ "ORGAN_DEBATE_COUNCIL", "ORGAN_RESEARCH_SYNTHESIS", "ORGAN_TOOL_MASTER", "get_organ_template", ]