What problem does it solve? Defining a new specialist subagent requires coordinating a base prompt, shared metadata, and three harness-specific JSON files while avoiding duplicate roles, wrong intelligence levels, and inconsistent trigger surfaces. This Skill scaffolds the full five-file agent definition with confirmed settings before anything is written. ## Core Features & Use Cases - Five-file agent scaffolding: Generates base.md plus frontmatter/meta.json, claude.json, codex.json, and grok.json under the owning plugin's agents/<name>/ directory. - Archetype-based configuration: Classifies the role as producer, critic, or orchestrator and derives intelligence level, permissionMode, memory, and isolation settings, confirming deviations with the user via a structured prompt. - Duplicate and overlap rejection: Searches existing agent descriptions and callers to reject roles already covered, routing modifications to update-agent instead. - Deterministic verification: Validates JSON with Python and runs Essential's stitch helper for all three harnesses into temporary outputs, checking placeholders, routing rows, and trigger non-overlap. - Use Case: When you need a new frontend-implementer subagent, invoke this Skill with the role description; it classifies the role, proposes intelligence and permission settings, confirms them, writes the five canonical files, updates the plugin's routing table, and verifies the stitched output. ## Quick Start Ask the agent to create a new specialist agent for your role, for example: create a new agent called frontend-implementer owned by the coding plugin.
