Compositions

Subagents + Skills

A role defines who should do the work; a skill defines the repeatable procedure. Keep the role lean and reuse the same reviewed skill across compatible agents.

Subagents + Worktrees

Worktrees isolate filesystem changes when parallel workers must modify code. Allocate branches and integration ownership before spawning.

Subagents + Hooks

Hooks can record child lifecycle events or check scope on completion. Verify that the product actually emits the relevant events for nested agents.

Subagents + MCP

Give specialists only the external tools their role needs. A research agentAgentA worker you delegate to: brief it, and it takes steps on its own. A chatbot answers; an agent acts.An LLM that runs tools in a loop toward a goal — it acts, checks the result, decides the next step, and repeats until done.Full definitionClose may read documentation; a release agent may inspect deployment state but should not inherit publish authority by default.

Subagents + Memory

Role-specific memory can preserve useful calibration, but it also fragments context. Put shared team truth in version-controlled instructions or a governed knowledge source.

Subagents + Independent Review

One agent implements; another reviews without receiving the implementer's reasoning. The parent resolves findings against code and tests. This reduces anchoring better than asking the same context to critique itself.