What problem does it solve? Long research and engineering sessions lose context when chat history resets, leaving goals, constraints, and prior conclusions untraceable. This Skill converts a new, inherited, or recovered project into a durable, file-based project contract that any future session can read and continue from. ## Core Features & Use Cases - Project Initialization: Creates a ds-lite.work-unit.v1 planning unit via ds_lite_state.py init, plus PROJECT.md, STATUS.md, and RESEARCH_MAP.md files capturing goals, hypotheses, constraints, and acceptance criteria. - Intake Audit for Existing Projects: Preserves existing conclusions, flags stale or unsupported claims, and reconciles non-terminal external tasks (tmux plans, running jobs) before planning new work. - State Graph Management: Maintains research/state/graph.json as machine-readable state with revision-checked mutations and a human-readable Markdown projection. - Use Case: You inherit a half-finished ML experiment folder with scattered notes and a running tmux job. Run intake to audit existing artifacts, reconcile the external task state, and produce a Mission Board with the active node, blockers, and next action. ## Quick Start Ask the assistant to inspect this project folder and turn your research question into a structured project handoff with goals, constraints, and a first action.
