Специализированные суб-агенты в Claude Code

orchestration

sub-agents is a Claude Code skill that serves as a reference guide for creating and configuring specialized AI agents with isolated contexts, custom system prompts, and restricted tool sets. Agents are defined as Markdown files stored in `.claude/agents/` (project-level) or `~/.claude/agents/` (user-level), and support model selection — Haiku for fast lookups, Sonnet for general tasks, and Opus for complex reasoning. The skill covers three built-in agents (General-purpose, Plan, Explore), ready-made tool combinations for code reviewers, debuggers, and implementers, plus agent chaining and resumable-agent patterns using agentId. Agents can be managed interactively via `/agents`, created manually as YAML-frontmatter files, or defined inline with the `--agents` CLI flag. Ideal for developers building multi-agent workflows in Claude Code who need clear role separation, least-privilege tool access, and reusable agent definitions across projects.