planlisted

# swarm:plan — plan, mock up, then have the swarm try to break the plan > Read the `swarm:orchestrate` skill first for the fan-out mechanics (team creation, briefs, blinded verification, monitoring). This skill is the **plan mode** layered on top: research deeply, **draw what the user will see**, produce a behavior-first change proposal, and validate it against independent agents who plan the same feature blind. You are planning: **$ARGUMENTS** The deliverable is not a paragraph of intentions — it is a **rigorous, behavior-first change proposal** plus **concrete mock-ups** for any surface a human will look at: a precise statement of the delta, the tasks to get there, the visual/UX shape, and the behavior it leaves behind. Then de-risk it by having the swarm independently arrive at their own plans and reconciling. **plan vs spec:** this skill is the *delta* (what we will build). `/swarm:spec` is the durable *is* (what the capability already guarantees) for other agents/humans so they do not invent wrong behavior. ## Existing work and the execution boundary Before planning, search open PRs and relevant tickets through the configured tracker (`tickets` skill), then read matches, linked changes, and active ownership. Link and extend prior work; coordinate overlapping work. A failed lookup is an unknown. During design iteration keep tracker discovery read-only: no new issues or subtasks, claims, or In Progress moves just to plan. Draft steps belong in the plan and local harn