subagent-artifact

name	subagent-artifact
description	Validate, diagnose, and understand subagent definitions (agents/*.md) — frontmatter format, delegation model, and vendor support across Claude Code, Cursor, and Codex. Use when this capability is needed.
metadata	{"author":"eyelock"}

Use this skill when working with agent definition files — validating frontmatter, diagnosing delegation failures, or understanding what each vendor supports.

Frontmatter fields (Claude Code — most complete support):

• name: required, agent identifier
• description: required, used to route work to this agent
• model: optional, override default model
• tools: optional, space-delimited allowed tools
• disallowedTools: optional, space-delimited blocked tools
• skills: optional, array of skill names to load
• maxTurns: optional, limit agent turns
• effort: optional
• memory, background, isolation: optional

Directory layout:

agents/
└── agent-name.md    # agent definition

Delegation model: An agent is invoked when the orchestrating agent (or user) routes work to it. The delegate receives the harness's AGENTS.md instructions, rules (inlined), and skills (listed by reference).

Vendor support matrix: See references/ — Codex does not support agents in plugins. Cursor support is partial. Claude Code has full support.

---

Source: eyelock/assistants — distributed by TomeVault.