What problem does it solve? AI coding agents lose track of multi-step work after a context reset or /clear, and flat markdown checklists cannot express task dependencies, so agents guess the wrong next step. Plandeck keeps the plan as a structured plan.yaml file on disk and renders it as a live Kanban board that recomputes itself on every edit. ## Core Features & Use Cases - Self-organizing Kanban board: Cards auto-promote from Backlog to Ready when their dependencies complete, the critical path is drawn in gold, and a banner names the single next action, all streamed live over SSE with zero dependencies. - Deterministic planning engine: Pure functions over YAML compute the ready-queue, points-weighted critical path, progress rollups, velocity-based ETA, and one tie-broken next action, with no AI or LLM involved. - Crash-proof continuity: A mission journal, last-known-good snapshots, a doctor restore command, and a NEXT.md breadcrumb let an agent recover its exact next move after /clear without re-reading the whole plan. - Use Case: An agent shipping an onboarding flow breaks the work into ten cards with depends_on edges, runs plandeck board to watch progress live, and after a context reset runs plandeck next to instantly resume the one active card. ## Quick Start Ask the agent to initialize a Plandeck plan in the current project, break the task into dependency-linked cards, and start the live board with plandeck board.