Skill Orchestrator

Problem

When a task spans product, finance, and technical domains, a single Skill often only answers one slice. Engineers may need to manually split the work, choose the right Skills, pass context, and reconcile conclusions. The Skill Orchestrator turns that kind of multi-domain goal into an executable orchestration plan.

How It Works

• Intent parsing: identifies task type, relevant domains, complexity, and constraints, such as a combined plan or a parallel chain.
• Skill discovery: scans local and project-level Skill directories, parses name, description, and location, ranks candidates by relevance, and falls back to the built-in matrix when needed.
• Orchestration planning: generates steps such as step-1, step-2a, and step-3, with dependencies, parallel or serial structure, execution mode, and risk level.
• Execution tracking: uses session_id and step_id to record progress for review and retry; a Checkpoint can be emitted for confirmation.
• Result merging: aggregates subtask outputs, detects numeric or directional conflicts, applies priority-based decisions, and cites source steps in the report.

Boundaries

It is well suited to complex cross-domain analysis and proposal generation across product, finance, engineering, and market tasks. For single-tool operations, such as editing one config item or running one command, calling the relevant Skill directly is simpler. Deployment, deletion, external API calls, or sensitive writes should keep Checkpoints and permission checks; bypassPermissions mode carries higher risk and should be used only with explicit user authorization and reversible scope.