intent-orchestration-contract

What problem does it solve?

This Skill provides a contract-first way to transform user intent and context into a structured, validated orchestration plan for multi-agent execution, instead of relying on ad-hoc string routing or hardcoded agent names.

Core Features & Use Cases

• Structured OrchestrationPlan: Defines a standardized plan format with tasks, strategy (single / ordered_parallel / sequential), expected outputs, dependencies, and fallback semantics.
• Local Plan Validation: Specifies schema validation plus capability/agent/expected-output checks before any execution is allowed.
• Safe LLM Planning: Constrains LLM planners to output valid JSON candidates that are always re-validated locally and never directly drive execution.
• Group & Mention Routing: Supports group conversations and @mention/manual/direct/auto planning modes with clear routing priorities and audit trace requirements.
• Fallback & Retry Guardrails: Defines bounded fallback modes with re-validation and SafeError behavior when all options fail.

Quick Start

Ask your Orchestrator to generate a structured OrchestrationPlan from PlannerInput for a multi-agent scenario, then run Plan Validation and only execute the plan if it passes all safety and capability checks.