What problem does it solve? Large coding efforts lack a shared, machine-readable plan that multiple agents and worktrees can pick up and continue. This Skill converts a goal into a canonical multi-phase plan with dependencies, lanes, and acceptance criteria, then registers it on the local Loop daemon so any agent can resume the work. ## Core Features & Use Cases - Phased plan authoring: Decomposes a goal into phases with imperative titles, dependency edges, testable Done-when criteria, and descriptive suggested branches, following the canonical format in references/plan-format.md. - Parallelization analysis: Applies a three-part independence test (no dependency, low file contention, no coordination tax) to decide whether phases split into parallel lanes for separate Conductor worktrees, defaulting to a single sequential lane. - Daemon registration and optional Kestral linking: Writes .loop/plan.md, registers it via loop-plan.sh on the local daemon at http://localhost:7717, and optionally publishes a Kestral project, plan document, and parent task with phase subtasks. - Use Case: Ask to plan a full-stack feature; the Skill explores the codebase, drafts vertical-slice phases across repositories, registers the plan locally, and reports the planId so you can run /loop-pickup or /loop-execute in any worktree. ## Quick Start Ask the agent to plan out your next feature or refactor as a multi-phase plan, for example by saying plan out the migration of our billing service into phases.