What problem does it solve? Coordinating work between two AI agents (Cowork on Claude Desktop and Claude Code CLI) on a solo developer project is error-prone without a shared protocol. This Skill enforces a structured five-stage file-based handoff workflow (directive, plan, review, report, decision) so tasks are issued, executed, verified, and closed without miscommunication or state corruption. ## Core Features & Use Cases - Five-Stage Handoff Protocol: Manages the lifecycle of handoff folders (HO-, PLG-, DOC-) through 01_directive, 02_plan, 03_review, 04_report, and 05_decision files with a defined state machine (DRAFT through CLOSED-MERGE/CLOSED-REJECT). - Role Separation Enforcement: Defines clear responsibilities where Cowork handles strategy, review, and INDEX.md updates while Claude Code handles planning, execution, and reporting, with unidirectional write rules per file. - Guardrails and Decision Trees: Provides a decision tree for four read-only commands (list, pickup, execute, report), an APPROVED gate check before execution, and operational rules covering NFD path handling, Kaggle CLI usage, and commit approval. - Use Case: A solo developer running Kaggle experiments asks Claude Code to pick up an issued handoff; the Skill directs it to read INDEX.md, create 02_plan.md from the template, wait for Cowork approval, execute, and write 04_report.md before closure. ## Quick Start Ask the assistant to list pending handoffs and pick up the next issued one from the handoff folder following the solo handoff protocol.
