Subagent
A specialized agent Claude Code can spawn to handle a focused task in isolation, with its own system prompt, tools, and context window.
Subagents solve two problems at once: context bloat and role blurring. Instead of stuffing your main Claude conversation with tool noise from a long investigation, you delegate the discrete piece of work to a subagent. It runs with its own focused system prompt, its own allowed tools, and returns a clean summary — keeping the main thread on track.
A subagent is defined by a markdown file with frontmatter (name, description, tools, model) and a system prompt body. Drop it into ~/.claude/agents/ globally or .claude/agents/ per-project. Claude Code surfaces it as a Task tool target; you can also chain subagents (an orchestrator agent calling specialist agents in parallel).
Use cases: PR security reviewer, accessibility auditor, performance profiler, test writer, doc generator, infrastructure inspector. ClaudeWave indexes 480+ open-source subagents you can drop into your project today.