Summary

Kiro has a working subagent system but with significant limitations compared to Claude Code's Task()/Agent() tools. These need to be documented for plugin authors and noted in the platform comparison.

Current Kiro subagent state

Working

• Primary agent invokes subagents by name from .kiro/agents/*.json
• Synchronous execution - parent waits, gets result, continues
• Direct CLI launch: kiro-cli --agent <name>
• Custom agent configs with tools, allowedTools, resources, hooks, MCP

Not working / Not implemented

• No parallel spawning (Enable Parallel Task Execution with Sub-Agent Sessions in Kiro kirodotdev/Kiro#2288 - open feature request)
• No async subagents (Multi-Agent Orchestration: Support sync/async subagents, native context handoff, and context-efficient coordinator workflow kirodotdev/Kiro#4262 - "actively thinking about")
• No permission inheritance (Subagents should inherit parent agent's allowedTools kirodotdev/Kiro#5449 - open)
• Spawning regression in v1.26.0 (Subagent failing to spawn with "unable to find agent with name <agent>" kirodotdev/Kiro#5743 - "unable to find agent")
• No Team/swarm pattern - no multi-agent coordination
• No SendMessage between agents - synchronous call-return only

What to document

In docs/CROSS_PLATFORM.md

Add a section on Kiro subagent capabilities and limitations vs other platforms.

In docs/INSTALLATION.md

Note that agentsys skills using parallel Task() calls will execute sequentially on Kiro.

In README.md

Update the platform comparison to note subagent support level.

In CLAUDE.md

Add Kiro subagent notes to the platform-aware section.

Platform comparison for subagents

Acceptance criteria

• Cross-platform docs updated with subagent comparison
• Plugin author guidance on multi-agent patterns per platform
• Kiro limitations clearly noted (no parallel, no messaging)
• Upstream Kiro issues referenced for tracking