Планирование задач и handoff-документы для агента

ai-agents

plan-agent is a Claude Code skill that generates detailed implementation plans and handoff documents from conversation context. In brownfield projects it reads a pre-built `codebase-map.md` instead of re-exploring the codebase, while greenfield scenarios start from scratch based on stated requirements. For complex or ambiguous tasks the agent enters a structured deep-interview loop, covering six areas: problem definition, user context, technical constraints, edge cases, success criteria, and tradeoffs — continuing until all areas have concrete answers or the user opts to proceed. The finished plan is written to `thoughts/shared/plans/PLAN-<description>.md` and a summary handoff is saved to the session handoff directory. It fits engineering teams that need to capture architectural decisions and pass work to an implementation agent without losing context.