I want to support subagents in Keen. But before that, we should run a thorough investigation on how existing coding agents do this and how we can leverage the unique strength of Keen Code. Explore Claude Code, Codex CLI, OpenCode and other coding agents. Then explain how they do subagents. Finally, suggest possible approach for Keen Code.
Save the outcome in .ai-interactions/tasks/issue-55 as output-1_investigation.md
Given how Keen handles context as per docs/turn-memory.md, does adding subagent add any value for context management?
is this a good idea to implement subagents in keen just like how skills are implemented? Through YAML config + subagent calling tool?